Muse Glimmer, A 30B parameter dense model swallowing a... 130,000 token context window using only 19.3 GB of VRAM (extreme efficiency). No KV cache quantization required. I just benched the new Muse Glimmer 30B (dense) on a single RTX 4090. We are pulling 3,100+ t/s prefill and 75 tokens/second decode. The throughput is violent. Meta superintelligence lab just open sourced this agentic beast, explicitly engineered to dominate 24GB consumer cards. I pulled the latest llama.cpp source on Ubuntu 22 (CUDA 13) to see if the specs were real. Fed it a 28k token prompt. Here is the exact llama.cpp God Stack and benchmarking breakdown: # 1. The Deep Context Run (No Speculative Decoding) The architecture uses a massive 16:1 GQA (Grouped Query Attention) ratio. This means the KV cache footprint is practically non existent. ./build/bin/llama-server -m Muse-Glimmer-30B-UD-Q4_K_XL.gguf -c 130000 -b 4096 -ub 4096 -ngl 99 --port 8080 Prefill: 3134.95 t/s Decode: 50.00 t/s VRAM: 19.34 GB (I hit 130k context on pristine, unquantized f16 cache and still had 4.5 GB of VRAM left over. Absolute witchcraft). # 2. The DFlash Speculative Overdrive Meta shipped this with a DFlash block diffusion drafter. Let's trade that extra VRAM for pure speed. ./build/bin/llama-server -m Muse-Glimmer-30B-UD-Q4_K_XL.gguf -md dflash-kquant.gguf --spec-type draft-dflash --spec-draft-n-max 3 -c 80000 -b 4096 -ub 4096 -ngl 99 --port 8080 Prefill: 1293.69 t/s Decode: 75.00 t/s VRAM: 23.93 GB (Maxed out on card) the dflash gguf is additional 1.6 GBs # The Architecture Insight (Muse Glimmer vs. Gemma 4 31B) If you look at my Gemma 4 31B tests from last week, getting 140k context required heavily degrading the memory with Q4 KV quantization (gemma 31b q4 can do only about 40k context with unquantized kv on a 24gb card). That "unzipping" overhead bottlenecked Gemma's MTP decode speeds down to 65 t/s. Muse Glimmer completely sidesteps this bottleneck. By using aggressive 16:1 GQA, it keeps the KV cache in native f16 format at massive context lengths. Flash Attention gets to run at maximum uncompressed speed, letting the DFlash drafter push decode safely to 75 t/s without compute lag. With a 76% on SWE Bench Verified and seamless local tool calling, this model looks promising. Unsloth's Hugging Face GGUF links, intelligence/agentic benchmark details, and inference throughput performance graphs are posted in the replies. For 24GB rig, what’s your current go to model?show more

Qwen 3.8 27B (dense) running on a single RTX... 4090 (24GB VRAM) at 65 tokens/sec decode with MTP! 260,000 context window or 65 tokens/sec decode with native MTP. The API cartel should be terrified. We are officially running frontier tier agentic AI (benchmarks comparable to claude opus 4.6 max) on a single consumer gaming GPU. I benchmarked Qwen3.8-27B on a single NVIDIA RTX 4090 (24GB VRAM, Ubuntu 22) using Unsloth’s Dynamic Q4_K_XL GGUF on the latest llama.cpp. Here is the complete benchmark breakdown across both Context Scaling and MTP Overdrive (28k prompt baseline): ### PART 1: The Context Scaling Matrix (Pure Throughput) # 1. Standard FP16 KV Cache (Unquantized): - 80k Context: 2,664.7 t/s prefill | 40.68 t/s decode | 22.36 GB VRAM - 100k Context: 2,678.7 t/s prefill | 40.89 t/s decode | 23.59 GB VRAM (100k is the hard ceiling for unquantized f16 KV in 24GB VRAM) # 2. Q8 Quantized KV Cache (-ctv q8_0 -ctk q8_0): - 130k Context: 2,639.1 t/s prefill | 40.96 t/s decode | 22.18 GB VRAM - 170k Context: 2,653.9 t/s prefill | 40.70 t/s decode | 23.68 GB VRAM (170k is the sweet spot for heavy agentic coding workflows) # 3. Q4 Quantized KV Cache (-ctv q4_0 -ctk q4_0): - 260k Context: 2,659.8 t/s prefill | 40.70 t/s decode | 23.00 GB VRAM Full 262k native context residing entirely in 24GB VRAM. Zero system RAM offload. Stress test with a monster 142k real-world prompt (-c 170000, Q8 KV): - Prefill: 1,829.50 tokens/s - Decode: 31.3 tokens/s - VRAM: 23.7 GB rock solid ### PART 2: Native MTP Overdrive (Trading Context for Speed) Since MTP heads are baked into the architecture, enabling native speculative drafting pushes decode speeds straight to 60 t/s with zero external draft model: # 1. MTP + Q8 KV Cache: - 80k Context: 2,370.66 t/s prefill | 59.25 t/s decode | 23.4 GB VRAM (MTP state buffers eat slightly more memory, making 80k the ceiling for Q8) # 2. MTP + Q4 KV Cache: - 130k Context: 2,391.09 t/s prefill | 60.10 t/s decode | 23.5 GB VRAM (Sweet spot: 130,000 context running at a screaming 60 tps decode) ### Qwen3.8-27B vs Muse Glimmer 30B Two days ago I benched Meta's Muse Glimmer 30B hitting 130k context unquantized (19.3 GB VRAM) pulling 50-75 t/s decode. If you own a single RTX 3090 or RTX 4090, you have zero excuse to burn API credits. ### The Reproduction llama.cpp flags: 1. Max Context Stack (260,000 Context @ 41 tps): ./build/bin/llama-server -m Qwen3.8-27B-UD-Q4_K_XL.gguf -c 260000 -ngl 99 --port 8080 -ctv q4_0 -ctk q4_0 2. MTP Overdrive Stack (130,000 Context @ 60 tps): ./build/bin/llama-server -m Qwen3.8-27B-UD-Q4_K_XL.gguf -c 130000 -ngl 99 --port 8080 -ctv q4_0 -ctk q4_0 --spec-type draft-mtp --spec-draft-n-max 4 --spec-draft-p-min 0.7 Unsloth's Hugging Face GGUF links, intelligence/agentic benchmark details, and performance charts are posted in the replies. Local compute is eating the cloud alive. How much monthly API spend does a 24GB setup like this actually replace for you?show more

