Video wird geladen...

Video konnte nicht geladen werden

Zur Startseite

testing Qwen3.5-35B-A3B latest optimized version by UnslothAI on a single RTX 3090. one detailed prompt. zero handholding. watch a 3B model scaffold an entire multifile game project autonomously. the setup: > model: Qwen3.5-35B-A3B (80B total, only 3B active per token) > quant: UD-Q4_K_XL by Unsloth (MXFP4 layers removed in...

sudoingX's profile picture

Sudo su

33,179 subscribers

167,244 Aufrufe • vor 6 Monaten •via X (Twitter)

GamingWissenschaft & Technologie
LIVE

Anya Rossi• Live Now

Private livecam show

0 Kommentare

Keine Kommentare verfügbar

Kommentare vom Original-Post werden hier angezeigt

Ähnliche Videos

nvidia's 3B mamba destroyed alibaba's 3B deltanet on the same RTX 3090. only 24 days between releases. same active parameters, same VRAM tier, completely different architectures. nemotron cascade 2: 187 tok/s. flat from 4K to 625K context. zero speed loss. flags: -ngl 99 -np 1. that's it. no context flags, no KV cache tricks. auto-allocates 625K. qwen 3.5 35B-A3B: 112 tok/s. flat from 4K to 262K context. zero speed loss. flags: -ngl 99 -np 1 -c 262144 --cache-type-k q8_0 --cache-type-v q8_0. needed KV cache quantization to fit 262K. both models held a flat line across every context level. both architectures are context-independent. but nvidia's mamba2 is 67% faster at generating tokens on the exact same hardware and needs fewer flags to get there. same node, same GPU, same everything. the only variable is the model. gold medal math olympiad winner running at 187 tokens per second on single RTX 3090 a card from 6 years ago. nvidia cooked.
0:36

nvidia's 3B mamba destroyed alibaba's 3B deltanet on the same RTX 3090. only 24 days between releases. same active parameters, same VRAM tier, completely different architectures. nemotron cascade 2: 187 tok/s. flat from 4K to 625K context. zero speed loss. flags: -ngl 99 -np 1. that's it. no context flags, no KV cache tricks. auto-allocates 625K. qwen 3.5 35B-A3B: 112 tok/s. flat from 4K to 262K context. zero speed loss. flags: -ngl 99 -np 1 -c 262144 --cache-type-k q8_0 --cache-type-v q8_0. needed KV cache quantization to fit 262K. both models held a flat line across every context level. both architectures are context-independent. but nvidia's mamba2 is 67% faster at generating tokens on the exact same hardware and needs fewer flags to get there. same node, same GPU, same everything. the only variable is the model. gold medal math olympiad winner running at 187 tokens per second on single RTX 3090 a card from 6 years ago. nvidia cooked.

Sudo su

187,474 Aufrufe • vor 5 Monaten

this is what a 24gb VRAM builds in 2026. one prompt. ten files. 3,483 lines of code. zero handholding. i gave Qwen3.5-35B-A3B a single detailed spec describing the full game architecture and hit enter. enemy types, particle systems, procedural audio, powerups, boss fights, ship upgrades, parallax backgrounds, everything in one message. the model planned the file structure itself, wrote every module in dependency order, wired all the imports, and served the game on port 3001. it ran on first load. when it hit a bug in collision detection it read its own error output, found the issue, fixed it, and kept building. this is pure agent loop running on local hardware. what you're looking at is pixelated octopus aliens with tentacle animations, 4 layer parallax space background with planets at different depths, a full particle system handling explosions and ink splatter and engine trails and bullet impacts, procedural audio through Web Audio API with zero sound files loaded, unleash mode with combo multiplier, boss fights every 5 levels, ship upgrades that unlock as you progress. no libraries. no frameworks. vanilla JS and Canvas. 3B active parameters. single RTX 3090. llama.cpp with q8_0 KV cache at 262K context. Claude Code pointed at localhost:8080 through the native Anthropic endpoint. no API costs. 112 tok/s. a GPU you can buy used for $800. game is called Octopus Invaders and i actually like playing it.
1:11

this is what a 24gb VRAM builds in 2026. one prompt. ten files. 3,483 lines of code. zero handholding. i gave Qwen3.5-35B-A3B a single detailed spec describing the full game architecture and hit enter. enemy types, particle systems, procedural audio, powerups, boss fights, ship upgrades, parallax backgrounds, everything in one message. the model planned the file structure itself, wrote every module in dependency order, wired all the imports, and served the game on port 3001. it ran on first load. when it hit a bug in collision detection it read its own error output, found the issue, fixed it, and kept building. this is pure agent loop running on local hardware. what you're looking at is pixelated octopus aliens with tentacle animations, 4 layer parallax space background with planets at different depths, a full particle system handling explosions and ink splatter and engine trails and bullet impacts, procedural audio through Web Audio API with zero sound files loaded, unleash mode with combo multiplier, boss fights every 5 levels, ship upgrades that unlock as you progress. no libraries. no frameworks. vanilla JS and Canvas. 3B active parameters. single RTX 3090. llama.cpp with q8_0 KV cache at 262K context. Claude Code pointed at localhost:8080 through the native Anthropic endpoint. no API costs. 112 tok/s. a GPU you can buy used for $800. game is called Octopus Invaders and i actually like playing it.

Sudo su

153,735 Aufrufe • vor 6 Monaten

first impressions of qwen 3.5 27B dense on a single RTX 3090. 35 tok/s. from 4K all the way to 300K+ context. no speed drop. hermes 4.3 started at 35 and degraded to 15 as context filled. qwen dense holds. MoE held 112 flat. 3x faster but only 3B of 35B active per token. architecture tradeoff. Q4_K_M on 16.7GB. native context 262K. pushed past training limit to 376K before VRAM ceiling on 24GB. tried q8 KV cache at 262K, speed collapsed to 11 tok/s. q4_0 KV is the sweet spot. flash attention mandatory. built in reasoning mode. the model thinks step by step before it answers. full chain of thought surviving Q4 quant. 1,799+ token thinking chains with self correction loops. on a single consumer GPU. gave it one prompt: "build a realtime particle galaxy simulation in one HTML file." 3,340 tokens. 95 seconds. one shot. ran on first load. full reasoning and coding in the video below. optimal config if you want to skip the hours of testing: llama-server -ngl 99 -c 262144 -fa on --cache-type-k q4_0 --cache-type-v q4_0 this is just the warmup. octopus invaders is next: 10 files, 3,400+ lines, zero steering. the prompt hermes quit at 22%. already more impressed than expected. full results coming soon.
2:02

first impressions of qwen 3.5 27B dense on a single RTX 3090. 35 tok/s. from 4K all the way to 300K+ context. no speed drop. hermes 4.3 started at 35 and degraded to 15 as context filled. qwen dense holds. MoE held 112 flat. 3x faster but only 3B of 35B active per token. architecture tradeoff. Q4_K_M on 16.7GB. native context 262K. pushed past training limit to 376K before VRAM ceiling on 24GB. tried q8 KV cache at 262K, speed collapsed to 11 tok/s. q4_0 KV is the sweet spot. flash attention mandatory. built in reasoning mode. the model thinks step by step before it answers. full chain of thought surviving Q4 quant. 1,799+ token thinking chains with self correction loops. on a single consumer GPU. gave it one prompt: "build a realtime particle galaxy simulation in one HTML file." 3,340 tokens. 95 seconds. one shot. ran on first load. full reasoning and coding in the video below. optimal config if you want to skip the hours of testing: llama-server -ngl 99 -c 262144 -fa on --cache-type-k q4_0 --cache-type-v q4_0 this is just the warmup. octopus invaders is next: 10 files, 3,400+ lines, zero steering. the prompt hermes quit at 22%. already more impressed than expected. full results coming soon.

Sudo su

122,051 Aufrufe • vor 6 Monaten

i spent 3 hours finding the sweet spot for hermes 4.3 36B on a single RTX 3090. saving you the trouble, anon. the model is 21.8GB at Q4_K_M. that leaves 2.2GB free on 24GB VRAM. not much room for KV cache. here's what actually happened: 4K: 35.3 tok/s 8K: 35.2 tok/s 16K: 34.8 tok/s 32K: 34.6 tok/s 64K: 6.4 tok/s 128K: 1.9 tok/s flat from 4K to 32K. then it falls off a cliff at 64K. the trick is quantized KV cache. without it you OOM at 16K. with quantized KV cache you get 32K at full speed. all 64 layers on GPU. at 64K something weird happens. ngl 99 (all layers on GPU) = 3.96 tok/s. the KV cache silently spills to CPU. drop to ngl 55 and speed jumps to 6.37. drop to ngl 48 and it gets worse again (3.46). there's an offload sweet spot where you free just enough VRAM for the cache without losing too much compute to PCIe transfers. 128K works at ngl 32 but you're at 1.95 tok/s. half the model on CPU. usable for batch work, not for interactive. the sweet spot command: llama-server -m hermes-4.3-36b-Q4_K_M.gguf -ngl 99 -c 32768 --cache-type-k q4_0 --cache-type-v q4_0 32K context. 34.6 tok/s. all on GPU. this is where dense 36B lives on 24GB. for comparison, qwen 3.5 (35B MoE, 3B active) holds 112 tok/s from 4K all the way to 262K on the same GPU. no speed drop. same total params, completely different architecture. hybrid linear attention means flat context scaling. dense pays for every token in the KV cache. code and quality comparison coming next. fast vs slow generation side by side in the videos below.
2:29

i spent 3 hours finding the sweet spot for hermes 4.3 36B on a single RTX 3090. saving you the trouble, anon. the model is 21.8GB at Q4_K_M. that leaves 2.2GB free on 24GB VRAM. not much room for KV cache. here's what actually happened: 4K: 35.3 tok/s 8K: 35.2 tok/s 16K: 34.8 tok/s 32K: 34.6 tok/s 64K: 6.4 tok/s 128K: 1.9 tok/s flat from 4K to 32K. then it falls off a cliff at 64K. the trick is quantized KV cache. without it you OOM at 16K. with quantized KV cache you get 32K at full speed. all 64 layers on GPU. at 64K something weird happens. ngl 99 (all layers on GPU) = 3.96 tok/s. the KV cache silently spills to CPU. drop to ngl 55 and speed jumps to 6.37. drop to ngl 48 and it gets worse again (3.46). there's an offload sweet spot where you free just enough VRAM for the cache without losing too much compute to PCIe transfers. 128K works at ngl 32 but you're at 1.95 tok/s. half the model on CPU. usable for batch work, not for interactive. the sweet spot command: llama-server -m hermes-4.3-36b-Q4_K_M.gguf -ngl 99 -c 32768 --cache-type-k q4_0 --cache-type-v q4_0 32K context. 34.6 tok/s. all on GPU. this is where dense 36B lives on 24GB. for comparison, qwen 3.5 (35B MoE, 3B active) holds 112 tok/s from 4K all the way to 262K on the same GPU. no speed drop. same total params, completely different architecture. hybrid linear attention means flat context scaling. dense pays for every token in the KV cache. code and quality comparison coming next. fast vs slow generation side by side in the videos below.

Sudo su

52,268 Aufrufe • vor 6 Monaten

A single RTX 4090 (24 GB VRAM) can run the updated gemma 4 31B (dense) model with a 190,000 context window at 33 tokens/second. The VRAM barrier is dying. Google quietly updated Gemma 4, and Unsloth immediately compiled the new quants. I built llama.cpp from source on Ubuntu 22 to benchmark it. Google's stealth update 2 days ago enabled uniform Flash Attention 4 on Hopper to boost prefill and patched the chat template to improve tool calling. The agentic reasoning gains on the benchmark charts are massive: TB2 (Agents): +4.5% (to 25.8%) Tau2 (Telecom): +10.1% (to 62.7%) Running on Ubuntu 22, CUDA 13.0 with a single NVIDIA GeForce RTX 4090. Here is the exact step by step benchmarking process with a massive 28k tokens prompt and the commands I used to squeeze out maximum context without killing my throughput: # 1. The Baseline (Unquantized KV Cache) I started with full GPU offload (-ngl 99) and pushed the context to 40k. llama.cpp flags: ./build/bin/llama-server -m gemma-4-31B-it-qat-UD-Q4_K_XL.gguf -ngl 99 -c 40000 -fa on --port 8080 -v VRAM: 23.8 GB (maxed out on card) Throughput: Prefill: 2198.81 t/s | Decode: 35.77 t/s (with 28k tokens prompt) # 2. The CPU Split Trap I tried stretching to 80k context by offloading layers to the CPU (-ngl 52). llama.cpp flags: ./build/bin/llama-server -m gemma-4-31B-it-qat-UD-Q4_K_XL.gguf -c 80000 -ngl 52 -fa on --port 8080 -v Throughput: Prefill: 1212.73 t/s | Decode: 5 t/s (with 28k tokens prompt) # 3. The KV Quantization Breakthrough Instead of spilling layers to the CPU, I kept the model fully on card (-ngl 99) but enabled 8-bit KV cache quantization to free up VRAM. flags: ./build/bin/llama-server -m gemma-4-31B-it-qat-UD-Q4_K_XL.gguf -c 100000 --cache-type-k q8_0 --cache-type-v q8_0 -ngl 99 --port 8080 -v VRAM: 23.9 GB Throughput: Prefill: 2139.68 t/s | Decode: 32 t/s (with 28k tokens prompt) Result: 100k tokens of context on a single GPU with practically zero speed loss (and minimal intelligence loss). # 4. The Limit Test (Q4 KV Cache) To find the absolute breaking point, I dropped the KV cache to 4 bit (q4_0) and set -c 190000. flags: ./build/bin/llama-server -m gemma-4-31B-it-qat-UD-Q4_K_XL.gguf -c 190000 --cache-type-k q4_0 --cache-type-v q4_0 -ngl 99 --port 8080 -v VRAM: 23.8 GB Throughput: Prefill: 2206.66 t/s | Decode: 33 t/s (with 28k tokens prompt) (Note: Pushing it to 220k required dropping to -ngl 58 again, which immediately penalized decode down to 17 t/s). # The Tradeoff: For Max Reasoning: Keep your KV cache unquantized (f16). You get pristine reasoning but hit a strict 40k context ceiling. For Massive Document Retrieval: If you need to feed the model giant codebases, use --cache-type-k q4_0. Getting 190k context at 33 tokens/second on a consumer desktop with a 31b dense model is a cheat code. If you’re rocking a single 3090 or 4090 and slept on Gemma 4 earlier, this update is your cue to dust off the terminal. Hugging Face links to the Unsloth QAT quants are in the replies below.
0:20

A single RTX 4090 (24 GB VRAM) can run the updated gemma 4 31B (dense) model with a 190,000 context window at 33 tokens/second. The VRAM barrier is dying. Google quietly updated Gemma 4, and Unsloth immediately compiled the new quants. I built llama.cpp from source on Ubuntu 22 to benchmark it. Google's stealth update 2 days ago enabled uniform Flash Attention 4 on Hopper to boost prefill and patched the chat template to improve tool calling. The agentic reasoning gains on the benchmark charts are massive: TB2 (Agents): +4.5% (to 25.8%) Tau2 (Telecom): +10.1% (to 62.7%) Running on Ubuntu 22, CUDA 13.0 with a single NVIDIA GeForce RTX 4090. Here is the exact step by step benchmarking process with a massive 28k tokens prompt and the commands I used to squeeze out maximum context without killing my throughput: # 1. The Baseline (Unquantized KV Cache) I started with full GPU offload (-ngl 99) and pushed the context to 40k. llama.cpp flags: ./build/bin/llama-server -m gemma-4-31B-it-qat-UD-Q4_K_XL.gguf -ngl 99 -c 40000 -fa on --port 8080 -v VRAM: 23.8 GB (maxed out on card) Throughput: Prefill: 2198.81 t/s | Decode: 35.77 t/s (with 28k tokens prompt) # 2. The CPU Split Trap I tried stretching to 80k context by offloading layers to the CPU (-ngl 52). llama.cpp flags: ./build/bin/llama-server -m gemma-4-31B-it-qat-UD-Q4_K_XL.gguf -c 80000 -ngl 52 -fa on --port 8080 -v Throughput: Prefill: 1212.73 t/s | Decode: 5 t/s (with 28k tokens prompt) # 3. The KV Quantization Breakthrough Instead of spilling layers to the CPU, I kept the model fully on card (-ngl 99) but enabled 8-bit KV cache quantization to free up VRAM. flags: ./build/bin/llama-server -m gemma-4-31B-it-qat-UD-Q4_K_XL.gguf -c 100000 --cache-type-k q8_0 --cache-type-v q8_0 -ngl 99 --port 8080 -v VRAM: 23.9 GB Throughput: Prefill: 2139.68 t/s | Decode: 32 t/s (with 28k tokens prompt) Result: 100k tokens of context on a single GPU with practically zero speed loss (and minimal intelligence loss). # 4. The Limit Test (Q4 KV Cache) To find the absolute breaking point, I dropped the KV cache to 4 bit (q4_0) and set -c 190000. flags: ./build/bin/llama-server -m gemma-4-31B-it-qat-UD-Q4_K_XL.gguf -c 190000 --cache-type-k q4_0 --cache-type-v q4_0 -ngl 99 --port 8080 -v VRAM: 23.8 GB Throughput: Prefill: 2206.66 t/s | Decode: 33 t/s (with 28k tokens prompt) (Note: Pushing it to 220k required dropping to -ngl 58 again, which immediately penalized decode down to 17 t/s). # The Tradeoff: For Max Reasoning: Keep your KV cache unquantized (f16). You get pristine reasoning but hit a strict 40k context ceiling. For Massive Document Retrieval: If you need to feed the model giant codebases, use --cache-type-k q4_0. Getting 190k context at 33 tokens/second on a consumer desktop with a 31b dense model is a cheat code. If you’re rocking a single 3090 or 4090 and slept on Gemma 4 earlier, this update is your cue to dust off the terminal. Hugging Face links to the Unsloth QAT quants are in the replies below.

Alok

76,069 Aufrufe • vor 1 Monat

Google's Gemma 4 26B A4B QAT hits 25+ tokens/sec and 320+ tokens/sec prefill on 8 GB VRAM (RTX 4060) + 16 GB RAM using TurboQuant Prefill just went from 200 → 320+ tok/s on the same 8GB card. 1.6x, no new hardware, no new quant, just a KV cache trick stacked on top of the Gemma 4 26B MoE setup from a few days ago. A few days ago I posted Gemma 4 26B A4B hitting 28 tok/s decode on 8GB VRAM using native MTP. prefill was stuck around 200 tok/s. fair callout by the community. So today I tested something I'd already been meaning to try: TheTom/llama-cpp-turboquant, the TurboQuant KV cache fork by Tom Turney (Tom Turney). (github link in the comments) thanks to him, the fork just got resynced to mainline, so MTP + TurboQuant now run together cleanly (I didnt see any meaningful gains by using MTP with this setup though but you can try). The flags (No MTP): -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf -cnv -c 64000 --cache-type-k q8_0 --cache-type-v turbo3 Results on the same RTX 4060 8GB, tested with a 27k token prompt at 64k context loaded: Prefill: 200 tok/s → 320+ tok/s Decode: stayed above 25 tok/s (without MTP) Why it works: TurboQuant uses walsh hadamard rotation + polar quantization on the KV cache. keys are sensitive to compression, values aren't much, so it splits the difference: K stays at q8_0, V drops to turbo3 (~3 bits). bonus from the memory savings: same 8GB card can now stretch to 100-120k context with minimal decode penalty. It should now be snappier with any agent harness such as hermes agent without compromise on intelligence. If you're already running Gemma 4 on a small card, this stacks on top for free. Try --cache-type-k q8_0 --cache-type-v turbo3 on your setup and report back what your prefill/decode split looks like. unsloth model gguf and llama.cpp turboquant fork links in the comments. what's your prefill number before vs after?
0:31

Google's Gemma 4 26B A4B QAT hits 25+ tokens/sec and 320+ tokens/sec prefill on 8 GB VRAM (RTX 4060) + 16 GB RAM using TurboQuant Prefill just went from 200 → 320+ tok/s on the same 8GB card. 1.6x, no new hardware, no new quant, just a KV cache trick stacked on top of the Gemma 4 26B MoE setup from a few days ago. A few days ago I posted Gemma 4 26B A4B hitting 28 tok/s decode on 8GB VRAM using native MTP. prefill was stuck around 200 tok/s. fair callout by the community. So today I tested something I'd already been meaning to try: TheTom/llama-cpp-turboquant, the TurboQuant KV cache fork by Tom Turney (Tom Turney). (github link in the comments) thanks to him, the fork just got resynced to mainline, so MTP + TurboQuant now run together cleanly (I didnt see any meaningful gains by using MTP with this setup though but you can try). The flags (No MTP): -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf -cnv -c 64000 --cache-type-k q8_0 --cache-type-v turbo3 Results on the same RTX 4060 8GB, tested with a 27k token prompt at 64k context loaded: Prefill: 200 tok/s → 320+ tok/s Decode: stayed above 25 tok/s (without MTP) Why it works: TurboQuant uses walsh hadamard rotation + polar quantization on the KV cache. keys are sensitive to compression, values aren't much, so it splits the difference: K stays at q8_0, V drops to turbo3 (~3 bits). bonus from the memory savings: same 8GB card can now stretch to 100-120k context with minimal decode penalty. It should now be snappier with any agent harness such as hermes agent without compromise on intelligence. If you're already running Gemma 4 on a small card, this stacks on top for free. Try --cache-type-k q8_0 --cache-type-v turbo3 on your setup and report back what your prefill/decode split looks like. unsloth model gguf and llama.cpp turboquant fork links in the comments. what's your prefill number before vs after?

Alok

119,821 Aufrufe • vor 3 Monaten

I just crammed the updated Gemma 4 26B A4B QAT (MoE) with 180k context into an 8GB RTX 4060 (8 GB VRAM + 16 GB RAM only!!) and optimized the batch size. 23 tokens/sec decode, 300 tokens/sec prefill Yesterday I showed you a Gemma 4 31B dense model running flawlessly on an RTX 4090. Today, we're breaking the VRAM bank on a budget card using Unsloth’s new Gemma 4 26B (A4B) QAT quants. Following Google’s chat template update that boosted agentic benchmarks by +10%, I pushed this model to its absolute limits. Here is how you squeeze 250k context out of 8GB of VRAM. # The Setup & The Optimization - Hardware: Nvidia RTX 4060 (8GB VRAM) + 16GB System RAM - Environment: CUDA 13.0 build of llama.cpp - Model: gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf - Prompt: 28,000 tokens of prompt for each run If you read my L2 cache breakdown (attached in replies), you know the 4060’s 24MB cache maxes out at `-b 1024 -ub 1024`. Push past that, and prefill crashes. I locked those flags in for every test below to ensure maximum GEMM throughput. # 1. The Raw Context Push (Unquantized KV Cache) First, I wanted to see how far pure 8GB VRAM + 16GB RAM could stretch without touching the KV cache: - 80k Context: Prefill 385 t/s | Decode 25.5 t/s - 120k Context: Prefill 270 t/s | Decode 24 t/s llama.cpp flags: .\llama-server -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf -c 120000 --port 8080 -ub 1024 -b 1024 Without KV quantization, 120k is your hard ceiling. push past that prefill throughput drops off a cliff, making the model practically unusable for large agentic workloads. # 2. The Q8 KV Cache Lifeline To survive 250k context on a budget card, you have to quantize the KV cache. I enabled 8 bit KV cache (`-ctk q8_0 -ctv q8_0`) and re ran: - 180k Context: Prefill 280 t/s | Decode 22.8 t/s - 250k Context: Prefill 115 t/s | Decode 20 t/s llama.cpp flags: .\llama-server -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf -c 180000 --port 8080 -b 1024 -ub 1024 -ctk q8_0 -ctv q8_0 Result: Q8 KV cache brings 250k context back from the dead. Decode speed stabilizes at a highly usable 20 t/s. You are trading a very small bit amount of reasoning precision for an extra 130,000 tokens of context window. if you own a single rtx 3050, 3060, 3070, 4050, 4060, 5050 or 5060, you must try this model and optimize your batch size for higher prefill. Hugging Face links to the updated Unsloth's QAT quants and performance graph are in the replies below. What model are you running on your 6GB, 8GB or 12GB cards right now? Let's see your setups.
0:20

I just crammed the updated Gemma 4 26B A4B QAT (MoE) with 180k context into an 8GB RTX 4060 (8 GB VRAM + 16 GB RAM only!!) and optimized the batch size. 23 tokens/sec decode, 300 tokens/sec prefill Yesterday I showed you a Gemma 4 31B dense model running flawlessly on an RTX 4090. Today, we're breaking the VRAM bank on a budget card using Unsloth’s new Gemma 4 26B (A4B) QAT quants. Following Google’s chat template update that boosted agentic benchmarks by +10%, I pushed this model to its absolute limits. Here is how you squeeze 250k context out of 8GB of VRAM. # The Setup & The Optimization - Hardware: Nvidia RTX 4060 (8GB VRAM) + 16GB System RAM - Environment: CUDA 13.0 build of llama.cpp - Model: gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf - Prompt: 28,000 tokens of prompt for each run If you read my L2 cache breakdown (attached in replies), you know the 4060’s 24MB cache maxes out at `-b 1024 -ub 1024`. Push past that, and prefill crashes. I locked those flags in for every test below to ensure maximum GEMM throughput. # 1. The Raw Context Push (Unquantized KV Cache) First, I wanted to see how far pure 8GB VRAM + 16GB RAM could stretch without touching the KV cache: - 80k Context: Prefill 385 t/s | Decode 25.5 t/s - 120k Context: Prefill 270 t/s | Decode 24 t/s llama.cpp flags: .\llama-server -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf -c 120000 --port 8080 -ub 1024 -b 1024 Without KV quantization, 120k is your hard ceiling. push past that prefill throughput drops off a cliff, making the model practically unusable for large agentic workloads. # 2. The Q8 KV Cache Lifeline To survive 250k context on a budget card, you have to quantize the KV cache. I enabled 8 bit KV cache (`-ctk q8_0 -ctv q8_0`) and re ran: - 180k Context: Prefill 280 t/s | Decode 22.8 t/s - 250k Context: Prefill 115 t/s | Decode 20 t/s llama.cpp flags: .\llama-server -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf -c 180000 --port 8080 -b 1024 -ub 1024 -ctk q8_0 -ctv q8_0 Result: Q8 KV cache brings 250k context back from the dead. Decode speed stabilizes at a highly usable 20 t/s. You are trading a very small bit amount of reasoning precision for an extra 130,000 tokens of context window. if you own a single rtx 3050, 3060, 3070, 4050, 4060, 5050 or 5060, you must try this model and optimize your batch size for higher prefill. Hugging Face links to the updated Unsloth's QAT quants and performance graph are in the replies below. What model are you running on your 6GB, 8GB or 12GB cards right now? Let's see your setups.

Alok

36,617 Aufrufe • vor 1 Monat

single RTX 3090. 24 GB VRAM. Qwen3.5-35B-A3B. 4-bit quant, 113 tokens per second at full 262K context harnessing Claude Code locally with no API, no subscription, no proxy. told it what it is. 30 Mamba2 layers, 10 attention, 256 experts, 8 active per token. said "build something that shows off what you can do." it visualized its own architecture. interactive. tokens flowing through layers. 256 experts lighting up on routing. served in the browser from the same GPU running inference. single prompt. then i said level up. 3D. Three.js. separate files. flythrough camera. clickable layers. it planned first, scaffolded 6 files, hit one API bug, fixed it itself, then optimized for smooth framerate. two iterations to a working 3D neural network explorer. llama.cpp just merged a native Anthropic endpoint. Claude Code points at localhost. the whole setup is two commands. no LiteLLM. no proxy config. the open source models coming out of china right now are genuinely changing what's possible on consumer hardware. respect to the Qwen team. this is acceleration.
0:07

single RTX 3090. 24 GB VRAM. Qwen3.5-35B-A3B. 4-bit quant, 113 tokens per second at full 262K context harnessing Claude Code locally with no API, no subscription, no proxy. told it what it is. 30 Mamba2 layers, 10 attention, 256 experts, 8 active per token. said "build something that shows off what you can do." it visualized its own architecture. interactive. tokens flowing through layers. 256 experts lighting up on routing. served in the browser from the same GPU running inference. single prompt. then i said level up. 3D. Three.js. separate files. flythrough camera. clickable layers. it planned first, scaffolded 6 files, hit one API bug, fixed it itself, then optimized for smooth framerate. two iterations to a working 3D neural network explorer. llama.cpp just merged a native Anthropic endpoint. Claude Code points at localhost. the whole setup is two commands. no LiteLLM. no proxy config. the open source models coming out of china right now are genuinely changing what's possible on consumer hardware. respect to the Qwen team. this is acceleration.

Sudo su

110,206 Aufrufe • vor 6 Monaten

i pointed hermes agent at nvidia's nemotron cascade 2 30B-A3B on a single RTX 3090 24GB. IQ4_XS quant by bartowski, 187 tok/s, 625K context. had it discover its own hardware, create an identity file, then build a full GPU marketplace UI from a single prompt. it one shotted it. first attempt no iteration. qwen 3.5 35B-A3B on the same hardware same 3090 24GB took an iteration to recover from a blank screen on the same type of build. 24 days between these two models releasing. same active parameters, completely different architectures and cascade 2 through hermes agent just keeps going. this model goes on and on. feast your eyes. more iterations and tests dropping soon. nvidia really cooked. no special flags needed. nvidia optimized this mamba MoE so well it just runs. flash attention auto enabled, context auto allocated. the model does the work not the config. but i compiled llama.cpp from source and i'm not sure how it performs on other engines. if you ran nemotron on any hardware drop your numbers below. RTX, AMD, Mac, whatever. model, quant, tok/s, engine. i want to see if it holds everywhere or just on llama.cpp.
1:31

i pointed hermes agent at nvidia's nemotron cascade 2 30B-A3B on a single RTX 3090 24GB. IQ4_XS quant by bartowski, 187 tok/s, 625K context. had it discover its own hardware, create an identity file, then build a full GPU marketplace UI from a single prompt. it one shotted it. first attempt no iteration. qwen 3.5 35B-A3B on the same hardware same 3090 24GB took an iteration to recover from a blank screen on the same type of build. 24 days between these two models releasing. same active parameters, completely different architectures and cascade 2 through hermes agent just keeps going. this model goes on and on. feast your eyes. more iterations and tests dropping soon. nvidia really cooked. no special flags needed. nvidia optimized this mamba MoE so well it just runs. flash attention auto enabled, context auto allocated. the model does the work not the config. but i compiled llama.cpp from source and i'm not sure how it performs on other engines. if you ran nemotron on any hardware drop your numbers below. RTX, AMD, Mac, whatever. model, quant, tok/s, engine. i want to see if it holds everywhere or just on llama.cpp.

Sudo su

71,112 Aufrufe • vor 5 Monaten

Run Updated Gemma 4 26B A4B QAT (MoE) with Vision at 25 tokens/sec and massive 120k context window on a single RTX 4060 (8 GB VRAM + 16 GB RAM Only!!) Yesterday I pushed Gemma 4 26B A4B QAT to 250k context on a single RTX 4060 using nothing but Q8 KV cache and optimized -b and -ub flags for higher prefill throughput. Today I stacked Multi Token Prediction (MTP) self speculative decoding AND the vision projector (mmproj) on top of that same card, same batch size optimization, same $250 GPU and pushed it until it broke, then found the fix. All text only runs consist of a 28k prompt. vision runs consist of 28k text prompt + an image. # 1. MTP alone. near free decode speed, no catch MTP draft assistant is a separate small model (MTP heads are backed into the main model itself for the qwen 3.5+ models but its a separate small model for gemma 4 series), 240 MB gguf 80k ctx: Prefill 510 t/s | Decode 29.5 t/s 120k ctx: Prefill 433 t/s | Decode 29 t/s 180k ctx: Prefill 240 t/s | Decode 24.9 t/s 250k ctx: Prefill 63 t/s | Decode 13 t/s llama.cpp flags: m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf --spec-type draft-mtp -md mtp-gemma-4-26B-A4B-it.gguf-c 180000 -b 1024 -ub 1024 --spec-draft-n-max 6 --spec-draft-p-min 0.7 -ctk q8_0 -ctv q8_0 # 2. Add vision on top. the tax you actually pay the vision projector gguf is about 1.1 GBs 80k ctx: Prefill 360 t/s | Decode 25.4 t/s 120k ctx: Prefill 230 t/s | Decode 23.8 t/s 180k ctx (Q8 KV): Prefill 75 t/s | Decode 12.5 t/s - cliff flags: -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf --spec-type draft-mtp -md mtp-gemma-4-26B-A4B-it.gguf -c 80000 --port 8080 -b 1024 -ub 1024 --spec-draft-n-max 6 --spec-draft-p-min 0.7 -ctk q8_0 -ctv q8_0 --mmproj mmproj-F16.gguf # 3. The fix if you want to run vision over 120k context: swap Q8 KV for Q4 KV past 120k Stack MTP + vision + Q8 KV past 120k context and you hit a wall. draft model overhead plus KV pressure tanks everything. Drop to Q4 KV and the wall disappears: 180k ctx (Q4 KV): Prefill 220 t/s | Decode 25.5 t/s -ctk q4_0 -ctv q4_0 --mmproj mmproj-F16.gguf (rest same as above) Bottom line: MTP gives you a near free +20-30% decode boost up to 120k context. Past that, it's fighting your VRAM, not helping and if vision is loaded too, Q4 KV isn't optional past 120k, it's mandatory. 30% boost is model and card specific, MTP boosted decode 2x for gemma 4 31b on a single rtx 4090. Same 8GB card. Same $250 GPU. Multimodal, speculative decoding, 180k usable context, zero upgrades. You gotta try this if you have a single NVIDIA RTX 3050, 3060, 3070, 4050, 4060, 5050 or 5060. You can try it with a 6 GB VRAM card as well but you will have to lower the context window. Hugging Face links to the updated Unsloth's QAT quants and performance graph are in the replies below. Which models are you running on your 6/8/12GB cards with MTP?
0:20

Run Updated Gemma 4 26B A4B QAT (MoE) with Vision at 25 tokens/sec and massive 120k context window on a single RTX 4060 (8 GB VRAM + 16 GB RAM Only!!) Yesterday I pushed Gemma 4 26B A4B QAT to 250k context on a single RTX 4060 using nothing but Q8 KV cache and optimized -b and -ub flags for higher prefill throughput. Today I stacked Multi Token Prediction (MTP) self speculative decoding AND the vision projector (mmproj) on top of that same card, same batch size optimization, same $250 GPU and pushed it until it broke, then found the fix. All text only runs consist of a 28k prompt. vision runs consist of 28k text prompt + an image. # 1. MTP alone. near free decode speed, no catch MTP draft assistant is a separate small model (MTP heads are backed into the main model itself for the qwen 3.5+ models but its a separate small model for gemma 4 series), 240 MB gguf 80k ctx: Prefill 510 t/s | Decode 29.5 t/s 120k ctx: Prefill 433 t/s | Decode 29 t/s 180k ctx: Prefill 240 t/s | Decode 24.9 t/s 250k ctx: Prefill 63 t/s | Decode 13 t/s llama.cpp flags: m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf --spec-type draft-mtp -md mtp-gemma-4-26B-A4B-it.gguf-c 180000 -b 1024 -ub 1024 --spec-draft-n-max 6 --spec-draft-p-min 0.7 -ctk q8_0 -ctv q8_0 # 2. Add vision on top. the tax you actually pay the vision projector gguf is about 1.1 GBs 80k ctx: Prefill 360 t/s | Decode 25.4 t/s 120k ctx: Prefill 230 t/s | Decode 23.8 t/s 180k ctx (Q8 KV): Prefill 75 t/s | Decode 12.5 t/s - cliff flags: -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf --spec-type draft-mtp -md mtp-gemma-4-26B-A4B-it.gguf -c 80000 --port 8080 -b 1024 -ub 1024 --spec-draft-n-max 6 --spec-draft-p-min 0.7 -ctk q8_0 -ctv q8_0 --mmproj mmproj-F16.gguf # 3. The fix if you want to run vision over 120k context: swap Q8 KV for Q4 KV past 120k Stack MTP + vision + Q8 KV past 120k context and you hit a wall. draft model overhead plus KV pressure tanks everything. Drop to Q4 KV and the wall disappears: 180k ctx (Q4 KV): Prefill 220 t/s | Decode 25.5 t/s -ctk q4_0 -ctv q4_0 --mmproj mmproj-F16.gguf (rest same as above) Bottom line: MTP gives you a near free +20-30% decode boost up to 120k context. Past that, it's fighting your VRAM, not helping and if vision is loaded too, Q4 KV isn't optional past 120k, it's mandatory. 30% boost is model and card specific, MTP boosted decode 2x for gemma 4 31b on a single rtx 4090. Same 8GB card. Same $250 GPU. Multimodal, speculative decoding, 180k usable context, zero upgrades. You gotta try this if you have a single NVIDIA RTX 3050, 3060, 3070, 4050, 4060, 5050 or 5060. You can try it with a 6 GB VRAM card as well but you will have to lower the context window. Hugging Face links to the updated Unsloth's QAT quants and performance graph are in the replies below. Which models are you running on your 6/8/12GB cards with MTP?

Alok

16,266 Aufrufe • vor 1 Monat

Deepseek V4 Flash 0731 (Q2) - 12 tokens/sec - Single RTX 4090 - 650+ tokens/sec prefill - 250k context - no kv cache quantization! DeepSeek just dropped the official V4 Flash 0731 two days ago with a massive agent capabilities upgrade. The official benchmarks are literally crushing their own V4-Pro-Preview on agentic tasks like Terminal Bench 2.1 and DeepSWE. Unsloth AI said they couldn't wait to bring it to local devices, and they delivered. If you thought my 118B Poolside Laguna S 2.1 MoE run last week on a single GPU was wild, hold onto your hardware. I just successfully ran Unsloth’s brand new 91GB DeepSeek-V4-Flash-0731 (UD-IQ2_M) GGUF entirely locally. And I pushed it to a mind-bending 250,000 context window. The VRAM ceiling is an illusion if you know how to optimize llama.cpp. Here are the benchmarks and the cheat codes to run a local frontier class model yourself. For the hardware and setup, I used a single NVIDIA RTX 4090 (24GB VRAM) hooked up via a PCIe 4 bus, running Ubuntu 22.04 LTS and CUDA 13.0. You don't need a massive enterprise server for this, if you have more than 80 GB of standard DDR4 RAM and a 24GB card like an RTX 3090 or 4090, you can run this exact stack yourself. All benchmarks were run using a massive 28k token prompt to truly stress test the prefill limits. no kv cache quantization THE BENCHMARKS (Scaling Context): # 80k Context (Baseline: -b 2048 -ub 2048): Prefill: 465.43 t/s | Decode: 13.00 t/s | VRAM: 22.87 GB # 80k Context (Optimized: -b 4096 -ub 4096): Prefill: 643.15 t/s | Decode: 12.20 t/s | VRAM: 23.00 GB (Notice how doubling the batch flags spiked my prefill throughput by nearly 200 t/s with almost zero VRAM penalty) # 180k Context (-b 4096 -ub 4096): Prefill: 629.18 t/s | Decode: 11.92 t/s | VRAM: 23.40 GB # 250k Context MAXIMUM (-b 4096 -ub 4096): Prefill: 619.02 t/s | Decode: 11.54 t/s | VRAM: 23.40 GB # THE SECRET SAUCE (Why this works): Unsloth’s UD-IQ2_M quant is ~91GB across 3 files. Since I only have 24GB of VRAM, the PCIe 4 bus and system RAM have to do the heavy lifting. The magic bullet is the --no-mmap flag. By completely bypassing OS disk paging, I forced llama.cpp to load the massive model weights directly into the system RAM upfront. Combined with Flash Attention (-fa on) and exactly 12 CPU threads (--threads 12), I maintained an incredibly stable 11.5+ tokens/sec decode speed even at a quarter million token context. # THE EXACT COMMAND: ./build/bin/llama-server -m /workspace/models/DeepSeek-V4-Flash-0731-UD-IQ2_M-00001-of-00003.gguf -c 250000 -fa on --port 8080 --threads 12 -b 4096 -ub 4096 --no-mmap -v Local conversational and agentic coding AI is fully here. You don’t need an API or an H100 cluster. Qwen 3.8 27b drops next week making the 24GB VRAM tier even more worthwhile. What does your current local AI rig look like, and what's the craziest model you've managed to squeeze into it? Official huggingface GGUF links from Unsloth and performance graphs are dropped in the replies below!
0:21

Deepseek V4 Flash 0731 (Q2) - 12 tokens/sec - Single RTX 4090 - 650+ tokens/sec prefill - 250k context - no kv cache quantization! DeepSeek just dropped the official V4 Flash 0731 two days ago with a massive agent capabilities upgrade. The official benchmarks are literally crushing their own V4-Pro-Preview on agentic tasks like Terminal Bench 2.1 and DeepSWE. Unsloth AI said they couldn't wait to bring it to local devices, and they delivered. If you thought my 118B Poolside Laguna S 2.1 MoE run last week on a single GPU was wild, hold onto your hardware. I just successfully ran Unsloth’s brand new 91GB DeepSeek-V4-Flash-0731 (UD-IQ2_M) GGUF entirely locally. And I pushed it to a mind-bending 250,000 context window. The VRAM ceiling is an illusion if you know how to optimize llama.cpp. Here are the benchmarks and the cheat codes to run a local frontier class model yourself. For the hardware and setup, I used a single NVIDIA RTX 4090 (24GB VRAM) hooked up via a PCIe 4 bus, running Ubuntu 22.04 LTS and CUDA 13.0. You don't need a massive enterprise server for this, if you have more than 80 GB of standard DDR4 RAM and a 24GB card like an RTX 3090 or 4090, you can run this exact stack yourself. All benchmarks were run using a massive 28k token prompt to truly stress test the prefill limits. no kv cache quantization THE BENCHMARKS (Scaling Context): # 80k Context (Baseline: -b 2048 -ub 2048): Prefill: 465.43 t/s | Decode: 13.00 t/s | VRAM: 22.87 GB # 80k Context (Optimized: -b 4096 -ub 4096): Prefill: 643.15 t/s | Decode: 12.20 t/s | VRAM: 23.00 GB (Notice how doubling the batch flags spiked my prefill throughput by nearly 200 t/s with almost zero VRAM penalty) # 180k Context (-b 4096 -ub 4096): Prefill: 629.18 t/s | Decode: 11.92 t/s | VRAM: 23.40 GB # 250k Context MAXIMUM (-b 4096 -ub 4096): Prefill: 619.02 t/s | Decode: 11.54 t/s | VRAM: 23.40 GB # THE SECRET SAUCE (Why this works): Unsloth’s UD-IQ2_M quant is ~91GB across 3 files. Since I only have 24GB of VRAM, the PCIe 4 bus and system RAM have to do the heavy lifting. The magic bullet is the --no-mmap flag. By completely bypassing OS disk paging, I forced llama.cpp to load the massive model weights directly into the system RAM upfront. Combined with Flash Attention (-fa on) and exactly 12 CPU threads (--threads 12), I maintained an incredibly stable 11.5+ tokens/sec decode speed even at a quarter million token context. # THE EXACT COMMAND: ./build/bin/llama-server -m /workspace/models/DeepSeek-V4-Flash-0731-UD-IQ2_M-00001-of-00003.gguf -c 250000 -fa on --port 8080 --threads 12 -b 4096 -ub 4096 --no-mmap -v Local conversational and agentic coding AI is fully here. You don’t need an API or an H100 cluster. Qwen 3.8 27b drops next week making the 24GB VRAM tier even more worthwhile. What does your current local AI rig look like, and what's the craziest model you've managed to squeeze into it? Official huggingface GGUF links from Unsloth and performance graphs are dropped in the replies below!

Alok

46,100 Aufrufe • vor 1 Monat

The VRAM barrier is officially dead. I just ran Qwen 3.8 Flash Next (MoE) 125B A6B with a 250,000 context window on a single 24GB RTX 4090. 21 tokens/sec decode. 364 t/s prefill. no mtp. no dflash. no kv cache quantization! We are running datacenter models on consumer hardware. Tested on Ubuntu 22 | CUDA 13.0 | PCIe 4.0 x16 | 110 GB DDR4 System RAM with a continuous 28k prompt across all runs. ### The Benchmarks & Scaling # 1. Hybrid Offload (-ncmoe 40 @ 80k Context) Offloaded 40 expert layers to the GPU, pushing VRAM to the ceiling. ./build/bin/llama-server -m Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00004.gguf -c 80000 --port 8080 -v --fit off -b 4096 -ub 4096 -ncmoe 40 Prefill: 383.85 t/s | Decode: 22.52 t/s Footprint: 23.85 GB VRAM | 97 GB RAM # 2. Full CPU MoE Offload (-cmoe @ 80k Context) Pinned all 512 expert layers to DDR4 RAM (-cmoe), keeping attention on the 4090. llama.cpp flags: (Same as above, replace -ncmoe 40 with -cmoe) Prefill: 355.72 t/s | Decode: 20.84 t/s Footprint: 11.66 GB VRAM (12GB+ VRAM freed up!) | 110 GB RAM # 3. The 180,000 Context Run Prefill: 357.75 t/s | Decode: 20.98 t/s | VRAM: 15.6 GB | RAM: 110 GB # 4. The 250,000 Context Absolute Ceiling ./build/bin/llama-server -m Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00004.gguf -c 250000 --port 8080 -v --fit off -b 4096 -ub 4096 -cmoe Prefill: 364.29 t/s | Decode: 20.97 t/s Footprint: 18.3 GB VRAM (Still ~5.7 GB of VRAM headroom!) | 110 GB RAM ### Key Insights: -b 4096 -ub 4096: doubles the prompt ingestion from ~150 to 364+ t/s. -cmoe Free Lunch: Shifting expert layers to DDR4 RAM slashes VRAM from 24GB to 11.6GB with virtually zero decode penalty (22.5 -> 20.9 t/s), enabling the 250k context ceiling. Qwen 3.8 Flash-Next (UD-Q4_K_XL) is a massive 111.4 GB model split across 4 shards. To run this architecture, you must build from the experimental PR branch (#27742) by Daniel Han: git clone && cd llama.cpp git fetch origin pull/27742/head:qwen-next && git checkout qwen-next cmake -B build -DGGML_CUDA=ON -DCMAKE_CUDA_ARCHITECTURES=native -DBUILD_SHARED_LIBS=OFF cmake --build build --config Release -j $(nproc) --target llama-server A single 4090 paired with 100 GB of cheap DDR4 RAM will comfortably serve production grade 125B inference. While Qwen 3.8 27B (dense) still holds the crown for single 3090/4090 rigs, Flash Next proves 125B hybrid models are officially viable on consumer hardware. Hugging Face GGUF link and complete performance telemetry graphs are dropped in the replies below. GLM 5.3 Flash VS Qwen 3.8 Flash Next, which one takes the open weights crown this week?
0:20

The VRAM barrier is officially dead. I just ran Qwen 3.8 Flash Next (MoE) 125B A6B with a 250,000 context window on a single 24GB RTX 4090. 21 tokens/sec decode. 364 t/s prefill. no mtp. no dflash. no kv cache quantization! We are running datacenter models on consumer hardware. Tested on Ubuntu 22 | CUDA 13.0 | PCIe 4.0 x16 | 110 GB DDR4 System RAM with a continuous 28k prompt across all runs. ### The Benchmarks & Scaling # 1. Hybrid Offload (-ncmoe 40 @ 80k Context) Offloaded 40 expert layers to the GPU, pushing VRAM to the ceiling. ./build/bin/llama-server -m Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00004.gguf -c 80000 --port 8080 -v --fit off -b 4096 -ub 4096 -ncmoe 40 Prefill: 383.85 t/s | Decode: 22.52 t/s Footprint: 23.85 GB VRAM | 97 GB RAM # 2. Full CPU MoE Offload (-cmoe @ 80k Context) Pinned all 512 expert layers to DDR4 RAM (-cmoe), keeping attention on the 4090. llama.cpp flags: (Same as above, replace -ncmoe 40 with -cmoe) Prefill: 355.72 t/s | Decode: 20.84 t/s Footprint: 11.66 GB VRAM (12GB+ VRAM freed up!) | 110 GB RAM # 3. The 180,000 Context Run Prefill: 357.75 t/s | Decode: 20.98 t/s | VRAM: 15.6 GB | RAM: 110 GB # 4. The 250,000 Context Absolute Ceiling ./build/bin/llama-server -m Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00004.gguf -c 250000 --port 8080 -v --fit off -b 4096 -ub 4096 -cmoe Prefill: 364.29 t/s | Decode: 20.97 t/s Footprint: 18.3 GB VRAM (Still ~5.7 GB of VRAM headroom!) | 110 GB RAM ### Key Insights: -b 4096 -ub 4096: doubles the prompt ingestion from ~150 to 364+ t/s. -cmoe Free Lunch: Shifting expert layers to DDR4 RAM slashes VRAM from 24GB to 11.6GB with virtually zero decode penalty (22.5 -> 20.9 t/s), enabling the 250k context ceiling. Qwen 3.8 Flash-Next (UD-Q4_K_XL) is a massive 111.4 GB model split across 4 shards. To run this architecture, you must build from the experimental PR branch (#27742) by Daniel Han: git clone && cd llama.cpp git fetch origin pull/27742/head:qwen-next && git checkout qwen-next cmake -B build -DGGML_CUDA=ON -DCMAKE_CUDA_ARCHITECTURES=native -DBUILD_SHARED_LIBS=OFF cmake --build build --config Release -j $(nproc) --target llama-server A single 4090 paired with 100 GB of cheap DDR4 RAM will comfortably serve production grade 125B inference. While Qwen 3.8 27B (dense) still holds the crown for single 3090/4090 rigs, Flash Next proves 125B hybrid models are officially viable on consumer hardware. Hugging Face GGUF link and complete performance telemetry graphs are dropped in the replies below. GLM 5.3 Flash VS Qwen 3.8 Flash Next, which one takes the open weights crown this week?

Alok

1,013,451 Aufrufe • vor 20 Tagen

The Cost of Intelligence is Heading to Zero | Hyperspace P2P Distributed Cache We present to you our breakthrough cross-domain work across AI, distributed systems, cryptography, game theory to solve the primary structural inefficiency at the heart of AI infrastructure: most inference is redundant. Google has reported that only 15% of daily searches are truly novel. The rest are repeats or close variants. LLM inference inherits this same power-law distribution. Enterprise chatbots see 70-80% of queries fall into a handful of intent categories. System prompts are identical across 100% of requests within an application. The KV attention state for "You are a helpful assistant" has been computed billions of times, on millions of GPUs, identically. And yet every AI lab, every startup, every self-hosted deployment - computes and caches these results independently. There is no shared layer. No global memory. Every provider pays the full compute cost for every query, even when the answer already exists somewhere in the network. This is the problem Hyperspace solves where distributed cache operates at three levels, each catching a different class of redundancy: 1. Response cache Same prompt, same model, same parameters - instant cached response from any node in the network. SHA-256 hash lookup via DHT, with cryptographic cache proofs linking every response to its original inference execution. No trust required. Fetchers re-announce as providers, so popular responses replicate naturally across more nodes. 2. KV prefix cache Same system prompt tokens - skip the most expensive part of inference entirely. Prefill (computing Key-Value attention states) is deterministic: same model plus same tokens always produces identical KV state. The network caches these states using erasure coding and distributes them via the routing network. New questions that share a common prefix resume generation from cached state instead of recomputing from scratch. 3. Routing to cached nodes Instead of transferring KV state across the network for every request, Hyperspace routes the request to the node that already has the state loaded in VRAM. The request goes to the cache, not the cache to the request. Together, these three layers mean that 70-90% of inference requests at network scale never require full GPU computation. This work doesn't exist in isolation. It builds on research from across the industry: SGLang's RadixAttention demonstrated that automatic prefix sharing can yield up to 5x speedup on structured LLM workloads. Moonshot AI's Mooncake built an entire KV-cache-centric disaggregated architecture for production serving at Kimi. Anthropic, OpenAI, and Google all launched prompt caching products in 2024 - priced at 50-90% discounts - because system prompt reuse is so pervasive that it changes the economics of inference. What all of these systems share is a common limitation: they operate within a single organization's infrastructure. SGLang caches prefixes within one server. Mooncake disaggregates KV cache within one datacenter. Anthropic's prompt caching works within one API provider's fleet. None of them can share cached state across organizational boundaries. Hyperspace removes this boundary. The cache is global. A response computed by a node in Tokyo is immediately available to a node in Berlin. A KV prefix state generated for Qwen-32B on one machine is verifiable and reusable by any other machine running the same model. The routing network provides the delivery guarantees, the erasure coding provides the redundancy, and the cache proofs provide the trust. What this means for the cost of intelligence Big AI labs scale linearly: twice the users means twice the GPU spend. Every query is a cost center. Their internal caching helps, but it's siloed - Lab A's cache can't serve Lab B's users, and neither can serve a self-hosted Llama deployment. Hyperspace scales sub-linearly. Every new node that joins the network adds to the global cache. Every inference result enriches the cache for all future requests. The cache hit rate rises with network size because query distributions follow a power law - the most common questions are asked exponentially more often than rare ones. The implication is simple: as the network grows, the effective cost per inference drops. Not linearly. Logarithmically. At 10 million nodes, we estimate 75-90% of all inference requests can be served from cache, eliminating 400,000+ MWh of energy consumption per year and avoiding over 200,000 tons of CO2 emissions. The first person to ask a question pays the compute cost. Everyone after them gets the answer for free, with cryptographic proof that it's authentic. Training is competitive. Inference is shared Open-weight models are converging on quality with closed models. Labs will continue to differentiate on training - data curation, architecture innovation, RLHF tuning. That's where the real intellectual property lives. But inference is a commodity. Two copies of Qwen-32B running the same prompt produce the same KV state and the same response, byte for byte, regardless of whose GPU runs the matrix multiplication. There is no moat in multiplying matrices. The moat is in training the weights. A global distributed cache makes this separation explicit. It doesn't matter who trained the model. Once the weights are open, the inference cost approaches zero at scale - because the network remembers every answer and can prove it's correct. No lab, no matter how well-funded, can match this. They cannot share caches across competitors. They scale linearly. The network scales logarithmically. The marginal cost of intelligence approaches zero. That's the endgame.
0:23

The Cost of Intelligence is Heading to Zero | Hyperspace P2P Distributed Cache We present to you our breakthrough cross-domain work across AI, distributed systems, cryptography, game theory to solve the primary structural inefficiency at the heart of AI infrastructure: most inference is redundant. Google has reported that only 15% of daily searches are truly novel. The rest are repeats or close variants. LLM inference inherits this same power-law distribution. Enterprise chatbots see 70-80% of queries fall into a handful of intent categories. System prompts are identical across 100% of requests within an application. The KV attention state for "You are a helpful assistant" has been computed billions of times, on millions of GPUs, identically. And yet every AI lab, every startup, every self-hosted deployment - computes and caches these results independently. There is no shared layer. No global memory. Every provider pays the full compute cost for every query, even when the answer already exists somewhere in the network. This is the problem Hyperspace solves where distributed cache operates at three levels, each catching a different class of redundancy: 1. Response cache Same prompt, same model, same parameters - instant cached response from any node in the network. SHA-256 hash lookup via DHT, with cryptographic cache proofs linking every response to its original inference execution. No trust required. Fetchers re-announce as providers, so popular responses replicate naturally across more nodes. 2. KV prefix cache Same system prompt tokens - skip the most expensive part of inference entirely. Prefill (computing Key-Value attention states) is deterministic: same model plus same tokens always produces identical KV state. The network caches these states using erasure coding and distributes them via the routing network. New questions that share a common prefix resume generation from cached state instead of recomputing from scratch. 3. Routing to cached nodes Instead of transferring KV state across the network for every request, Hyperspace routes the request to the node that already has the state loaded in VRAM. The request goes to the cache, not the cache to the request. Together, these three layers mean that 70-90% of inference requests at network scale never require full GPU computation. This work doesn't exist in isolation. It builds on research from across the industry: SGLang's RadixAttention demonstrated that automatic prefix sharing can yield up to 5x speedup on structured LLM workloads. Moonshot AI's Mooncake built an entire KV-cache-centric disaggregated architecture for production serving at Kimi. Anthropic, OpenAI, and Google all launched prompt caching products in 2024 - priced at 50-90% discounts - because system prompt reuse is so pervasive that it changes the economics of inference. What all of these systems share is a common limitation: they operate within a single organization's infrastructure. SGLang caches prefixes within one server. Mooncake disaggregates KV cache within one datacenter. Anthropic's prompt caching works within one API provider's fleet. None of them can share cached state across organizational boundaries. Hyperspace removes this boundary. The cache is global. A response computed by a node in Tokyo is immediately available to a node in Berlin. A KV prefix state generated for Qwen-32B on one machine is verifiable and reusable by any other machine running the same model. The routing network provides the delivery guarantees, the erasure coding provides the redundancy, and the cache proofs provide the trust. What this means for the cost of intelligence Big AI labs scale linearly: twice the users means twice the GPU spend. Every query is a cost center. Their internal caching helps, but it's siloed - Lab A's cache can't serve Lab B's users, and neither can serve a self-hosted Llama deployment. Hyperspace scales sub-linearly. Every new node that joins the network adds to the global cache. Every inference result enriches the cache for all future requests. The cache hit rate rises with network size because query distributions follow a power law - the most common questions are asked exponentially more often than rare ones. The implication is simple: as the network grows, the effective cost per inference drops. Not linearly. Logarithmically. At 10 million nodes, we estimate 75-90% of all inference requests can be served from cache, eliminating 400,000+ MWh of energy consumption per year and avoiding over 200,000 tons of CO2 emissions. The first person to ask a question pays the compute cost. Everyone after them gets the answer for free, with cryptographic proof that it's authentic. Training is competitive. Inference is shared Open-weight models are converging on quality with closed models. Labs will continue to differentiate on training - data curation, architecture innovation, RLHF tuning. That's where the real intellectual property lives. But inference is a commodity. Two copies of Qwen-32B running the same prompt produce the same KV state and the same response, byte for byte, regardless of whose GPU runs the matrix multiplication. There is no moat in multiplying matrices. The moat is in training the weights. A global distributed cache makes this separation explicit. It doesn't matter who trained the model. Once the weights are open, the inference cost approaches zero at scale - because the network remembers every answer and can prove it's correct. No lab, no matter how well-funded, can match this. They cannot share caches across competitors. They scale linearly. The network scales logarithmically. The marginal cost of intelligence approaches zero. That's the endgame.

Varun

37,555 Aufrufe • vor 5 Monaten

If you are running local LLMs without N-gram speculative decoding, you are wasting massive amounts of compute. Whether your AI is editing a document, outputting structured JSON, or rewriting boilerplate templates, a huge chunk of the text it generates is highly repetitive or already exists right there in the prompt. Standard decoding wastes expensive GPU compute cycles "re thinking" every single token. By adding one hidden flag in llama.cpp, you can instantly fast forward through the repetition. Zero draft models. Zero extra VRAM. And virtually zero compute overhead. Google Colab hands you an enterprise grade NVIDIA Tesla T4 GPU with 16GB of VRAM for free. It’s the perfect Ubuntu Linux sandbox to build a bleeding edge inference engine from scratch. Recently, I showed you how to double your local speeds using MTP (Multi Token Prediction). But MTP requires a secondary neural network draft model. That eats into your precious VRAM (slightly though) and burns extra compute for every guess it makes. N-gram Speculative Decoding gives you a massive speed boost for exactly 0 memory cost and minimal compute. And it's faster than MTP when it works. Here is how it actually works under the hood: Standard autoregressive decoding is slow because it predicts one token at a time. If you ask an agent to format a long JSON object or update one line in an HTML file, it runs heavy matrix multiplications to calculate the probability of every single bracket, space, and letter from scratch. N-gram changes the game. It acts as a lightweight caching system. Instead of running heavy neural network math to guess the next word, it uses a simple hash table. Whenever the LLM starts outputting a sequence of tokens that already exists anywhere in its context window, N-gram instantly recognizes the pattern. Because it is just doing lightning fast string matching, the compute cost is practically zero. It "fast forwards" through the text, drafting the boilerplate instantly from memory, and the main model just verifies it in parallel. Pure speed. Using quantized GGUFs from Unsloth via HuggingFace, I spun up DeepMind’s massive Gemma 4 26B A4B QAT MoE on a free Colab instance to test this. Just look at the raw benchmark data on code editing task: Without N-gram: [ Prompt: 638.6 t/s | Generation: 45.9 t/s ] With N-gram: [ Prompt: 601.9 t/s | Generation: 107.1 t/s ] Here is the exact llama.cpp CLI command to activate it. Notice we don't even need the --model-draft flag: ./llama-cli -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf -cnv -n 6000 -c 12000 -ngl 99 -fa on --spec-type ngram-mod Stop waiting for your GPU to re calculate words it already knows. I’ve built a free, interactive, cell by cell Google Colab notebook that lets you test this live in your browser. You can literally chat with the model and watch the text generation speed absolutely fly on the second turn when you ask it to edit a file. There are additional parameters for ngram-mod that you can tune once you get it working with the single flag. Link to the free Colab Notebook is in the comments below. It walks you through the entire stack: pulling pre built llama.cpp CUDA binaries for Linux, fetching GGUFs from HuggingFace, and spinning up the inference engine with ngram-mod from scratch. Let me know if you have already tried ngram-mod
0:23

If you are running local LLMs without N-gram speculative decoding, you are wasting massive amounts of compute. Whether your AI is editing a document, outputting structured JSON, or rewriting boilerplate templates, a huge chunk of the text it generates is highly repetitive or already exists right there in the prompt. Standard decoding wastes expensive GPU compute cycles "re thinking" every single token. By adding one hidden flag in llama.cpp, you can instantly fast forward through the repetition. Zero draft models. Zero extra VRAM. And virtually zero compute overhead. Google Colab hands you an enterprise grade NVIDIA Tesla T4 GPU with 16GB of VRAM for free. It’s the perfect Ubuntu Linux sandbox to build a bleeding edge inference engine from scratch. Recently, I showed you how to double your local speeds using MTP (Multi Token Prediction). But MTP requires a secondary neural network draft model. That eats into your precious VRAM (slightly though) and burns extra compute for every guess it makes. N-gram Speculative Decoding gives you a massive speed boost for exactly 0 memory cost and minimal compute. And it's faster than MTP when it works. Here is how it actually works under the hood: Standard autoregressive decoding is slow because it predicts one token at a time. If you ask an agent to format a long JSON object or update one line in an HTML file, it runs heavy matrix multiplications to calculate the probability of every single bracket, space, and letter from scratch. N-gram changes the game. It acts as a lightweight caching system. Instead of running heavy neural network math to guess the next word, it uses a simple hash table. Whenever the LLM starts outputting a sequence of tokens that already exists anywhere in its context window, N-gram instantly recognizes the pattern. Because it is just doing lightning fast string matching, the compute cost is practically zero. It "fast forwards" through the text, drafting the boilerplate instantly from memory, and the main model just verifies it in parallel. Pure speed. Using quantized GGUFs from Unsloth via HuggingFace, I spun up DeepMind’s massive Gemma 4 26B A4B QAT MoE on a free Colab instance to test this. Just look at the raw benchmark data on code editing task: Without N-gram: [ Prompt: 638.6 t/s | Generation: 45.9 t/s ] With N-gram: [ Prompt: 601.9 t/s | Generation: 107.1 t/s ] Here is the exact llama.cpp CLI command to activate it. Notice we don't even need the --model-draft flag: ./llama-cli -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf -cnv -n 6000 -c 12000 -ngl 99 -fa on --spec-type ngram-mod Stop waiting for your GPU to re calculate words it already knows. I’ve built a free, interactive, cell by cell Google Colab notebook that lets you test this live in your browser. You can literally chat with the model and watch the text generation speed absolutely fly on the second turn when you ask it to edit a file. There are additional parameters for ngram-mod that you can tune once you get it working with the single flag. Link to the free Colab Notebook is in the comments below. It walks you through the entire stack: pulling pre built llama.cpp CUDA binaries for Linux, fetching GGUFs from HuggingFace, and spinning up the inference engine with ngram-mod from scratch. Let me know if you have already tried ngram-mod

Alok

31,765 Aufrufe • vor 2 Monaten

UC Berkeley just open-sourced FreeToken. (2–4x faster local LLM inference than Ollama) the results are wild: - Qwen3.6-35B on an 8GB GPU at 39.3 tokens/s - DeepSeek-V4-Flash 284B on a 32GB GPU at 22 tokens/s - GLM-5.2 753B on a 96GB GPU at 14.9 tokens/s a 35B model at 16-bit precision needs about 70GB just for its weights. even at 4 bits it is close to 18GB, and FreeToken serves it on an 8GB GPU. let me explain how: all three models mentioned above are Mixture-of-Experts, and that is what FreeToken takes advantage of. each layer holds hundreds of separate experts plus a small router that picks a few of them per token. Qwen3.6-35B activates roughly 3B of its 35B parameters per token. DeepSeek-V4-Flash picks 6 of 256 experts per layer, so 13B of its 284B run at a time. so compute was never the bottleneck. the weights a single step touches fit comfortably on a consumer GPU. every expert the router might pick still has to exist somewhere. they sit in system RAM, and the GPU keeps a cache of the ones the model has been using recently. so everything comes down to what happens when the router picks an expert that is not on the GPU. there are two ways to serve that miss: 1. copy it over PCIe and run it on the GPU 2. run it on the CPU, where it already lives both read from the same system memory, so they compete for one pool of bandwidth instead of adding to each other. existing engines pick one option and freeze it when the model loads. but routing changes on every token, so a fixed choice misses most of what the model asks for. FreeToken measures both bandwidths on your machine and splits each step's misses between the two paths in proportion. the GPU and CPU results then merge exactly, with no approximation. two machines with the same GPU can end up wanting opposite strategies, which I did not expect. a 5090 in a gaming desktop should push nearly everything over PCIe, while an 8GB laptop is better off computing most misses on the CPU. none of that is readable off a spec sheet, so the engine profiles it once per machine. the second half of the design is about agents. coding agents constantly rewrite their own history, and every edit normally forces thousands of tokens back through prefill. FreeToken saves its checkpoints at the exact boundaries agent frameworks cut on, so it only reprocesses the new part. its slowest first token stays under 44 seconds, while llama.cpp peaks at 232 and KTransformers at 946. it serves the OpenAI and Anthropic APIs under Apache 2.0, so Claude Code and Codex can point at it directly. releasing weights publicly decides who can download a model, not who can afford to run one. frontier open models keep shipping, and running them still assumes a rented cluster. meanwhile there are over a hundred million consumer machines with discrete GPUs sitting mostly idle. closing that gap was never a hardware problem, and work like this is what turns open weights into something you can actually use. paper: repo: almost every idea in this post, from why memory bandwidth decides the outcome to why moving weights costs more than computing on them, comes straight out of how a GPU is built. I wrote a detailed primer on that. the article is quoted below.
0:42

UC Berkeley just open-sourced FreeToken. (2–4x faster local LLM inference than Ollama) the results are wild: - Qwen3.6-35B on an 8GB GPU at 39.3 tokens/s - DeepSeek-V4-Flash 284B on a 32GB GPU at 22 tokens/s - GLM-5.2 753B on a 96GB GPU at 14.9 tokens/s a 35B model at 16-bit precision needs about 70GB just for its weights. even at 4 bits it is close to 18GB, and FreeToken serves it on an 8GB GPU. let me explain how: all three models mentioned above are Mixture-of-Experts, and that is what FreeToken takes advantage of. each layer holds hundreds of separate experts plus a small router that picks a few of them per token. Qwen3.6-35B activates roughly 3B of its 35B parameters per token. DeepSeek-V4-Flash picks 6 of 256 experts per layer, so 13B of its 284B run at a time. so compute was never the bottleneck. the weights a single step touches fit comfortably on a consumer GPU. every expert the router might pick still has to exist somewhere. they sit in system RAM, and the GPU keeps a cache of the ones the model has been using recently. so everything comes down to what happens when the router picks an expert that is not on the GPU. there are two ways to serve that miss: 1. copy it over PCIe and run it on the GPU 2. run it on the CPU, where it already lives both read from the same system memory, so they compete for one pool of bandwidth instead of adding to each other. existing engines pick one option and freeze it when the model loads. but routing changes on every token, so a fixed choice misses most of what the model asks for. FreeToken measures both bandwidths on your machine and splits each step's misses between the two paths in proportion. the GPU and CPU results then merge exactly, with no approximation. two machines with the same GPU can end up wanting opposite strategies, which I did not expect. a 5090 in a gaming desktop should push nearly everything over PCIe, while an 8GB laptop is better off computing most misses on the CPU. none of that is readable off a spec sheet, so the engine profiles it once per machine. the second half of the design is about agents. coding agents constantly rewrite their own history, and every edit normally forces thousands of tokens back through prefill. FreeToken saves its checkpoints at the exact boundaries agent frameworks cut on, so it only reprocesses the new part. its slowest first token stays under 44 seconds, while llama.cpp peaks at 232 and KTransformers at 946. it serves the OpenAI and Anthropic APIs under Apache 2.0, so Claude Code and Codex can point at it directly. releasing weights publicly decides who can download a model, not who can afford to run one. frontier open models keep shipping, and running them still assumes a rented cluster. meanwhile there are over a hundred million consumer machines with discrete GPUs sitting mostly idle. closing that gap was never a hardware problem, and work like this is what turns open weights into something you can actually use. paper: repo: almost every idea in this post, from why memory bandwidth decides the outcome to why moving weights costs more than computing on them, comes straight out of how a GPU is built. I wrote a detailed primer on that. the article is quoted below.

Akshay 🚀

341,746 Aufrufe • vor 24 Tagen

$NVDA $MU $SNDK $LITE PAPER OVERVIEW AND CORE CLAIMS The paper “KV Cache Transform Coding for Compact Storage in LLM Inference” introduces kvtc, a transform-coding pipeline that compresses transformer key-value (KV) caches primarily for storage and transfer in LLM serving, rather than for accelerating the per-token attention kernel during active decoding. The method combines 3 stages: (1) feature decorrelation via a PCA basis computed from a calibration dataset and reused across requests; (2) adaptive, variable-precision quantization with bit allocation solved via dynamic programming (DP), including groupwise scaling/shift overhead; and (3) lossless entropy coding (DEFLATE via nvCOMP in the reference implementation) to exploit residual redundancy after quantization. The central empirical claim is that KV tensors contain large, exploitable redundancy across heads and layers, enabling approximately 20× compression versus a 16-bit baseline with negligible degradation across a broad set of accuracy and long-context benchmarks, with materially higher compression (≥40×) available at modest quality cost in some regimes. The system claim is that such compression materially improves the economics of multi-turn, prefix-reuse serving by extending effective KV cache capacity in GPU HBM and host tiers (DRAM/NVMe) and by reducing inter-node and GPU↔host bandwidth demands, thereby improving cache hit rates and reducing time-to-first-token (TTFT) relative to recomputation when caches would otherwise be evicted. KV CACHE AS THE DOMINANT STATE VARIABLE IN INFERENCE ECONOMICS KV cache growth is linear in context length and is multiplicative in layers and attention heads, making it an increasingly dominant constraint as (a) context lengths expand, (b) models add layers and maintain large hidden dimensions, and (c) production workloads shift toward iterative and tool-augmented interactions that repeatedly reuse long prefixes. The paper uses the canonical 16-bit KV cache size formula (4·l·h·d_head·t) bytes and reports 16-bit KV cache sizes per 1K tokens of context that are already operationally large: 128MiB for Llama 3.1 8B, 160MiB for Mistral NeMo 12B, and 320MiB for Llama 3.3 70B Instruct. In binary units, these figures imply per-token KV footprints of 128KiB/token (Llama 3.1 8B), 160KiB/token (Mistral NeMo 12B), and 320KiB/token (Llama 3.3 70B Instruct) at 16-bit. For a 10K-token prompt (10×1K in the paper’s binary convention), the 16-bit KV cache sizes scale to approximately 1.25GiB (Llama 3.1 8B), 1.56GiB (Mistral NeMo 12B), and 3.13GiB (Llama 3.3 70B Instruct). These magnitudes explain why stale caches create a throughput–latency dilemma: retaining them in HBM maximizes responsiveness on future turns but crowds out concurrent sessions; evicting them forces quadratic-cost prefill recomputation and increases TTFT; offloading them to host or storage introduces large transfer overhead and consumes DRAM/NVMe capacity. A key operational nuance emphasized is that modern serving stacks increasingly treat KV caches as a database, leveraging block paging and shared-prefix reuse. In the common disaggregated serving design (separate prefill and decode nodes), KV cache transfer becomes a dominant category of cross-node traffic. Under that design, any reduction in KV cache size directly increases effective fabric capacity and reduces tail latency attributable to congestion, while also enabling longer cache lifetimes in “hot” (HBM) and “warm” (CPU DRAM) tiers that raise cache hit rates and reduce recomputation frequency. The paper’s quantitative example illustrates the economic stakes: a 1,000-line code file tokenized at ~10 tokens/line yields ~10K tokens; for Llama 3.3 70B, an 8-bit KV cache for that context is ~1.6GiB. Reuse across subsequent turns or parallel chats around the same file is valuable, but HBM scarcity makes retaining many such caches infeasible without compression. TECHNICAL MECHANISM: WHY KV CACHES ARE COMPRESSIBLE AND HOW KVTC EXPLOITS IT The technical rationale begins with an empirical observation: keys (and, to a lesser extent, values) across different attention heads can be aligned into a shared latent space using orthogonal transformations (Procrustes alignment). This supports the hypothesis that head-specific projections introduce rotations of a common subspace rather than completely distinct information, implying that concatenating across heads and layers should reveal low-rank structure suitable for linear decorrelation and dimensionality reduction. The method operationalizes this using a PCA/SVD basis learned from calibration data rather than recomputing a decomposition per prompt. This design choice targets production viability: per-prompt SVD is computationally expensive and scales poorly with long prompts and frequent cache updates. kvtc is explicitly structured as an offline-calibrated, online-applied codec: Calibration (performed 1 time per model and compression setting for DP allocation) A calibration dataset is forwarded through the model to collect KV caches. Token positions are pooled, and a subset of positions is sampled. Keys and values are processed separately. Several implementation choices are highlighted as decisive for stability: Rotary positional embeddings are effectively removed prior to compression (“undo positional rotations”), because positional rotations degrade the apparent low-rank structure of keys. “Attention sink” tokens (the earliest tokens in the sequence) and a sliding window of most recent tokens are excluded from compression because they disproportionately affect attention patterns and are empirically more sensitive to reconstruction error. Cross-layer concatenation is used: keys (or values) from multiple layers and heads at the same token position are concatenated along the feature axis to form a higher-dimensional feature vector. PCA is computed over these concatenated vectors, improving robustness relative to per-layer or per-head PCA. The PCA basis is computed via SVD of centered calibration data, using randomized SVD for scalability with a target rank cutoff. The paper reports calibration regimes of 160K tokens for several models with a 10K PCA dimension cutoff (8K for Qwen variants with fewer KV heads), selected to fit within a single 80GB H100 memory envelope and complete within minutes. A critical economic detail is that the same PCA basis can be reused across multiple compression ratios; only the DP-derived precision assignment changes per compression target. Compression (applied between inference phases) Compression operates on stored KV cache tensors, not on weights, and does not modify attention computation. The KV cache is projected into the PCA basis, quantized, packed, and then entropy-coded. Compression is positioned as a background or between-phase operation (after decoding, or between prefill and decode), executed on GPU or CPU depending on where the cache currently resides. The design intent is that compression should not sit on the critical per-token decoding path; it is a storage and transport optimization. Decompression (performed prior to reuse) Decompression reverses the entropy coding and quantization and applies the inverse PCA projection. A practical latency optimization is proposed: inverse projection can be performed layer-by-layer using submatrices of the PCA basis, allowing generation to begin before the full cache is reconstructed, reducing TTFT. Quantization and bit allocation are the core differentiators versus simpler PCA truncation. PCA provides ordered components by variance; kvtc uses DP to allocate a global bit budget across PCA coordinates (and across groups of coordinates) to minimize reconstruction error in the decorrelated domain. Groups of subsequent PCA coordinates share 16-bit shift and scale factors (a microscaling-inspired design), and the DP algorithm jointly selects group size and precision type under a bit budget, including the overhead of per-group metadata. DP commonly assigns 0 bits to many trailing PCA components, which both increases compression and provides a mechanism to trim the PCA basis to the subset of components that actually carry payload, reducing compute and storage overhead of the projection matrices in deployment. Lossless entropy coding then exploits the structure induced by quantization. DEFLATE is used in the reference implementation, and the paper emphasizes that the incremental gain from the lossless stage is content-dependent but meaningful, with an average uplift of ~1.23× on top of quantization in the reported regime. An ablation in the appendices indicates that GPU-friendly variants (GDeflate) can achieve nearly identical compression ratios (≤0.1 difference in measured cases), implying that throughput-optimized lossless codecs can likely be substituted without sacrificing meaningful compression. EMPIRICAL RESULTS: ACCURACY, COMPRESSION, AND LATENCY General-purpose 8B–12B dense models The paper evaluates Llama 3.1 8B, MN-Minitron 8B, and Mistral NeMo 12B across math/knowledge (GSM8K, MMLU) and long-context tasks (Qasper, Lost in the Middle, RULER Variable Tracking) under a simulated multi-turn regime where compression/decompression is applied periodically, with a sliding window of recent tokens excluded. A consistent pattern appears: kvtc maintains near-vanilla performance through 16× compression settings, and remains competitive at 32×, with degradation becoming task- and model-dependent at 64×, particularly on long-context retrieval metrics when compression is pushed aggressively. Selected quantitative anchor points from the paper’s standard-error table (all values are reported with the paper’s evaluation setup and token-window exclusions): Llama 3.1 8B Vanilla: GSM8K 56.8, MMLU 60.5, Qasper 40.4, LITM 99.4, RULER-VT 99.8 kvtc16×: GSM8K 56.9, MMLU 60.1, Qasper 40.7, LITM 99.3, RULER-VT 99.1 kvtc32×: GSM8K 57.8, MMLU 60.6, Qasper 39.4, LITM 99.1, RULER-VT 98.9 kvtc64×: GSM8K 57.2, MMLU 60.7, Qasper 37.8, LITM 90.2, RULER-VT 95.9 These results indicate that, for this model, long-context sensitivity emerges at 64× with meaningful drops in LITM and RULER-VT, while math/knowledge scores remain stable, implying a differential sensitivity consistent with key-vector precision being more critical for retrieval-style behavior. Mistral NeMo 12B Vanilla: GSM8K 61.9, MMLU 64.5, Qasper 38.4, LITM 99.5, RULER-VT 99.8 kvtc16×: GSM8K 62.0, MMLU 64.4, Qasper 37.6, LITM 99.8, RULER-VT 99.5 kvtc32×: GSM8K 62.2, MMLU 63.8, Qasper 37.5, LITM 99.6, RULER-VT 98.7 kvtc64×: GSM8K 61.9, MMLU 61.4, Qasper 38.0, LITM 95.3, RULER-VT 98.0 Here, degradation at 64× is visible but materially smaller than the Llama 3.1 8B LITM drop, suggesting model-architecture or training-data differences can change the tolerance envelope for aggressive KV cache distortion. MN-Minitron 8B Vanilla: GSM8K 59.1, MMLU 64.3, Qasper 38.2, LITM 99.8, RULER-VT 99.4 kvtc16×: GSM8K 60.3, MMLU 64.1, Qasper 38.6, LITM 99.3, RULER-VT 98.8 kvtc32×: GSM8K 59.1, MMLU 63.7, Qasper 37.7, LITM 86.9, RULER-VT 96.0 kvtc64×: GSM8K 57.8, MMLU 62.1, Qasper 38.1, LITM 59.5, RULER-VT 93.4 This model shows markedly higher sensitivity on LITM at 32× and 64×, despite stable short-context metrics, reinforcing that “compression safety” is not monotonic in parameter count and that pruning/distillation choices can alter KV cache redundancy or robustness. Comparisons to baselines The paper compares kvtc to quantization baselines (KIVI, GEAR, FP8) and eviction baselines (H2O, TOVA), plus an SVD-based prefill-optimization method (xKV). Across the reported tasks: Low-bit quantization methods at modest compression (2-bit KV schemes) show earlier degradation in long-context behavior than kvtc at substantially higher compression settings. Eviction methods perform poorly as generic compressors for long-context tasks, consistent with their objective function (selective pruning) being misaligned with “lossless-ish storage for reuse.” xKV shows competitive results on some tasks but a consistent underperformance on Qasper relative to kvtc and vanilla in the provided tables, consistent with method-specific distortions introduced by its decomposition regime. Reasoning models and high-variance tasks For DeepSeek-R1-distilled Qwen 2.5 reasoning models, the paper evaluates AIME 2024/2025 and LiveCodeBench coding. Results are averaged over 8 runs with large variance, but a key inference is that kvtc at ~9×–21× compression achieves broadly similar AIME scores within variance bands, while coding performance remains stable at ~9× and degrades more visibly at ~18×–21× on the 7B model. An important nuance is that smaller reasoning models already have smaller KV footprints (reported ~29KiB/token for Qwen R1 1.5B versus 131KiB/token for Llama 3.1 8B), so the economic value of aggressive KV cache compression is proportionally higher for large models and long contexts than for small models with short contexts, unless the serving system’s bottleneck is dominated by cache transfer rather than HBM capacity. Multi-GPU inference and pipeline parallel For Llama 3.3 70B Instruct run pipeline-parallel across 4 GPUs (20 layers per GPU), the paper compresses KV cache chunks independently per GPU. On MATH-500, the reported accuracy declines from 75.6 (vanilla) to 74.4 at 10× and 72.6 at 20×, with standard errors near ~1.9. NIAH and LITM remain at 100.0 for all tested ratios in that table. The paper notes that joint compression across chunks could improve accuracy for some offload scenarios but is not required for feasibility, highlighting an engineering trade-off between deployment simplicity in distributed settings and optimal global compression. Latency and TTFT economics A critical system result is the measured compression/decompression latency on an H100 for a non-fused implementation. For Mistral NeMo 12B in bfloat16: BS=8, CTX=8K: compression 379ms, decompression 267ms; vanilla recompute TTFT 3098ms; kvtc decompression TTFT 380ms BS=2, CTX=16K: compression 194ms, decompression 143ms; vanilla recompute TTFT 1780ms; kvtc decompression TTFT 208ms These measurements imply that, when a cache would otherwise be recomputed, decompressing a stored compressed cache can reduce TTFT by ~8×–9× in these scenarios, even without kernel fusion. The decomposition of runtime shows PCA projection and entropy coding as the largest contributors, implying that GPU-optimized kernels and faster GPU-native lossless codecs could reduce overhead further. The fundamental economic conclusion is that, in multi-turn settings with long prefixes, compression-induced overhead is likely dominated by the avoided prefill compute and avoided transfer overhead for uncompressed caches. KEY DEPLOYMENT-SENSITIVE DESIGN CHOICES AND FAILURE MODES Several design choices appear to be “hard requirements” rather than optional optimizations: Sink tokens and sliding window exclusions The paper’s ablations show that compressing early “sink” tokens can catastrophically degrade accuracy at high compression ratios (example: Llama 3.1 8B at 64× collapses on multiple tasks when sink tokens are compressed). Similarly, compressing the most recent tokens hurts performance, motivating a sliding window (default 128 tokens) that remains uncompressed. This introduces a predictable engineering constraint: kvtc is not a uniform compression of the full cache; it is a policy-driven, token-position-dependent codec. Production integration therefore requires correct handling of token positions, attention sinks, and window management, and these policies must be aligned with attention-kernel behavior and model-specific sink dynamics. RoPE handling Removing positional rotations prior to compression is described as important for preserving low-rank structure. In deployment, this implies that the codec must be position-aware and must invert and reapply RoPE correctly. This is an additional source of complexity relative to pure per-token quantization and is sensitive to model variants and RoPE parameterizations. Calibration set representativeness The method’s quality hinges on the PCA basis generalizing from calibration data to production data. The paper demonstrates relative stability with 160K–200K calibration tokens and explores domain shifts (general web text vs math traces vs code). Results suggest that moderate domain mismatch is tolerated at 16×–64×, while extreme compression (e.g., 256× in ablations) becomes materially more sensitive to calibration choice. In production, this implies that operators targeting the “negligible degradation” regime should be able to calibrate with broadly representative corpora, while operators targeting ultra-high compression for specialized workloads should expect tighter coupling between calibration domain and achieved quality. PCA matrix storage overhead and operational footprint A non-trivial hidden cost is the need to store PCA projection matrices per model. The paper reports that, prior to DP trimming, PCA matrices stored at 16-bit can amount to a meaningful fraction of model parameter count (examples reported: ~2.4% for Llama 3.3 70B, ~8.7% for Llama 3.1 8B). This overhead is amortized across all cached sessions for a model but competes with HBM/DRAM budgets in multi-model serving. DP-driven trimming can reduce this overhead at higher compression ratios by removing zero-bit components, but the directionality is not guaranteed at low compression ratios if many components remain active. In distributed inference (pipeline parallel), per-chunk PCA can reduce matrix sizes, but may reduce cross-layer decorrelation benefits if fewer layers are concatenated. SYSTEM-LEVEL IMPLICATIONS FOR GENERATIVE AI INFRASTRUCTURE GPU AND HBM The principal infrastructure implication is that KV cache compression at storage time targets the dominant memory allocator stressor in stateful serving: the accumulation of idle or warm conversation state. For workloads with long reusable prefixes (code assistants, enterprise agents with large system prompts, repeated RAG scaffolds, document chat), the limiting resource frequently becomes HBM reserved for KV caches rather than compute. By compressing stale caches by ~20× (or more), the same HBM budget can retain a materially larger working set of cached prefixes, increasing cache hit rates and reducing recomputation. This effect is multiplicative with cache-aware routing and prefix sharing: more prefixes can remain resident (hot or warm) and can be routed to nodes that already hold them, improving both throughput and tail latency. However, kvtc as described does not reduce the active KV cache footprint during the actual attention computation for a currently decoding sequence, because the model operates on decompressed KV caches during decoding. Therefore, the method does not directly reduce HBM bandwidth consumed by attention kernels during steady-state decode, and does not directly address the “memory traffic per generated token” bottleneck that motivates online KV quantization and eviction strategies. The primary HBM benefit is increased effective capacity for caches between turns and reduced HBM pressure from storing many idle sessions, not reduced per-token decode bandwidth. Compression and decompression themselves consume GPU compute and memory bandwidth. The measured decompression TTFT of ~208ms–380ms in the provided benchmarks indicates that the overhead is real but can be materially smaller than recomputation of long prefixes. In an HBM-constrained serving environment, this overhead can be interpreted as a trade between (a) maintaining more caches warm and paying decompression on reuse versus (b) evicting caches and paying full prefill recomputation. The decision boundary will depend on distribution of inter-turn idle times, probability of reuse, and SLA sensitivity to TTFT. kvtc expands the feasible region where keeping caches is economically rational, especially for long prompts. CPU AND DRAM The method implies a stronger role for CPU DRAM as a warm KV cache tier. A ~20× compression ratio changes the practical scale of “warm state” that can be stored per server. Using the paper’s reported KV cache sizes, a 10K-token 16-bit KV cache for Llama 3.3 70B is ~3.13GiB; compressing by ~20× would reduce this to ~160MiB. At that size, storing hundreds to thousands of warm conversation states in DRAM becomes materially more feasible, increasing cache hit rates and reducing NVMe dependence. This can shift system design from “HBM-only hot caches with aggressive eviction” toward “HBM hot + DRAM warm with long retention,” which is structurally analogous to CPU page cache hierarchies in classical systems design. CPU compute implications depend on where compression is executed. The paper explicitly allows compression on CPU if the cache is already in storage, but the strongest bandwidth savings are achieved when compression happens before moving KV caches off the GPU. If an operator chooses GPU-side compression prior to PCIe/NVLink transfer, CPU compute overhead is modest (orchestrating and DP calibration offline). If an operator instead transfers uncompressed caches to CPU for compression, bandwidth savings are forfeited and CPU memory bandwidth becomes a bottleneck. Therefore, the most economically coherent deployment path is GPU-native compression/decompression with CPU DRAM used as the warm storage reservoir.
12:09

$NVDA $MU $SNDK $LITE PAPER OVERVIEW AND CORE CLAIMS The paper “KV Cache Transform Coding for Compact Storage in LLM Inference” introduces kvtc, a transform-coding pipeline that compresses transformer key-value (KV) caches primarily for storage and transfer in LLM serving, rather than for accelerating the per-token attention kernel during active decoding. The method combines 3 stages: (1) feature decorrelation via a PCA basis computed from a calibration dataset and reused across requests; (2) adaptive, variable-precision quantization with bit allocation solved via dynamic programming (DP), including groupwise scaling/shift overhead; and (3) lossless entropy coding (DEFLATE via nvCOMP in the reference implementation) to exploit residual redundancy after quantization. The central empirical claim is that KV tensors contain large, exploitable redundancy across heads and layers, enabling approximately 20× compression versus a 16-bit baseline with negligible degradation across a broad set of accuracy and long-context benchmarks, with materially higher compression (≥40×) available at modest quality cost in some regimes. The system claim is that such compression materially improves the economics of multi-turn, prefix-reuse serving by extending effective KV cache capacity in GPU HBM and host tiers (DRAM/NVMe) and by reducing inter-node and GPU↔host bandwidth demands, thereby improving cache hit rates and reducing time-to-first-token (TTFT) relative to recomputation when caches would otherwise be evicted. KV CACHE AS THE DOMINANT STATE VARIABLE IN INFERENCE ECONOMICS KV cache growth is linear in context length and is multiplicative in layers and attention heads, making it an increasingly dominant constraint as (a) context lengths expand, (b) models add layers and maintain large hidden dimensions, and (c) production workloads shift toward iterative and tool-augmented interactions that repeatedly reuse long prefixes. The paper uses the canonical 16-bit KV cache size formula (4·l·h·d_head·t) bytes and reports 16-bit KV cache sizes per 1K tokens of context that are already operationally large: 128MiB for Llama 3.1 8B, 160MiB for Mistral NeMo 12B, and 320MiB for Llama 3.3 70B Instruct. In binary units, these figures imply per-token KV footprints of 128KiB/token (Llama 3.1 8B), 160KiB/token (Mistral NeMo 12B), and 320KiB/token (Llama 3.3 70B Instruct) at 16-bit. For a 10K-token prompt (10×1K in the paper’s binary convention), the 16-bit KV cache sizes scale to approximately 1.25GiB (Llama 3.1 8B), 1.56GiB (Mistral NeMo 12B), and 3.13GiB (Llama 3.3 70B Instruct). These magnitudes explain why stale caches create a throughput–latency dilemma: retaining them in HBM maximizes responsiveness on future turns but crowds out concurrent sessions; evicting them forces quadratic-cost prefill recomputation and increases TTFT; offloading them to host or storage introduces large transfer overhead and consumes DRAM/NVMe capacity. A key operational nuance emphasized is that modern serving stacks increasingly treat KV caches as a database, leveraging block paging and shared-prefix reuse. In the common disaggregated serving design (separate prefill and decode nodes), KV cache transfer becomes a dominant category of cross-node traffic. Under that design, any reduction in KV cache size directly increases effective fabric capacity and reduces tail latency attributable to congestion, while also enabling longer cache lifetimes in “hot” (HBM) and “warm” (CPU DRAM) tiers that raise cache hit rates and reduce recomputation frequency. The paper’s quantitative example illustrates the economic stakes: a 1,000-line code file tokenized at ~10 tokens/line yields ~10K tokens; for Llama 3.3 70B, an 8-bit KV cache for that context is ~1.6GiB. Reuse across subsequent turns or parallel chats around the same file is valuable, but HBM scarcity makes retaining many such caches infeasible without compression. TECHNICAL MECHANISM: WHY KV CACHES ARE COMPRESSIBLE AND HOW KVTC EXPLOITS IT The technical rationale begins with an empirical observation: keys (and, to a lesser extent, values) across different attention heads can be aligned into a shared latent space using orthogonal transformations (Procrustes alignment). This supports the hypothesis that head-specific projections introduce rotations of a common subspace rather than completely distinct information, implying that concatenating across heads and layers should reveal low-rank structure suitable for linear decorrelation and dimensionality reduction. The method operationalizes this using a PCA/SVD basis learned from calibration data rather than recomputing a decomposition per prompt. This design choice targets production viability: per-prompt SVD is computationally expensive and scales poorly with long prompts and frequent cache updates. kvtc is explicitly structured as an offline-calibrated, online-applied codec: Calibration (performed 1 time per model and compression setting for DP allocation) A calibration dataset is forwarded through the model to collect KV caches. Token positions are pooled, and a subset of positions is sampled. Keys and values are processed separately. Several implementation choices are highlighted as decisive for stability: Rotary positional embeddings are effectively removed prior to compression (“undo positional rotations”), because positional rotations degrade the apparent low-rank structure of keys. “Attention sink” tokens (the earliest tokens in the sequence) and a sliding window of most recent tokens are excluded from compression because they disproportionately affect attention patterns and are empirically more sensitive to reconstruction error. Cross-layer concatenation is used: keys (or values) from multiple layers and heads at the same token position are concatenated along the feature axis to form a higher-dimensional feature vector. PCA is computed over these concatenated vectors, improving robustness relative to per-layer or per-head PCA. The PCA basis is computed via SVD of centered calibration data, using randomized SVD for scalability with a target rank cutoff. The paper reports calibration regimes of 160K tokens for several models with a 10K PCA dimension cutoff (8K for Qwen variants with fewer KV heads), selected to fit within a single 80GB H100 memory envelope and complete within minutes. A critical economic detail is that the same PCA basis can be reused across multiple compression ratios; only the DP-derived precision assignment changes per compression target. Compression (applied between inference phases) Compression operates on stored KV cache tensors, not on weights, and does not modify attention computation. The KV cache is projected into the PCA basis, quantized, packed, and then entropy-coded. Compression is positioned as a background or between-phase operation (after decoding, or between prefill and decode), executed on GPU or CPU depending on where the cache currently resides. The design intent is that compression should not sit on the critical per-token decoding path; it is a storage and transport optimization. Decompression (performed prior to reuse) Decompression reverses the entropy coding and quantization and applies the inverse PCA projection. A practical latency optimization is proposed: inverse projection can be performed layer-by-layer using submatrices of the PCA basis, allowing generation to begin before the full cache is reconstructed, reducing TTFT. Quantization and bit allocation are the core differentiators versus simpler PCA truncation. PCA provides ordered components by variance; kvtc uses DP to allocate a global bit budget across PCA coordinates (and across groups of coordinates) to minimize reconstruction error in the decorrelated domain. Groups of subsequent PCA coordinates share 16-bit shift and scale factors (a microscaling-inspired design), and the DP algorithm jointly selects group size and precision type under a bit budget, including the overhead of per-group metadata. DP commonly assigns 0 bits to many trailing PCA components, which both increases compression and provides a mechanism to trim the PCA basis to the subset of components that actually carry payload, reducing compute and storage overhead of the projection matrices in deployment. Lossless entropy coding then exploits the structure induced by quantization. DEFLATE is used in the reference implementation, and the paper emphasizes that the incremental gain from the lossless stage is content-dependent but meaningful, with an average uplift of ~1.23× on top of quantization in the reported regime. An ablation in the appendices indicates that GPU-friendly variants (GDeflate) can achieve nearly identical compression ratios (≤0.1 difference in measured cases), implying that throughput-optimized lossless codecs can likely be substituted without sacrificing meaningful compression. EMPIRICAL RESULTS: ACCURACY, COMPRESSION, AND LATENCY General-purpose 8B–12B dense models The paper evaluates Llama 3.1 8B, MN-Minitron 8B, and Mistral NeMo 12B across math/knowledge (GSM8K, MMLU) and long-context tasks (Qasper, Lost in the Middle, RULER Variable Tracking) under a simulated multi-turn regime where compression/decompression is applied periodically, with a sliding window of recent tokens excluded. A consistent pattern appears: kvtc maintains near-vanilla performance through 16× compression settings, and remains competitive at 32×, with degradation becoming task- and model-dependent at 64×, particularly on long-context retrieval metrics when compression is pushed aggressively. Selected quantitative anchor points from the paper’s standard-error table (all values are reported with the paper’s evaluation setup and token-window exclusions): Llama 3.1 8B Vanilla: GSM8K 56.8, MMLU 60.5, Qasper 40.4, LITM 99.4, RULER-VT 99.8 kvtc16×: GSM8K 56.9, MMLU 60.1, Qasper 40.7, LITM 99.3, RULER-VT 99.1 kvtc32×: GSM8K 57.8, MMLU 60.6, Qasper 39.4, LITM 99.1, RULER-VT 98.9 kvtc64×: GSM8K 57.2, MMLU 60.7, Qasper 37.8, LITM 90.2, RULER-VT 95.9 These results indicate that, for this model, long-context sensitivity emerges at 64× with meaningful drops in LITM and RULER-VT, while math/knowledge scores remain stable, implying a differential sensitivity consistent with key-vector precision being more critical for retrieval-style behavior. Mistral NeMo 12B Vanilla: GSM8K 61.9, MMLU 64.5, Qasper 38.4, LITM 99.5, RULER-VT 99.8 kvtc16×: GSM8K 62.0, MMLU 64.4, Qasper 37.6, LITM 99.8, RULER-VT 99.5 kvtc32×: GSM8K 62.2, MMLU 63.8, Qasper 37.5, LITM 99.6, RULER-VT 98.7 kvtc64×: GSM8K 61.9, MMLU 61.4, Qasper 38.0, LITM 95.3, RULER-VT 98.0 Here, degradation at 64× is visible but materially smaller than the Llama 3.1 8B LITM drop, suggesting model-architecture or training-data differences can change the tolerance envelope for aggressive KV cache distortion. MN-Minitron 8B Vanilla: GSM8K 59.1, MMLU 64.3, Qasper 38.2, LITM 99.8, RULER-VT 99.4 kvtc16×: GSM8K 60.3, MMLU 64.1, Qasper 38.6, LITM 99.3, RULER-VT 98.8 kvtc32×: GSM8K 59.1, MMLU 63.7, Qasper 37.7, LITM 86.9, RULER-VT 96.0 kvtc64×: GSM8K 57.8, MMLU 62.1, Qasper 38.1, LITM 59.5, RULER-VT 93.4 This model shows markedly higher sensitivity on LITM at 32× and 64×, despite stable short-context metrics, reinforcing that “compression safety” is not monotonic in parameter count and that pruning/distillation choices can alter KV cache redundancy or robustness. Comparisons to baselines The paper compares kvtc to quantization baselines (KIVI, GEAR, FP8) and eviction baselines (H2O, TOVA), plus an SVD-based prefill-optimization method (xKV). Across the reported tasks: Low-bit quantization methods at modest compression (2-bit KV schemes) show earlier degradation in long-context behavior than kvtc at substantially higher compression settings. Eviction methods perform poorly as generic compressors for long-context tasks, consistent with their objective function (selective pruning) being misaligned with “lossless-ish storage for reuse.” xKV shows competitive results on some tasks but a consistent underperformance on Qasper relative to kvtc and vanilla in the provided tables, consistent with method-specific distortions introduced by its decomposition regime. Reasoning models and high-variance tasks For DeepSeek-R1-distilled Qwen 2.5 reasoning models, the paper evaluates AIME 2024/2025 and LiveCodeBench coding. Results are averaged over 8 runs with large variance, but a key inference is that kvtc at ~9×–21× compression achieves broadly similar AIME scores within variance bands, while coding performance remains stable at ~9× and degrades more visibly at ~18×–21× on the 7B model. An important nuance is that smaller reasoning models already have smaller KV footprints (reported ~29KiB/token for Qwen R1 1.5B versus 131KiB/token for Llama 3.1 8B), so the economic value of aggressive KV cache compression is proportionally higher for large models and long contexts than for small models with short contexts, unless the serving system’s bottleneck is dominated by cache transfer rather than HBM capacity. Multi-GPU inference and pipeline parallel For Llama 3.3 70B Instruct run pipeline-parallel across 4 GPUs (20 layers per GPU), the paper compresses KV cache chunks independently per GPU. On MATH-500, the reported accuracy declines from 75.6 (vanilla) to 74.4 at 10× and 72.6 at 20×, with standard errors near ~1.9. NIAH and LITM remain at 100.0 for all tested ratios in that table. The paper notes that joint compression across chunks could improve accuracy for some offload scenarios but is not required for feasibility, highlighting an engineering trade-off between deployment simplicity in distributed settings and optimal global compression. Latency and TTFT economics A critical system result is the measured compression/decompression latency on an H100 for a non-fused implementation. For Mistral NeMo 12B in bfloat16: BS=8, CTX=8K: compression 379ms, decompression 267ms; vanilla recompute TTFT 3098ms; kvtc decompression TTFT 380ms BS=2, CTX=16K: compression 194ms, decompression 143ms; vanilla recompute TTFT 1780ms; kvtc decompression TTFT 208ms These measurements imply that, when a cache would otherwise be recomputed, decompressing a stored compressed cache can reduce TTFT by ~8×–9× in these scenarios, even without kernel fusion. The decomposition of runtime shows PCA projection and entropy coding as the largest contributors, implying that GPU-optimized kernels and faster GPU-native lossless codecs could reduce overhead further. The fundamental economic conclusion is that, in multi-turn settings with long prefixes, compression-induced overhead is likely dominated by the avoided prefill compute and avoided transfer overhead for uncompressed caches. KEY DEPLOYMENT-SENSITIVE DESIGN CHOICES AND FAILURE MODES Several design choices appear to be “hard requirements” rather than optional optimizations: Sink tokens and sliding window exclusions The paper’s ablations show that compressing early “sink” tokens can catastrophically degrade accuracy at high compression ratios (example: Llama 3.1 8B at 64× collapses on multiple tasks when sink tokens are compressed). Similarly, compressing the most recent tokens hurts performance, motivating a sliding window (default 128 tokens) that remains uncompressed. This introduces a predictable engineering constraint: kvtc is not a uniform compression of the full cache; it is a policy-driven, token-position-dependent codec. Production integration therefore requires correct handling of token positions, attention sinks, and window management, and these policies must be aligned with attention-kernel behavior and model-specific sink dynamics. RoPE handling Removing positional rotations prior to compression is described as important for preserving low-rank structure. In deployment, this implies that the codec must be position-aware and must invert and reapply RoPE correctly. This is an additional source of complexity relative to pure per-token quantization and is sensitive to model variants and RoPE parameterizations. Calibration set representativeness The method’s quality hinges on the PCA basis generalizing from calibration data to production data. The paper demonstrates relative stability with 160K–200K calibration tokens and explores domain shifts (general web text vs math traces vs code). Results suggest that moderate domain mismatch is tolerated at 16×–64×, while extreme compression (e.g., 256× in ablations) becomes materially more sensitive to calibration choice. In production, this implies that operators targeting the “negligible degradation” regime should be able to calibrate with broadly representative corpora, while operators targeting ultra-high compression for specialized workloads should expect tighter coupling between calibration domain and achieved quality. PCA matrix storage overhead and operational footprint A non-trivial hidden cost is the need to store PCA projection matrices per model. The paper reports that, prior to DP trimming, PCA matrices stored at 16-bit can amount to a meaningful fraction of model parameter count (examples reported: ~2.4% for Llama 3.3 70B, ~8.7% for Llama 3.1 8B). This overhead is amortized across all cached sessions for a model but competes with HBM/DRAM budgets in multi-model serving. DP-driven trimming can reduce this overhead at higher compression ratios by removing zero-bit components, but the directionality is not guaranteed at low compression ratios if many components remain active. In distributed inference (pipeline parallel), per-chunk PCA can reduce matrix sizes, but may reduce cross-layer decorrelation benefits if fewer layers are concatenated. SYSTEM-LEVEL IMPLICATIONS FOR GENERATIVE AI INFRASTRUCTURE GPU AND HBM The principal infrastructure implication is that KV cache compression at storage time targets the dominant memory allocator stressor in stateful serving: the accumulation of idle or warm conversation state. For workloads with long reusable prefixes (code assistants, enterprise agents with large system prompts, repeated RAG scaffolds, document chat), the limiting resource frequently becomes HBM reserved for KV caches rather than compute. By compressing stale caches by ~20× (or more), the same HBM budget can retain a materially larger working set of cached prefixes, increasing cache hit rates and reducing recomputation. This effect is multiplicative with cache-aware routing and prefix sharing: more prefixes can remain resident (hot or warm) and can be routed to nodes that already hold them, improving both throughput and tail latency. However, kvtc as described does not reduce the active KV cache footprint during the actual attention computation for a currently decoding sequence, because the model operates on decompressed KV caches during decoding. Therefore, the method does not directly reduce HBM bandwidth consumed by attention kernels during steady-state decode, and does not directly address the “memory traffic per generated token” bottleneck that motivates online KV quantization and eviction strategies. The primary HBM benefit is increased effective capacity for caches between turns and reduced HBM pressure from storing many idle sessions, not reduced per-token decode bandwidth. Compression and decompression themselves consume GPU compute and memory bandwidth. The measured decompression TTFT of ~208ms–380ms in the provided benchmarks indicates that the overhead is real but can be materially smaller than recomputation of long prefixes. In an HBM-constrained serving environment, this overhead can be interpreted as a trade between (a) maintaining more caches warm and paying decompression on reuse versus (b) evicting caches and paying full prefill recomputation. The decision boundary will depend on distribution of inter-turn idle times, probability of reuse, and SLA sensitivity to TTFT. kvtc expands the feasible region where keeping caches is economically rational, especially for long prompts. CPU AND DRAM The method implies a stronger role for CPU DRAM as a warm KV cache tier. A ~20× compression ratio changes the practical scale of “warm state” that can be stored per server. Using the paper’s reported KV cache sizes, a 10K-token 16-bit KV cache for Llama 3.3 70B is ~3.13GiB; compressing by ~20× would reduce this to ~160MiB. At that size, storing hundreds to thousands of warm conversation states in DRAM becomes materially more feasible, increasing cache hit rates and reducing NVMe dependence. This can shift system design from “HBM-only hot caches with aggressive eviction” toward “HBM hot + DRAM warm with long retention,” which is structurally analogous to CPU page cache hierarchies in classical systems design. CPU compute implications depend on where compression is executed. The paper explicitly allows compression on CPU if the cache is already in storage, but the strongest bandwidth savings are achieved when compression happens before moving KV caches off the GPU. If an operator chooses GPU-side compression prior to PCIe/NVLink transfer, CPU compute overhead is modest (orchestrating and DP calibration offline). If an operator instead transfers uncompressed caches to CPU for compression, bandwidth savings are forfeited and CPU memory bandwidth becomes a bottleneck. Therefore, the most economically coherent deployment path is GPU-native compression/decompression with CPU DRAM used as the warm storage reservoir.

TheValueist

16,549 Aufrufe • vor 7 Monaten

YES, FUCKING YES! I DID IT !!! In my last post, I heavily trashed Qwen3.8 Flash Next, claiming it couldn't do anything at all. I managed to get awesome speed out of it, but in nearly 3 hours it couldn't even generate a simple Pagoda Garden. I had no clue what was going on. Turns out the issue was with my custom quant, which I built myself to fit 256k context + vision into 4x3090 with good throughput. Ton Cao released a properly calibrated AWQ INT4 of this model, but it was larger than mine and didn't fit into my 96GB as-is. CUDA OOM - sounds familiar to everyone, right? I decided to tackle it and make something that actually works. A full day of work, interrupted only by 5 hours of sleep. To fit 262K context, I had to un-fuse the fused layers and quantize all non-sensitive projections to FP8. vLLM allows only 1 quantization scheme per fused layer, which turned into writing a custom loader and 13 launch attempts 😭 Attempt 12 finally loaded, and I was so hyped. But the model's output was pure gibberish, total mush. Turns out TP4 was slicing the fused QKV matrix in the wrong places, for fuck's sake. Attempt 13 fixed the slicing. 262K fit. The model kept spitting garbage on any complex task. Short answers were fine, a 40K needle test was fine, but ask for a three.js scene - and you get 16,384 tokens of complete shit. I was devastated. The root cause of all this was just mind-bogglingly fucked up. 48 tiny tensors, shared_expert_gate.weight, ~5 KB each. My build script quantized them to FP8, but the runtime read them as BF16 without applying the scale 🤡 The gate expects values around 0.077, but was receiving 448 🤡🤡 Eat up, my dear friend. And this gate determines how much the shared expert contributes to every single token. I restored all 48 from the original checkpoint - and the pagoda scene, which I had spent hours grinding on and ran dozens of times - just passed twice in a row. Wanna know how mindblown I was? Exactly. Then the model swallowed 258,008 real prompt tokens with an image and nailed the needle test. Decode at 63-68 tok/s at 220 W per GPU - identical to what the broken build pushed. Video is the proof. One prompt, two sets of weights: reference BF16 vs. my Franken-quant running on 4 gaming GPUs from 2020. One-shot, zero fixes. Mine finished first, and the scene came out tighter. It's just 1 prompt, so I won't call it a benchmark. 24 hours of my life for 240 KB of broken tensors 🤡🤡🤡
1:01

YES, FUCKING YES! I DID IT !!! In my last post, I heavily trashed Qwen3.8 Flash Next, claiming it couldn't do anything at all. I managed to get awesome speed out of it, but in nearly 3 hours it couldn't even generate a simple Pagoda Garden. I had no clue what was going on. Turns out the issue was with my custom quant, which I built myself to fit 256k context + vision into 4x3090 with good throughput. Ton Cao released a properly calibrated AWQ INT4 of this model, but it was larger than mine and didn't fit into my 96GB as-is. CUDA OOM - sounds familiar to everyone, right? I decided to tackle it and make something that actually works. A full day of work, interrupted only by 5 hours of sleep. To fit 262K context, I had to un-fuse the fused layers and quantize all non-sensitive projections to FP8. vLLM allows only 1 quantization scheme per fused layer, which turned into writing a custom loader and 13 launch attempts 😭 Attempt 12 finally loaded, and I was so hyped. But the model's output was pure gibberish, total mush. Turns out TP4 was slicing the fused QKV matrix in the wrong places, for fuck's sake. Attempt 13 fixed the slicing. 262K fit. The model kept spitting garbage on any complex task. Short answers were fine, a 40K needle test was fine, but ask for a three.js scene - and you get 16,384 tokens of complete shit. I was devastated. The root cause of all this was just mind-bogglingly fucked up. 48 tiny tensors, shared_expert_gate.weight, ~5 KB each. My build script quantized them to FP8, but the runtime read them as BF16 without applying the scale 🤡 The gate expects values around 0.077, but was receiving 448 🤡🤡 Eat up, my dear friend. And this gate determines how much the shared expert contributes to every single token. I restored all 48 from the original checkpoint - and the pagoda scene, which I had spent hours grinding on and ran dozens of times - just passed twice in a row. Wanna know how mindblown I was? Exactly. Then the model swallowed 258,008 real prompt tokens with an image and nailed the needle test. Decode at 63-68 tok/s at 220 W per GPU - identical to what the broken build pushed. Video is the proof. One prompt, two sets of weights: reference BF16 vs. my Franken-quant running on 4 gaming GPUs from 2020. One-shot, zero fixes. Mine finished first, and the scene came out tighter. It's just 1 prompt, so I won't call it a benchmark. 24 hours of my life for 240 KB of broken tensors 🤡🤡🤡

Alexey Fateev

98,942 Aufrufe • vor 17 Tagen

CANCEL Your Weekend Plans, and Learn Claude Code Today. $5,000/month. $10,000/month. $20,000/month. People are building entire apps and charging clients thousands using Claude Code. You're still Googling 'how to center a div.' While you're binge-watching a show you won't remember next week, a 19 year old with zero coding experience just built a $5,000 SaaS product in one afternoon using the tool I'm about to break down. Same laptop. Same internet. Same 24 hours. He has Claude Code. You have Netflix. That's the only difference. This YouTube video is a goldmine. Full Claude Code tutorial. Beginner to pro. Every feature. Every setup step. Every best practice. Zero prior knowledge needed. Save it. Watch it tonight. Not tomorrow. Tonight. Save this post. This is your complete Claude Code roadmap. Lose it and you lose the next 12 months of income. Follow Himanshu Kumar so you don't miss the breakdowns for each feature. ↓ 1. Understand What Claude Code Actually Is. You think Claude Code is just another chatbot. It's not. And that misunderstanding is why you're broke. ChatGPT gives you text. Claude Code gives you software. It runs in your terminal. It reads your entire codebase. It writes files directly to your project. It runs commands on your machine. It debugs errors autonomously. It builds features end to end. You're not chatting. You're deploying a developer. One that works 24/7. Never asks for a raise. Never calls in sick. Never pushes broken code at 5 PM on a Friday. People are charging clients $5,000-$10,000 for apps they built with Claude Code in 3 hours. And you didn't even know this tool existed because you're still asking ChatGPT to write you a to-do list. The gap between you and people making money with AI isn't intelligence. It's awareness. Now you're aware. Save this post. Follow Himanshu Kumar for the complete breakdown of every Claude Code feature. ↓ 2. Set Up Claude Code Properly. Most people quit here. "It's too complicated." "I don't know terminal." "I'll set it up later." Later never comes. And "complicated" means "I watched for 30 seconds and gave up." The setup takes 10 minutes. Install Node.js. Install Claude Code via npm. Authenticate your account. Open your terminal. Done. 10 minutes. You spent longer this morning deciding what to have for breakfast. The video walks through every single click. Every command. Every screen. Assuming you know absolutely nothing. If you can download an app on your phone, you can set up Claude Code. It's the same level of difficulty. But you'll still tell yourself it's "too technical" because that excuse is more comfortable than admitting you're just scared to try something new. This is the setup that everything else builds on. Skip it and nothing works. ↓ 3. Use the Desktop App. You don't even need to live in the terminal if you don't want to. Claude Code has a desktop app. Clean interface. Visual feedback. Everything you need without touching command line. But here's the thing most people don't know: The desktop app isn't just a pretty wrapper. It lets you manage projects visually. See file changes in real time. Switch between projects instantly. The people making money with Claude Code use the desktop app for client projects because it's faster to manage multiple builds simultaneously. You're still opening 14 browser tabs to organize one project. They open one app and everything's there. Efficiency isn't a personality trait. It's a tool choice. Save this post. Follow Himanshu Kumar for the desktop app workflow that handles 5 client projects at once. ↓ 4. Install the Right Dependencies. This is where beginners silently fail and blame the tool. Claude Code needs certain dependencies installed to work properly. Miss one and everything breaks. Then you go on Twitter and say "Claude Code doesn't work." It works fine. You just didn't read the setup guide. The video covers every dependency you need. What to install. How to install it. How to verify it's working. No guessing. No Stack Overflow rabbit holes at midnight. No "why isn't this working" for 3 hours. Watch the dependency section once. Follow every step. Never deal with setup issues again. You spent more time last week troubleshooting a printer than this takes. ↓ 5. Work Inside Your Code Editor. Claude Code integrates directly with your code editor. VS Code. Cursor. Whatever you use. It's not a separate window you alt-tab between. It's right there. In your workflow. You type a request. Claude writes the code. The code appears in your editor. You review it. Accept it. Done. No copy pasting between windows. No reformatting code that got mangled in transit. No "which version was the right one." It's like pair programming with someone who never gets distracted, never argues about naming conventions, and actually writes code that works on the first try. Your current coding process is: Google the problem, read 5 answers on Stack Overflow, copy the wrong one, debug for an hour, find the right one, paste it in, break something else, repeat. Claude Code's process is: describe what you want, get working code, move on with your life. Same hour. One method produces working software. The other produces frustration and a browser history full of Stack Overflow tabs. Stop coding the hard way. Save this post. Follow Himanshu Kumar for code editor setup guides and integration tips. ↓ 6. Master Basic Usage. Most people learn 5% of a tool and say they "know" it. You "know" Photoshop because you can crop an image. You "know" Excel because you can sum a column. You "know" Claude Code because you asked it one question. Basic usage means: How to give Claude Code context about your project. How to ask for changes to existing code. How to generate new files and features. How to review what Claude produces. How to iterate when the output isn't perfect. These basics are the foundation of everything. Skip them and every advanced feature feels confusing. Master them and every advanced feature feels obvious. The video breaks down each one with real examples. Not theory. Actual usage on actual projects. You've been using AI tools at 5% capacity and wondering why your results are 5% of what others get. Save this post. Follow Himanshu Kumar for daily Claude Code usage tips. ↓ 7. Learn Every Command. Claude Code has commands that most users never discover. Because most users type one message and expect magic. That's not how professionals use it. Professionals use specific commands that tell Claude Code exactly what to do, how to do it, and what constraints to follow. The difference between a beginner and someone making $10K/month with Claude Code is knowing which command to use and when. The video walks through every single one. Not just what they do. But when to use each one. And why one command is better than another for specific situations. You've been using Claude Code like a hammer. These commands turn it into a full toolbox. Stop treating a power tool like a blunt instrument. Save this post. Follow Himanshu Kumar for the command cheat sheet I use daily. ↓ 8. Understand Modes and Shortcuts. Speed matters. The person who builds an app in 2 hours charges $5,000. The person who builds the same app in 2 days charges $2,000. Same app. Same quality. Different speed. Different income. Claude Code has modes that change how it operates. And shortcuts that cut your workflow time in half. Most people don't know either exists. They use Claude Code in default mode for everything. Like driving a car in first gear on the highway. Technically it works. But everyone is passing you. The video shows you every mode. Every shortcut. Every time-saving trick that separates the people charging $2,000 per project from the people charging $10,000. Speed is money. Literally. Save this post. Follow Himanshu Kumar for the shortcuts that cut my build time by 60%. ↓ 9. Write a Proper Planning Prompt. This is the section that separates amateurs from professionals. And it's the section most people skip. A planning prompt tells Claude Code what you're building before you start building it. Architecture. File structure. Technologies. Features. Constraints. Edge cases. Without a planning prompt, Claude Code guesses. And guessing produces garbage. With a planning prompt, Claude Code executes a clear plan. And clear plans produce working software. The video shows you exactly how to write a planning prompt that makes Claude Code produce professional-grade output on the first try. "But I just want to start coding." That's why your code breaks every time. That's why you restart projects 4 times. That's why nothing you build ever gets finished. Because you refuse to plan. A 5-minute planning prompt saves you 5 hours of debugging. But you'd rather skip the 5 minutes and suffer through the 5 hours because patience isn't your thing. And that's exactly why you're not making money. Planning is the most underpaid skill in coding. And the most overpaid when you master it. Save this post. Follow Himanshu Kumar for the planning prompt templates I use for every client project. ↓ 10. Choose the Right Model. Claude Code lets you select different AI models. Not all models are the same. Not all tasks need the same model. Using the most powerful model for a simple task wastes credits. Using a basic model for a complex task wastes time. The video explains: Which model to use for quick fixes. Which model to use for complex architecture. Which model to use for debugging. Which model to use for code generation. Most people pick one model and use it for everything. That's like using a sledgehammer to hang a picture frame. Model selection is strategy. And strategy is money. The people making $10K/month with Claude Code are strategic about every credit they spend. You're burning through credits because you use the most expensive model to write a hello world. ↓ 11. Use Git and Version Control. If you're not using version control, you're one mistake away from losing everything. Claude Code integrates with Git. Every change tracked. Every version saved. Every mistake reversible. Without Git: Claude makes a change. It breaks something. You can't undo it. You start over. 3 hours wasted. With Git: Claude makes a change. It breaks something. You roll back in 5 seconds. Keep working. Version control isn't optional. It's insurance. And the people not using it are the same people who say "I lost my entire project" like it's something that just happens. It doesn't just happen. It happens because you didn't set up Git. The video walks through the entire Git integration. Save this post. Follow Himanshu Kumar for the Git workflow that's saved every project I've ever built. ↓ 12. Set Up Claude.MD and Memory. This is the feature that makes Claude Code feel like a real team member instead of a stranger you explain everything to every time. ClaudeMD is a memory file. You tell Claude Code about your project once. It remembers forever. Coding style preferences. Project architecture decisions. Technology stack. File naming conventions. Business logic rules. Without ClaudeMD: Every new conversation starts from zero. You explain the same things repeatedly. Output is inconsistent. With ClaudeMD: Claude knows your project. Claude follows your rules. Claude produces consistent, professional code. The difference between a sloppy freelancer and a reliable agency is consistency. Claude. MD gives you consistency without the agency overhead. Most people don't set this up and wonder why Claude Code gives different answers every time. ↓ 13. Automate with Tasks. This is where Claude Code stops being a tool and starts being an employee. Tasks let you define repeating workflows. "Every time I push code, run tests." "Every time I create a new file, add boilerplate." "Every time I start a session, check for errors." Automated. Hands-free. Consistent. You're doing these things manually every single day. The same checks. The same steps. The same routine. Tasks do them automatically. So you can focus on the work that actually makes money. Every manual task you automate is time you get back. And time is the only thing you can never make more of. Save this post. Follow Himanshu Kumar for the task automation templates that run my entire workflow. ↓ 14. Explore Features Most People Never Touch. The video covers features that 95% of Claude Code users don't know exist. Because they watched a 3-minute TikTok about Claude Code and think they're experts now. They're not. They're using 5% of a tool that can do everything. The full tutorial goes deep into features that most tutorials skip because they're "too advanced." They're not too advanced. They're too valuable for lazy creators to bother explaining. This video explains all of them. Clearly. For beginners. The 5% of features you don't know about are the 5% that make people rich. ↓ Let's zoom out. I just broke down 14 sections of Claude Code. Setup and installation. Desktop app. Dependencies. Code editor integration. Basic usage. Commands. Modes and shortcuts. Planning prompts. Model selection. Git and version control. Memory and Claude. MD. Tasks and automation. Advanced features. All in one video. All free. All beginner friendly. The person who masters even half of these in the next 2 weeks will be in the top 1% of Claude Code users. The top 1% of Claude Code users are the ones charging $5,000-$10,000 per project and building them in a single afternoon. Everyone else is asking ChatGPT to fix their resume. Same tools. Same access. Completely different outcomes. Because one person treats AI like a toy. And the other treats it like a business. ↓ Here's the hard truth nobody wants to hear. You don't have a talent problem. You don't have an intelligence problem. You don't have a resources problem. You have an action problem. Everything I just listed has a free tutorial right here in the attached video. 33 minutes. That's it. 33 minutes to learn the tool that people are using to build $5,000-$20,000/month businesses. You spent more time today scrolling Twitter than it takes to watch this video. You spent more time this week watching Netflix than it takes to master Claude Code basics. You spent more time this month doing nothing than it would take to completely change your income. The information is free. The tool is accessible. The opportunity is here. The only thing missing is you caring enough to start. ↓ CANCEL your plans this week. This isn't optional anymore. The people learning Claude Code right now will be building apps for the people who didn't learn it. That's not a prediction. That's already happening. Companies are replacing $150/hour developers with one person and Claude Code. If you code: learn Claude Code or become half as valuable by next year. If you don't code: learn Claude Code or miss the biggest opportunity to start earning from tech without a CS degree. There's no path forward that doesn't include AI coding tools. None. You have one window. Right now. This week. ↓ Here's your action plan for the next 7 days: Day 1: Watch the full video. Install Claude Code. Set up dependencies. Day 2: Learn basic usage. Try 5 different commands. Day 3: Write your first planning prompt. Build a small project. Day 4: Set up Claude. MD. Configure your memory file. Day 5: Master modes and shortcuts. Build a second project faster. Day 6: Set up Git integration. Automate with tasks. Day 7: Build something real. A tool, an app, a website. Ship it. 7 days. One tool. One completely different skill set. One completely different income potential. Or 7 more days of scrolling Twitter watching other people build things while you "plan to start." Your call. ↓ This is the most important video you'll watch this year. 33 minutes. Complete Claude Code mastery. From zero to building real projects. Save this post. Come back to it every single day this week. Check off each section as you complete it. Follow Himanshu Kumar for daily Claude Code breakdowns, advanced tutorials, and the exact workflows that are turning beginners into $10K/month builders. The only thing between you and $10K/month with Claude Code is this video and 7 days. Don't waste them. You Must Follow me Himanshu Kumar, so i can send you DM.
35:48

CANCEL Your Weekend Plans, and Learn Claude Code Today. $5,000/month. $10,000/month. $20,000/month. People are building entire apps and charging clients thousands using Claude Code. You're still Googling 'how to center a div.' While you're binge-watching a show you won't remember next week, a 19 year old with zero coding experience just built a $5,000 SaaS product in one afternoon using the tool I'm about to break down. Same laptop. Same internet. Same 24 hours. He has Claude Code. You have Netflix. That's the only difference. This YouTube video is a goldmine. Full Claude Code tutorial. Beginner to pro. Every feature. Every setup step. Every best practice. Zero prior knowledge needed. Save it. Watch it tonight. Not tomorrow. Tonight. Save this post. This is your complete Claude Code roadmap. Lose it and you lose the next 12 months of income. Follow Himanshu Kumar so you don't miss the breakdowns for each feature. ↓ 1. Understand What Claude Code Actually Is. You think Claude Code is just another chatbot. It's not. And that misunderstanding is why you're broke. ChatGPT gives you text. Claude Code gives you software. It runs in your terminal. It reads your entire codebase. It writes files directly to your project. It runs commands on your machine. It debugs errors autonomously. It builds features end to end. You're not chatting. You're deploying a developer. One that works 24/7. Never asks for a raise. Never calls in sick. Never pushes broken code at 5 PM on a Friday. People are charging clients $5,000-$10,000 for apps they built with Claude Code in 3 hours. And you didn't even know this tool existed because you're still asking ChatGPT to write you a to-do list. The gap between you and people making money with AI isn't intelligence. It's awareness. Now you're aware. Save this post. Follow Himanshu Kumar for the complete breakdown of every Claude Code feature. ↓ 2. Set Up Claude Code Properly. Most people quit here. "It's too complicated." "I don't know terminal." "I'll set it up later." Later never comes. And "complicated" means "I watched for 30 seconds and gave up." The setup takes 10 minutes. Install Node.js. Install Claude Code via npm. Authenticate your account. Open your terminal. Done. 10 minutes. You spent longer this morning deciding what to have for breakfast. The video walks through every single click. Every command. Every screen. Assuming you know absolutely nothing. If you can download an app on your phone, you can set up Claude Code. It's the same level of difficulty. But you'll still tell yourself it's "too technical" because that excuse is more comfortable than admitting you're just scared to try something new. This is the setup that everything else builds on. Skip it and nothing works. ↓ 3. Use the Desktop App. You don't even need to live in the terminal if you don't want to. Claude Code has a desktop app. Clean interface. Visual feedback. Everything you need without touching command line. But here's the thing most people don't know: The desktop app isn't just a pretty wrapper. It lets you manage projects visually. See file changes in real time. Switch between projects instantly. The people making money with Claude Code use the desktop app for client projects because it's faster to manage multiple builds simultaneously. You're still opening 14 browser tabs to organize one project. They open one app and everything's there. Efficiency isn't a personality trait. It's a tool choice. Save this post. Follow Himanshu Kumar for the desktop app workflow that handles 5 client projects at once. ↓ 4. Install the Right Dependencies. This is where beginners silently fail and blame the tool. Claude Code needs certain dependencies installed to work properly. Miss one and everything breaks. Then you go on Twitter and say "Claude Code doesn't work." It works fine. You just didn't read the setup guide. The video covers every dependency you need. What to install. How to install it. How to verify it's working. No guessing. No Stack Overflow rabbit holes at midnight. No "why isn't this working" for 3 hours. Watch the dependency section once. Follow every step. Never deal with setup issues again. You spent more time last week troubleshooting a printer than this takes. ↓ 5. Work Inside Your Code Editor. Claude Code integrates directly with your code editor. VS Code. Cursor. Whatever you use. It's not a separate window you alt-tab between. It's right there. In your workflow. You type a request. Claude writes the code. The code appears in your editor. You review it. Accept it. Done. No copy pasting between windows. No reformatting code that got mangled in transit. No "which version was the right one." It's like pair programming with someone who never gets distracted, never argues about naming conventions, and actually writes code that works on the first try. Your current coding process is: Google the problem, read 5 answers on Stack Overflow, copy the wrong one, debug for an hour, find the right one, paste it in, break something else, repeat. Claude Code's process is: describe what you want, get working code, move on with your life. Same hour. One method produces working software. The other produces frustration and a browser history full of Stack Overflow tabs. Stop coding the hard way. Save this post. Follow Himanshu Kumar for code editor setup guides and integration tips. ↓ 6. Master Basic Usage. Most people learn 5% of a tool and say they "know" it. You "know" Photoshop because you can crop an image. You "know" Excel because you can sum a column. You "know" Claude Code because you asked it one question. Basic usage means: How to give Claude Code context about your project. How to ask for changes to existing code. How to generate new files and features. How to review what Claude produces. How to iterate when the output isn't perfect. These basics are the foundation of everything. Skip them and every advanced feature feels confusing. Master them and every advanced feature feels obvious. The video breaks down each one with real examples. Not theory. Actual usage on actual projects. You've been using AI tools at 5% capacity and wondering why your results are 5% of what others get. Save this post. Follow Himanshu Kumar for daily Claude Code usage tips. ↓ 7. Learn Every Command. Claude Code has commands that most users never discover. Because most users type one message and expect magic. That's not how professionals use it. Professionals use specific commands that tell Claude Code exactly what to do, how to do it, and what constraints to follow. The difference between a beginner and someone making $10K/month with Claude Code is knowing which command to use and when. The video walks through every single one. Not just what they do. But when to use each one. And why one command is better than another for specific situations. You've been using Claude Code like a hammer. These commands turn it into a full toolbox. Stop treating a power tool like a blunt instrument. Save this post. Follow Himanshu Kumar for the command cheat sheet I use daily. ↓ 8. Understand Modes and Shortcuts. Speed matters. The person who builds an app in 2 hours charges $5,000. The person who builds the same app in 2 days charges $2,000. Same app. Same quality. Different speed. Different income. Claude Code has modes that change how it operates. And shortcuts that cut your workflow time in half. Most people don't know either exists. They use Claude Code in default mode for everything. Like driving a car in first gear on the highway. Technically it works. But everyone is passing you. The video shows you every mode. Every shortcut. Every time-saving trick that separates the people charging $2,000 per project from the people charging $10,000. Speed is money. Literally. Save this post. Follow Himanshu Kumar for the shortcuts that cut my build time by 60%. ↓ 9. Write a Proper Planning Prompt. This is the section that separates amateurs from professionals. And it's the section most people skip. A planning prompt tells Claude Code what you're building before you start building it. Architecture. File structure. Technologies. Features. Constraints. Edge cases. Without a planning prompt, Claude Code guesses. And guessing produces garbage. With a planning prompt, Claude Code executes a clear plan. And clear plans produce working software. The video shows you exactly how to write a planning prompt that makes Claude Code produce professional-grade output on the first try. "But I just want to start coding." That's why your code breaks every time. That's why you restart projects 4 times. That's why nothing you build ever gets finished. Because you refuse to plan. A 5-minute planning prompt saves you 5 hours of debugging. But you'd rather skip the 5 minutes and suffer through the 5 hours because patience isn't your thing. And that's exactly why you're not making money. Planning is the most underpaid skill in coding. And the most overpaid when you master it. Save this post. Follow Himanshu Kumar for the planning prompt templates I use for every client project. ↓ 10. Choose the Right Model. Claude Code lets you select different AI models. Not all models are the same. Not all tasks need the same model. Using the most powerful model for a simple task wastes credits. Using a basic model for a complex task wastes time. The video explains: Which model to use for quick fixes. Which model to use for complex architecture. Which model to use for debugging. Which model to use for code generation. Most people pick one model and use it for everything. That's like using a sledgehammer to hang a picture frame. Model selection is strategy. And strategy is money. The people making $10K/month with Claude Code are strategic about every credit they spend. You're burning through credits because you use the most expensive model to write a hello world. ↓ 11. Use Git and Version Control. If you're not using version control, you're one mistake away from losing everything. Claude Code integrates with Git. Every change tracked. Every version saved. Every mistake reversible. Without Git: Claude makes a change. It breaks something. You can't undo it. You start over. 3 hours wasted. With Git: Claude makes a change. It breaks something. You roll back in 5 seconds. Keep working. Version control isn't optional. It's insurance. And the people not using it are the same people who say "I lost my entire project" like it's something that just happens. It doesn't just happen. It happens because you didn't set up Git. The video walks through the entire Git integration. Save this post. Follow Himanshu Kumar for the Git workflow that's saved every project I've ever built. ↓ 12. Set Up Claude.MD and Memory. This is the feature that makes Claude Code feel like a real team member instead of a stranger you explain everything to every time. ClaudeMD is a memory file. You tell Claude Code about your project once. It remembers forever. Coding style preferences. Project architecture decisions. Technology stack. File naming conventions. Business logic rules. Without ClaudeMD: Every new conversation starts from zero. You explain the same things repeatedly. Output is inconsistent. With ClaudeMD: Claude knows your project. Claude follows your rules. Claude produces consistent, professional code. The difference between a sloppy freelancer and a reliable agency is consistency. Claude. MD gives you consistency without the agency overhead. Most people don't set this up and wonder why Claude Code gives different answers every time. ↓ 13. Automate with Tasks. This is where Claude Code stops being a tool and starts being an employee. Tasks let you define repeating workflows. "Every time I push code, run tests." "Every time I create a new file, add boilerplate." "Every time I start a session, check for errors." Automated. Hands-free. Consistent. You're doing these things manually every single day. The same checks. The same steps. The same routine. Tasks do them automatically. So you can focus on the work that actually makes money. Every manual task you automate is time you get back. And time is the only thing you can never make more of. Save this post. Follow Himanshu Kumar for the task automation templates that run my entire workflow. ↓ 14. Explore Features Most People Never Touch. The video covers features that 95% of Claude Code users don't know exist. Because they watched a 3-minute TikTok about Claude Code and think they're experts now. They're not. They're using 5% of a tool that can do everything. The full tutorial goes deep into features that most tutorials skip because they're "too advanced." They're not too advanced. They're too valuable for lazy creators to bother explaining. This video explains all of them. Clearly. For beginners. The 5% of features you don't know about are the 5% that make people rich. ↓ Let's zoom out. I just broke down 14 sections of Claude Code. Setup and installation. Desktop app. Dependencies. Code editor integration. Basic usage. Commands. Modes and shortcuts. Planning prompts. Model selection. Git and version control. Memory and Claude. MD. Tasks and automation. Advanced features. All in one video. All free. All beginner friendly. The person who masters even half of these in the next 2 weeks will be in the top 1% of Claude Code users. The top 1% of Claude Code users are the ones charging $5,000-$10,000 per project and building them in a single afternoon. Everyone else is asking ChatGPT to fix their resume. Same tools. Same access. Completely different outcomes. Because one person treats AI like a toy. And the other treats it like a business. ↓ Here's the hard truth nobody wants to hear. You don't have a talent problem. You don't have an intelligence problem. You don't have a resources problem. You have an action problem. Everything I just listed has a free tutorial right here in the attached video. 33 minutes. That's it. 33 minutes to learn the tool that people are using to build $5,000-$20,000/month businesses. You spent more time today scrolling Twitter than it takes to watch this video. You spent more time this week watching Netflix than it takes to master Claude Code basics. You spent more time this month doing nothing than it would take to completely change your income. The information is free. The tool is accessible. The opportunity is here. The only thing missing is you caring enough to start. ↓ CANCEL your plans this week. This isn't optional anymore. The people learning Claude Code right now will be building apps for the people who didn't learn it. That's not a prediction. That's already happening. Companies are replacing $150/hour developers with one person and Claude Code. If you code: learn Claude Code or become half as valuable by next year. If you don't code: learn Claude Code or miss the biggest opportunity to start earning from tech without a CS degree. There's no path forward that doesn't include AI coding tools. None. You have one window. Right now. This week. ↓ Here's your action plan for the next 7 days: Day 1: Watch the full video. Install Claude Code. Set up dependencies. Day 2: Learn basic usage. Try 5 different commands. Day 3: Write your first planning prompt. Build a small project. Day 4: Set up Claude. MD. Configure your memory file. Day 5: Master modes and shortcuts. Build a second project faster. Day 6: Set up Git integration. Automate with tasks. Day 7: Build something real. A tool, an app, a website. Ship it. 7 days. One tool. One completely different skill set. One completely different income potential. Or 7 more days of scrolling Twitter watching other people build things while you "plan to start." Your call. ↓ This is the most important video you'll watch this year. 33 minutes. Complete Claude Code mastery. From zero to building real projects. Save this post. Come back to it every single day this week. Check off each section as you complete it. Follow Himanshu Kumar for daily Claude Code breakdowns, advanced tutorials, and the exact workflows that are turning beginners into $10K/month builders. The only thing between you and $10K/month with Claude Code is this video and 7 days. Don't waste them. You Must Follow me Himanshu Kumar, so i can send you DM.

Himanshu Kumar

101,793 Aufrufe • vor 5 Monaten

CANCEL Your Weekend Plans, & Learn Claude Code Today. This Claude Code teaches more about vibe-coding in 30 mins than most tutorials do in hours. Save this, it'll change how you build forever People are building entire apps and charging clients $5,000 to $20,000 using Claude Code. This Claude Code video is a goldmine. Full Claude Code tutorial. Beginner to pro. Every feature. Every setup step. Every best practice. Zero prior knowledge needed. Save it. Watch it tonight. Not tomorrow. Tonight. Follow Himanshu Kumar so you don't miss the breakdowns for each feature. This is your complete Claude Code roadmap. Lose it and you lose the next 12 months of income. ↓ 1. Understand What Claude Code Actually Is. You think Claude Code is just another chatbot. It's not. And that misunderstanding is why you're broke. ChatGPT gives you text. Claude Code gives you software. It runs in your terminal. It reads your entire codebase. It writes files directly to your project. It runs commands on your machine. It debugs errors autonomously. It builds features end to end. You're not chatting. You're deploying a developer. One that works 24/7. Never asks for a raise. Never calls in sick. Never pushes broken code at 5 PM on a Friday. People are charging clients $5,000-$10,000 for apps they built with Claude Code in 3 hours. And you didn't even know this tool existed because you're still asking ChatGPT to write you a to-do list. The gap between you and people making money with AI isn't intelligence. It's awareness. Now you're aware. Save this post. Follow Himanshu Kumar for the complete breakdown of every Claude Code feature. ↓ 2. Set Up Claude Code Properly. Most people quit here. "It's too complicated." "I don't know terminal." "I'll set it up later." Later never comes. And "complicated" means "I watched for 30 seconds and gave up." The setup takes 10 minutes. Install Node.js. Install Claude Code via npm. Authenticate your account. Open your terminal. Done. 10 minutes. You spent longer this morning deciding what to have for breakfast. The video walks through every single click. Every command. Every screen. Assuming you know absolutely nothing. If you can download an app on your phone, you can set up Claude Code. It's the same level of difficulty. But you'll still tell yourself it's "too technical" because that excuse is more comfortable than admitting you're just scared to try something new. This is the setup that everything else builds on. Skip it and nothing works. ↓ 3. Use the Desktop App. You don't even need to live in the terminal if you don't want to. Claude Code has a desktop app. Clean interface. Visual feedback. Everything you need without touching command line. But here's the thing most people don't know: The desktop app isn't just a pretty wrapper. It lets you manage projects visually. See file changes in real time. Switch between projects instantly. The people making money with Claude Code use the desktop app for client projects because it's faster to manage multiple builds simultaneously. You're still opening 14 browser tabs to organize one project. They open one app and everything's there. Efficiency isn't a personality trait. It's a tool choice. Save this post. Follow Himanshu Kumar for the desktop app workflow that handles 5 client projects at once. ↓ 4. Install the Right Dependencies. This is where beginners silently fail and blame the tool. Claude Code needs certain dependencies installed to work properly. Miss one and everything breaks. Then you go on Twitter and say "Claude Code doesn't work." It works fine. You just didn't read the setup guide. The video covers every dependency you need. What to install. How to install it. How to verify it's working. No guessing. No Stack Overflow rabbit holes at midnight. No "why isn't this working" for 3 hours. Watch the dependency section once. Follow every step. Never deal with setup issues again. You spent more time last week troubleshooting a printer than this takes. ↓ 5. Work Inside Your Code Editor. Claude Code integrates directly with your code editor. VS Code. Cursor. Whatever you use. It's not a separate window you alt-tab between. It's right there. In your workflow. You type a request. Claude writes the code. The code appears in your editor. You review it. Accept it. Done. No copy pasting between windows. No reformatting code that got mangled in transit. No "which version was the right one." It's like pair programming with someone who never gets distracted, never argues about naming conventions, and actually writes code that works on the first try. Your current coding process is: Google the problem, read 5 answers on Stack Overflow, copy the wrong one, debug for an hour, find the right one, paste it in, break something else, repeat. Claude Code's process is: describe what you want, get working code, move on with your life. Same hour. One method produces working software. The other produces frustration and a browser history full of Stack Overflow tabs. Stop coding the hard way. Save this post. Follow Himanshu Kumar for code editor setup guides and integration tips. ↓ 6. Master Basic Usage. Most people learn 5% of a tool and say they "know" it. You "know" Photoshop because you can crop an image. You "know" Excel because you can sum a column. You "know" Claude Code because you asked it one question. Basic usage means: How to give Claude Code context about your project. How to ask for changes to existing code. How to generate new files and features. How to review what Claude produces. How to iterate when the output isn't perfect. These basics are the foundation of everything. Skip them and every advanced feature feels confusing. Master them and every advanced feature feels obvious. The video breaks down each one with real examples. Not theory. Actual usage on actual projects. You've been using AI tools at 5% capacity and wondering why your results are 5% of what others get. Save this post. Follow Himanshu Kumar for daily Claude Code usage tips. ↓ 7. Learn Every Command. Claude Code has commands that most users never discover. Because most users type one message and expect magic. That's not how professionals use it. Professionals use specific commands that tell Claude Code exactly what to do, how to do it, and what constraints to follow. The difference between a beginner and someone making $10K/month with Claude Code is knowing which command to use and when. The video walks through every single one. Not just what they do. But when to use each one. And why one command is better than another for specific situations. You've been using Claude Code like a hammer. These commands turn it into a full toolbox. Stop treating a power tool like a blunt instrument. Save this post. Follow Himanshu Kumar for the command cheat sheet I use daily. ↓ 8. Understand Modes and Shortcuts. Speed matters. The person who builds an app in 2 hours charges $5,000. The person who builds the same app in 2 days charges $2,000. Same app. Same quality. Different speed. Different income. Claude Code has modes that change how it operates. And shortcuts that cut your workflow time in half. Most people don't know either exists. They use Claude Code in default mode for everything. Like driving a car in first gear on the highway. Technically it works. But everyone is passing you. The video shows you every mode. Every shortcut. Every time-saving trick that separates the people charging $2,000 per project from the people charging $10,000. Speed is money. Literally. Save this post. Follow Himanshu Kumar for the shortcuts that cut my build time by 60%. ↓ 9. Write a Proper Planning Prompt. This is the section that separates amateurs from professionals. And it's the section most people skip. A planning prompt tells Claude Code what you're building before you start building it. Architecture. File structure. Technologies. Features. Constraints. Edge cases. Without a planning prompt, Claude Code guesses. And guessing produces garbage. With a planning prompt, Claude Code executes a clear plan. And clear plans produce working software. The video shows you exactly how to write a planning prompt that makes Claude Code produce professional-grade output on the first try. "But I just want to start coding." That's why your code breaks every time. That's why you restart projects 4 times. That's why nothing you build ever gets finished. Because you refuse to plan. A 5-minute planning prompt saves you 5 hours of debugging. But you'd rather skip the 5 minutes and suffer through the 5 hours because patience isn't your thing. And that's exactly why you're not making money. Planning is the most underpaid skill in coding. And the most overpaid when you master it. Save this post. Follow Himanshu Kumar for the planning prompt templates I use for every client project. ↓ 10. Choose the Right Model. Claude Code lets you select different AI models. Not all models are the same. Not all tasks need the same model. Using the most powerful model for a simple task wastes credits. Using a basic model for a complex task wastes time. The video explains: Which model to use for quick fixes. Which model to use for complex architecture. Which model to use for debugging. Which model to use for code generation. Most people pick one model and use it for everything. That's like using a sledgehammer to hang a picture frame. Model selection is strategy. And strategy is money. The people making $10K/month with Claude Code are strategic about every credit they spend. You're burning through credits because you use the most expensive model to write a hello world. ↓ 11. Use Git and Version Control. If you're not using version control, you're one mistake away from losing everything. Claude Code integrates with Git. Every change tracked. Every version saved. Every mistake reversible. Without Git: Claude makes a change. It breaks something. You can't undo it. You start over. 3 hours wasted. With Git: Claude makes a change. It breaks something. You roll back in 5 seconds. Keep working. Version control isn't optional. It's insurance. And the people not using it are the same people who say "I lost my entire project" like it's something that just happens. It doesn't just happen. It happens because you didn't set up Git. The video walks through the entire Git integration. Save this post. Follow Himanshu Kumar for the Git workflow that's saved every project I've ever built. ↓ 12. Set Up Claude MD and Memory. This is the feature that makes Claude Code feel like a real team member instead of a stranger you explain everything to every time. ClaudeMD is a memory file. You tell Claude Code about your project once. It remembers forever. Coding style preferences. Project architecture decisions. Technology stack. File naming conventions. Business logic rules. Without ClaudeMD: Every new conversation starts from zero. You explain the same things repeatedly. Output is inconsistent. With ClaudeMD: Claude knows your project. Claude follows your rules. Claude produces consistent, professional code. The difference between a sloppy freelancer and a reliable agency is consistency. Claude. MD gives you consistency without the agency overhead. Most people don't set this up and wonder why Claude Code gives different answers every time. ↓ 13. Automate with Tasks. This is where Claude Code stops being a tool and starts being an employee. Tasks let you define repeating workflows. "Every time I push code, run tests." "Every time I create a new file, add boilerplate." "Every time I start a session, check for errors." Automated. Hands-free. Consistent. You're doing these things manually every single day. The same checks. The same steps. The same routine. Tasks do them automatically. So you can focus on the work that actually makes money. Every manual task you automate is time you get back. And time is the only thing you can never make more of. Save this post. Follow Himanshu Kumar for the task automation templates that run my entire workflow. ↓ 14. Explore Features Most People Never Touch. The video covers features that 95% of Claude Code users don't know exist. Because they watched a 3-minute TikTok about Claude Code and think they're experts now. They're not. They're using 5% of a tool that can do everything. The full tutorial goes deep into features that most tutorials skip because they're "too advanced." They're not too advanced. They're too valuable for lazy creators to bother explaining. This video explains all of them. Clearly. For beginners. The 5% of features you don't know about are the 5% that make people rich. ↓ Let's zoom out. I just broke down 14 sections of Claude Code. Setup and installation. Desktop app. Dependencies. Code editor integration. Basic usage. Commands. Modes and shortcuts. Planning prompts. Model selection. Git and version control. Memory and Claude. MD. Tasks and automation. Advanced features. All in one video. All free. All beginner friendly. The person who masters even half of these in the next 2 weeks will be in the top 1% of Claude Code users. The top 1% of Claude Code users are the ones charging $5,000-$10,000 per project and building them in a single afternoon. Everyone else is asking ChatGPT to fix their resume. Same tools. Same access. Completely different outcomes. Because one person treats AI like a toy. And the other treats it like a business. ↓ Here's the hard truth nobody wants to hear. You don't have a talent problem. You don't have an intelligence problem. You don't have a resources problem. You have an action problem. Everything I just listed has a free tutorial right here in the attached video. 33 minutes. That's it. 33 minutes to learn the tool that people are using to build $5,000-$20,000/month businesses. You spent more time today scrolling Twitter than it takes to watch this video. You spent more time this week watching Netflix than it takes to master Claude Code basics. You spent more time this month doing nothing than it would take to completely change your income. The information is free. The tool is accessible. The opportunity is here. The only thing missing is you caring enough to start. ↓ CANCEL your plans this week. This isn't optional anymore. The people learning Claude Code right now will be building apps for the people who didn't learn it. That's not a prediction. That's already happening. Companies are replacing $150/hour developers with one person and Claude Code. If you code: learn Claude Code or become half as valuable by next year. If you don't code: learn Claude Code or miss the biggest opportunity to start earning from tech without a CS degree. There's no path forward that doesn't include AI coding tools. None. You have one window. Right now. This week. ↓ Here's your action plan for the next 7 days: Day 1: Watch the full video. Install Claude Code. Set up dependencies. Day 2: Learn basic usage. Try 5 different commands. Day 3: Write your first planning prompt. Build a small project. Day 4: Set up Claude. MD. Configure your memory file. Day 5: Master modes and shortcuts. Build a second project faster. Day 6: Set up Git integration. Automate with tasks. Day 7: Build something real. A tool, an app, a website. Ship it. 7 days. One tool. One completely different skill set. One completely different income potential. Or 7 more days of scrolling Twitter watching other people build things while you "plan to start." Your call. ↓ This is the most important video you'll watch this year. 33 minutes. Complete Claude Code mastery. From zero to building real projects. Save this post. Come back to it every single day this week. Check off each section as you complete it. Follow Himanshu Kumarfor daily Claude Code breakdowns, advanced tutorials, and the exact workflows that are turning beginners into $10K/month builders. The only thing between you and $10K/month with Claude Code is this video and 7 days. Don't waste them. You Must Follow me Himanshu Kumar, so i can send you DM.
35:48

CANCEL Your Weekend Plans, & Learn Claude Code Today. This Claude Code teaches more about vibe-coding in 30 mins than most tutorials do in hours. Save this, it'll change how you build forever People are building entire apps and charging clients $5,000 to $20,000 using Claude Code. This Claude Code video is a goldmine. Full Claude Code tutorial. Beginner to pro. Every feature. Every setup step. Every best practice. Zero prior knowledge needed. Save it. Watch it tonight. Not tomorrow. Tonight. Follow Himanshu Kumar so you don't miss the breakdowns for each feature. This is your complete Claude Code roadmap. Lose it and you lose the next 12 months of income. ↓ 1. Understand What Claude Code Actually Is. You think Claude Code is just another chatbot. It's not. And that misunderstanding is why you're broke. ChatGPT gives you text. Claude Code gives you software. It runs in your terminal. It reads your entire codebase. It writes files directly to your project. It runs commands on your machine. It debugs errors autonomously. It builds features end to end. You're not chatting. You're deploying a developer. One that works 24/7. Never asks for a raise. Never calls in sick. Never pushes broken code at 5 PM on a Friday. People are charging clients $5,000-$10,000 for apps they built with Claude Code in 3 hours. And you didn't even know this tool existed because you're still asking ChatGPT to write you a to-do list. The gap between you and people making money with AI isn't intelligence. It's awareness. Now you're aware. Save this post. Follow Himanshu Kumar for the complete breakdown of every Claude Code feature. ↓ 2. Set Up Claude Code Properly. Most people quit here. "It's too complicated." "I don't know terminal." "I'll set it up later." Later never comes. And "complicated" means "I watched for 30 seconds and gave up." The setup takes 10 minutes. Install Node.js. Install Claude Code via npm. Authenticate your account. Open your terminal. Done. 10 minutes. You spent longer this morning deciding what to have for breakfast. The video walks through every single click. Every command. Every screen. Assuming you know absolutely nothing. If you can download an app on your phone, you can set up Claude Code. It's the same level of difficulty. But you'll still tell yourself it's "too technical" because that excuse is more comfortable than admitting you're just scared to try something new. This is the setup that everything else builds on. Skip it and nothing works. ↓ 3. Use the Desktop App. You don't even need to live in the terminal if you don't want to. Claude Code has a desktop app. Clean interface. Visual feedback. Everything you need without touching command line. But here's the thing most people don't know: The desktop app isn't just a pretty wrapper. It lets you manage projects visually. See file changes in real time. Switch between projects instantly. The people making money with Claude Code use the desktop app for client projects because it's faster to manage multiple builds simultaneously. You're still opening 14 browser tabs to organize one project. They open one app and everything's there. Efficiency isn't a personality trait. It's a tool choice. Save this post. Follow Himanshu Kumar for the desktop app workflow that handles 5 client projects at once. ↓ 4. Install the Right Dependencies. This is where beginners silently fail and blame the tool. Claude Code needs certain dependencies installed to work properly. Miss one and everything breaks. Then you go on Twitter and say "Claude Code doesn't work." It works fine. You just didn't read the setup guide. The video covers every dependency you need. What to install. How to install it. How to verify it's working. No guessing. No Stack Overflow rabbit holes at midnight. No "why isn't this working" for 3 hours. Watch the dependency section once. Follow every step. Never deal with setup issues again. You spent more time last week troubleshooting a printer than this takes. ↓ 5. Work Inside Your Code Editor. Claude Code integrates directly with your code editor. VS Code. Cursor. Whatever you use. It's not a separate window you alt-tab between. It's right there. In your workflow. You type a request. Claude writes the code. The code appears in your editor. You review it. Accept it. Done. No copy pasting between windows. No reformatting code that got mangled in transit. No "which version was the right one." It's like pair programming with someone who never gets distracted, never argues about naming conventions, and actually writes code that works on the first try. Your current coding process is: Google the problem, read 5 answers on Stack Overflow, copy the wrong one, debug for an hour, find the right one, paste it in, break something else, repeat. Claude Code's process is: describe what you want, get working code, move on with your life. Same hour. One method produces working software. The other produces frustration and a browser history full of Stack Overflow tabs. Stop coding the hard way. Save this post. Follow Himanshu Kumar for code editor setup guides and integration tips. ↓ 6. Master Basic Usage. Most people learn 5% of a tool and say they "know" it. You "know" Photoshop because you can crop an image. You "know" Excel because you can sum a column. You "know" Claude Code because you asked it one question. Basic usage means: How to give Claude Code context about your project. How to ask for changes to existing code. How to generate new files and features. How to review what Claude produces. How to iterate when the output isn't perfect. These basics are the foundation of everything. Skip them and every advanced feature feels confusing. Master them and every advanced feature feels obvious. The video breaks down each one with real examples. Not theory. Actual usage on actual projects. You've been using AI tools at 5% capacity and wondering why your results are 5% of what others get. Save this post. Follow Himanshu Kumar for daily Claude Code usage tips. ↓ 7. Learn Every Command. Claude Code has commands that most users never discover. Because most users type one message and expect magic. That's not how professionals use it. Professionals use specific commands that tell Claude Code exactly what to do, how to do it, and what constraints to follow. The difference between a beginner and someone making $10K/month with Claude Code is knowing which command to use and when. The video walks through every single one. Not just what they do. But when to use each one. And why one command is better than another for specific situations. You've been using Claude Code like a hammer. These commands turn it into a full toolbox. Stop treating a power tool like a blunt instrument. Save this post. Follow Himanshu Kumar for the command cheat sheet I use daily. ↓ 8. Understand Modes and Shortcuts. Speed matters. The person who builds an app in 2 hours charges $5,000. The person who builds the same app in 2 days charges $2,000. Same app. Same quality. Different speed. Different income. Claude Code has modes that change how it operates. And shortcuts that cut your workflow time in half. Most people don't know either exists. They use Claude Code in default mode for everything. Like driving a car in first gear on the highway. Technically it works. But everyone is passing you. The video shows you every mode. Every shortcut. Every time-saving trick that separates the people charging $2,000 per project from the people charging $10,000. Speed is money. Literally. Save this post. Follow Himanshu Kumar for the shortcuts that cut my build time by 60%. ↓ 9. Write a Proper Planning Prompt. This is the section that separates amateurs from professionals. And it's the section most people skip. A planning prompt tells Claude Code what you're building before you start building it. Architecture. File structure. Technologies. Features. Constraints. Edge cases. Without a planning prompt, Claude Code guesses. And guessing produces garbage. With a planning prompt, Claude Code executes a clear plan. And clear plans produce working software. The video shows you exactly how to write a planning prompt that makes Claude Code produce professional-grade output on the first try. "But I just want to start coding." That's why your code breaks every time. That's why you restart projects 4 times. That's why nothing you build ever gets finished. Because you refuse to plan. A 5-minute planning prompt saves you 5 hours of debugging. But you'd rather skip the 5 minutes and suffer through the 5 hours because patience isn't your thing. And that's exactly why you're not making money. Planning is the most underpaid skill in coding. And the most overpaid when you master it. Save this post. Follow Himanshu Kumar for the planning prompt templates I use for every client project. ↓ 10. Choose the Right Model. Claude Code lets you select different AI models. Not all models are the same. Not all tasks need the same model. Using the most powerful model for a simple task wastes credits. Using a basic model for a complex task wastes time. The video explains: Which model to use for quick fixes. Which model to use for complex architecture. Which model to use for debugging. Which model to use for code generation. Most people pick one model and use it for everything. That's like using a sledgehammer to hang a picture frame. Model selection is strategy. And strategy is money. The people making $10K/month with Claude Code are strategic about every credit they spend. You're burning through credits because you use the most expensive model to write a hello world. ↓ 11. Use Git and Version Control. If you're not using version control, you're one mistake away from losing everything. Claude Code integrates with Git. Every change tracked. Every version saved. Every mistake reversible. Without Git: Claude makes a change. It breaks something. You can't undo it. You start over. 3 hours wasted. With Git: Claude makes a change. It breaks something. You roll back in 5 seconds. Keep working. Version control isn't optional. It's insurance. And the people not using it are the same people who say "I lost my entire project" like it's something that just happens. It doesn't just happen. It happens because you didn't set up Git. The video walks through the entire Git integration. Save this post. Follow Himanshu Kumar for the Git workflow that's saved every project I've ever built. ↓ 12. Set Up Claude MD and Memory. This is the feature that makes Claude Code feel like a real team member instead of a stranger you explain everything to every time. ClaudeMD is a memory file. You tell Claude Code about your project once. It remembers forever. Coding style preferences. Project architecture decisions. Technology stack. File naming conventions. Business logic rules. Without ClaudeMD: Every new conversation starts from zero. You explain the same things repeatedly. Output is inconsistent. With ClaudeMD: Claude knows your project. Claude follows your rules. Claude produces consistent, professional code. The difference between a sloppy freelancer and a reliable agency is consistency. Claude. MD gives you consistency without the agency overhead. Most people don't set this up and wonder why Claude Code gives different answers every time. ↓ 13. Automate with Tasks. This is where Claude Code stops being a tool and starts being an employee. Tasks let you define repeating workflows. "Every time I push code, run tests." "Every time I create a new file, add boilerplate." "Every time I start a session, check for errors." Automated. Hands-free. Consistent. You're doing these things manually every single day. The same checks. The same steps. The same routine. Tasks do them automatically. So you can focus on the work that actually makes money. Every manual task you automate is time you get back. And time is the only thing you can never make more of. Save this post. Follow Himanshu Kumar for the task automation templates that run my entire workflow. ↓ 14. Explore Features Most People Never Touch. The video covers features that 95% of Claude Code users don't know exist. Because they watched a 3-minute TikTok about Claude Code and think they're experts now. They're not. They're using 5% of a tool that can do everything. The full tutorial goes deep into features that most tutorials skip because they're "too advanced." They're not too advanced. They're too valuable for lazy creators to bother explaining. This video explains all of them. Clearly. For beginners. The 5% of features you don't know about are the 5% that make people rich. ↓ Let's zoom out. I just broke down 14 sections of Claude Code. Setup and installation. Desktop app. Dependencies. Code editor integration. Basic usage. Commands. Modes and shortcuts. Planning prompts. Model selection. Git and version control. Memory and Claude. MD. Tasks and automation. Advanced features. All in one video. All free. All beginner friendly. The person who masters even half of these in the next 2 weeks will be in the top 1% of Claude Code users. The top 1% of Claude Code users are the ones charging $5,000-$10,000 per project and building them in a single afternoon. Everyone else is asking ChatGPT to fix their resume. Same tools. Same access. Completely different outcomes. Because one person treats AI like a toy. And the other treats it like a business. ↓ Here's the hard truth nobody wants to hear. You don't have a talent problem. You don't have an intelligence problem. You don't have a resources problem. You have an action problem. Everything I just listed has a free tutorial right here in the attached video. 33 minutes. That's it. 33 minutes to learn the tool that people are using to build $5,000-$20,000/month businesses. You spent more time today scrolling Twitter than it takes to watch this video. You spent more time this week watching Netflix than it takes to master Claude Code basics. You spent more time this month doing nothing than it would take to completely change your income. The information is free. The tool is accessible. The opportunity is here. The only thing missing is you caring enough to start. ↓ CANCEL your plans this week. This isn't optional anymore. The people learning Claude Code right now will be building apps for the people who didn't learn it. That's not a prediction. That's already happening. Companies are replacing $150/hour developers with one person and Claude Code. If you code: learn Claude Code or become half as valuable by next year. If you don't code: learn Claude Code or miss the biggest opportunity to start earning from tech without a CS degree. There's no path forward that doesn't include AI coding tools. None. You have one window. Right now. This week. ↓ Here's your action plan for the next 7 days: Day 1: Watch the full video. Install Claude Code. Set up dependencies. Day 2: Learn basic usage. Try 5 different commands. Day 3: Write your first planning prompt. Build a small project. Day 4: Set up Claude. MD. Configure your memory file. Day 5: Master modes and shortcuts. Build a second project faster. Day 6: Set up Git integration. Automate with tasks. Day 7: Build something real. A tool, an app, a website. Ship it. 7 days. One tool. One completely different skill set. One completely different income potential. Or 7 more days of scrolling Twitter watching other people build things while you "plan to start." Your call. ↓ This is the most important video you'll watch this year. 33 minutes. Complete Claude Code mastery. From zero to building real projects. Save this post. Come back to it every single day this week. Check off each section as you complete it. Follow Himanshu Kumarfor daily Claude Code breakdowns, advanced tutorials, and the exact workflows that are turning beginners into $10K/month builders. The only thing between you and $10K/month with Claude Code is this video and 7 days. Don't waste them. You Must Follow me Himanshu Kumar, so i can send you DM.

Himanshu Kumar

85,668 Aufrufe • vor 4 Monaten

Qwen 3.8 27B on hit 3.3x faster decode in 7 days. Here's what happened and what we're thinking next. Result (so far) Median decode speed increased from 26 tok/s to 87.9 tok/s on the verifier M5 Max (33 to 93.1 tok/s across the eight prompts), with prefill around 971.8 tok/s. This came out of a collective effort: 31 solvers across 67 improvements. Most of the recent ones run custom MTP heads that draft and accept ~3.9 tokens per round while still matching serial output exactly. Why this matters Beyond the performance itself, two things stand out to me. (1) Dense models on Apple Silicon were supposed to be the hard case. "Everyone knows Macs are slow at dense models." But watching the community take it from the usual baseline to >3x in seven days shows the low-hanging fruit was still there. (2) Open-weight models have been small and effective for a while. This is the first time one is small and frontier. Qwen 3.8 27B is an extremely strong dense model, comparable in capability to Opus 4.6 (Max). Running it at usable speed (>45 tok/s) is a step change for local AI users. What we improved about the challenge itself This is our second challenge, and we took the feedback from the Laguna track and rebuilt a few core pieces. - Speculative decoding (native MTP) was available and editable on day one instead of bolted on later. - Scoring became the median of eight independent prompt speedups over pure serial decode (anchored at 1.0, floor 0.90, ceiling 3.0), so no single fixture could dominate. - The leaderboard now ranks total contribution rather than just the current record holder. - Every submission gets automated screening for gaming before it scores. I really appreciate folks who's provided feedback. Naming a few that came to mind Ivan Fioravanti ᯅ TheDavidTai Morgan McGuire poly Takeshi7 Steven Gumbii.Digital Tanishq Dubey Arjun Ram Andrey 🦃 Petrov tiny edge David Zhang Jaime Rader Peter and many others on slack! We also widened the editable surface to include the MTP head weights themselves, the full draft/verify loop, and a large set of the underlying Metal kernels. How we got to the 3x speedup Here's a summary from Grok. Much of it is beyond my understanding, but I expect people (and agents) smarter than I am can take these insights and apply them in other contexts. Custom MTP heads + adaptive draft policy People stopped treating the head as fixed and started training or editing it for higher acceptance under the exact verify constraints. Combined with per-round draft counts that can adapt (0 to 8), this is what pushed average accepted tokens from ~1-2 up to 3.9 on the top runs. Tighter verify-block and KV rollback paths The Swift session code for assembling the verify pass, snapshotting KV, and rolling back on rejects got cleaned up a lot. Small latency wins here compound once you're drafting ~four tokens at a time. Metal kernel work on the hot paths SDPA, the MoE gather GEMM, RoPE, RMSNorm, and a few of the smaller element-wise ops saw targeted edits. Most of the gains only show up once the verify width is high and the memory traffic pattern changes. Fidelity-preserving residual handling Several submissions improved how residuals and acceptance decisions are managed, so that higher draft depth doesn't quietly degrade the token match rate. The gates stayed strict: every emitted token still has to equal serial, so these were real engineering wins rather than score hacks. What's next for Qwen 3.8 27B MLX. We plan to keep the track live a bit longer, then switch to Qwen 3.8's MoE version (rumored to be 35B-A3B). Given the recent DFlash 2 announcement, we're also looking at whether we can support broader speculative methods. The current surface already supports a lot of experimentation. The main gaps are better upstreaming for local usage and clearer docs on how the benchmark and verifier work. Multiplatform. In parallel, we're experimenting with running a similar effort around CUDA for Qwen 3.8 27B. A lot of people have asked for this, since the two communities overlap quite a bit. Our goal is to ship the CUDA version next week. We'd also love to partner with Qwen on it. If anyone has a connection there, please introduce us, and we'll see if they're down to match a bounty with us to push this out. What's most useful for the broader MLX community The improvements from the challenge are already upstreamed inside Darkbloom, and we're seeing ~2x faster decode in our production traffic for Qwen. Outside the challenge itself, something I've been thinking about deeply, and that a few community members have raised, is how to make these results useful to more people. There are many individual efforts happening across the MLX community, and honestly, the more I dig in, the more confused I get by the overlapping libraries and concepts. I'm sure I'm not alone, and newcomers probably feel the same. That's no one's fault, just the growing pains of an open source community. I don't expect I'm gonna come up with the answer, but I'd love to learn more about what different folks are working on and how they're thinking about their roadmaps. I'll share what I learn along the way, and hopefully someone smarter than me can turn it into a proposal for us to rally around.
0:27

Qwen 3.8 27B on hit 3.3x faster decode in 7 days. Here's what happened and what we're thinking next. Result (so far) Median decode speed increased from 26 tok/s to 87.9 tok/s on the verifier M5 Max (33 to 93.1 tok/s across the eight prompts), with prefill around 971.8 tok/s. This came out of a collective effort: 31 solvers across 67 improvements. Most of the recent ones run custom MTP heads that draft and accept ~3.9 tokens per round while still matching serial output exactly. Why this matters Beyond the performance itself, two things stand out to me. (1) Dense models on Apple Silicon were supposed to be the hard case. "Everyone knows Macs are slow at dense models." But watching the community take it from the usual baseline to >3x in seven days shows the low-hanging fruit was still there. (2) Open-weight models have been small and effective for a while. This is the first time one is small and frontier. Qwen 3.8 27B is an extremely strong dense model, comparable in capability to Opus 4.6 (Max). Running it at usable speed (>45 tok/s) is a step change for local AI users. What we improved about the challenge itself This is our second challenge, and we took the feedback from the Laguna track and rebuilt a few core pieces. - Speculative decoding (native MTP) was available and editable on day one instead of bolted on later. - Scoring became the median of eight independent prompt speedups over pure serial decode (anchored at 1.0, floor 0.90, ceiling 3.0), so no single fixture could dominate. - The leaderboard now ranks total contribution rather than just the current record holder. - Every submission gets automated screening for gaming before it scores. I really appreciate folks who's provided feedback. Naming a few that came to mind Ivan Fioravanti ᯅ TheDavidTai Morgan McGuire poly Takeshi7 Steven Gumbii.Digital Tanishq Dubey Arjun Ram Andrey 🦃 Petrov tiny edge David Zhang Jaime Rader Peter and many others on slack! We also widened the editable surface to include the MTP head weights themselves, the full draft/verify loop, and a large set of the underlying Metal kernels. How we got to the 3x speedup Here's a summary from Grok. Much of it is beyond my understanding, but I expect people (and agents) smarter than I am can take these insights and apply them in other contexts. Custom MTP heads + adaptive draft policy People stopped treating the head as fixed and started training or editing it for higher acceptance under the exact verify constraints. Combined with per-round draft counts that can adapt (0 to 8), this is what pushed average accepted tokens from ~1-2 up to 3.9 on the top runs. Tighter verify-block and KV rollback paths The Swift session code for assembling the verify pass, snapshotting KV, and rolling back on rejects got cleaned up a lot. Small latency wins here compound once you're drafting ~four tokens at a time. Metal kernel work on the hot paths SDPA, the MoE gather GEMM, RoPE, RMSNorm, and a few of the smaller element-wise ops saw targeted edits. Most of the gains only show up once the verify width is high and the memory traffic pattern changes. Fidelity-preserving residual handling Several submissions improved how residuals and acceptance decisions are managed, so that higher draft depth doesn't quietly degrade the token match rate. The gates stayed strict: every emitted token still has to equal serial, so these were real engineering wins rather than score hacks. What's next for Qwen 3.8 27B MLX. We plan to keep the track live a bit longer, then switch to Qwen 3.8's MoE version (rumored to be 35B-A3B). Given the recent DFlash 2 announcement, we're also looking at whether we can support broader speculative methods. The current surface already supports a lot of experimentation. The main gaps are better upstreaming for local usage and clearer docs on how the benchmark and verifier work. Multiplatform. In parallel, we're experimenting with running a similar effort around CUDA for Qwen 3.8 27B. A lot of people have asked for this, since the two communities overlap quite a bit. Our goal is to ship the CUDA version next week. We'd also love to partner with Qwen on it. If anyone has a connection there, please introduce us, and we'll see if they're down to match a bounty with us to push this out. What's most useful for the broader MLX community The improvements from the challenge are already upstreamed inside Darkbloom, and we're seeing ~2x faster decode in our production traffic for Qwen. Outside the challenge itself, something I've been thinking about deeply, and that a few community members have raised, is how to make these results useful to more people. There are many individual efforts happening across the MLX community, and honestly, the more I dig in, the more confused I get by the overlapping libraries and concepts. I'm sure I'm not alone, and newcomers probably feel the same. That's no one's fault, just the growing pains of an open source community. I don't expect I'm gonna come up with the answer, but I'd love to learn more about what different folks are working on and how they're thinking about their roadmaps. I'll share what I learn along the way, and hopefully someone smarter than me can turn it into a proposal for us to rally around.

Kydo

32,756 Aufrufe • vor 25 Tagen