50% more context unlocked for Qwen 3.8 27b Q4_K_XL... dflash 2 on a single RTX 4090 (24 GB VRAM) I found a hidden VRAM tax in llama.cpp. By combining my custom 2 bit DFlash 2 drafter with one overlooked server flag, I just unlocked another +80,000 tokens of context. Qwen3.8-27B is now running a massive 250,000 context at 75 tokens/s on a single RTX 4090. Here is the secret: By default, `llama-server` reserves massive chunks of your VRAM to handle multiple concurrent users (batching). If you are running a single user session, you are bleeding memory for features you aren't using. By passing the `--parallel 1` flag, you force the engine to dedicate 100% of your 24GB VRAM buffer to a single user. When we combine the VRAM saved by our Q2_K 2-bit drafter with the VRAM saved by `--parallel 1`, the context ceilings absolutely explode: Note: all benchmarks carried out with a massive 28k prompt. Ubuntu 22. ### THE NEW 24GB PHYSICAL LIMITS (Single RTX 4090): # 1. The "Repo Swallower" (Q4 KV Cache): - Context: 250,000 tokens (Up from 170k!) - Speed: 73.66 t/s decode | 1,608 t/s prefill - Peak VRAM: 23.8 GB # 2. The "High-Precision SWE" (Q8 KV Cache): - Context: 150,000 tokens (Up from 100k!) - Speed: 75.01 t/s decode | 1,667 t/s prefill - Peak VRAM: 23.9 GB # 3. The "Pristine Attention" (Unquantized FP16 KV): - Context: 90,000 tokens - Speed: 80.58 t/s decode | 1,699 t/s prefill - Peak VRAM: 23.92 GB ### HOW TO RUN THE 250K GOD STACK TODAY: (Requires PR #27342 + my Q2_K Hugging Face drafter) llama.cpp flags: ./build/bin/llama-server -m Qwen3.8-27B-UD-Q4_K_XL.gguf -md Qwen3.8-27B-DFlash2-Q2_K.gguf --spec-type draft-dflash --spec-draft-n-max 3 -c 250000 -ngl 99 --parallel 1 --port 8080 -ctv q4_0 -ctk q4_0 We are pushing a quarter million tokens of context with speculative DFlash 2 decoding at 73 tokens/second on a single consumer gaming GPU. I dropped my custom 2 bit Hugging Face GGUF links, visual performance graphs, and the PR #27342 build instructions in the replies below. If you own a single RTX 3090 or 4090, it is officially time to cancel your API subscriptions and let local silicon eat the cloud. how much monthly API spend does an optimized 4090 rig like this actually replace for you?show more

Qwen 3.8 27B Q4_K_M - 90 tokens/sec on a... single NVIDIA RTX 4090 (24 GB VRAM) with Dflash2! (MTP 60 tps -> 90 tps Dflash2!!!!) Local AI moves so fast (literally!) it’s terrifying. Z lab just dropped DFlash 2 for Qwen 3.8 27b and Muse Glimmer. I patched llama.cpp (PR #27342) and paired it with Unsloth’s Qwen 3.8 27B UD-Q4_K_XL quant. The result? Lossless 90 tokens/s decode. My last post highlighted native MTP hitting 60 t/s at 130,000 context. But DFlash 2 just completely shattered that ceiling. By using parallel block diffusion drafting (predicting whole blocks of tokens in a single pass using dynamic convolutions), DFlash achieves a massive 5.39 token acceptance rate. THE ALPHA TWEAK: `n-max 7` eats too much VRAM for draft states. But if you drop the draft limit to `--spec-draft-n-max 4`, you slash the VRAM overhead and actually increase the throughput. Here is the new 24GB VRAM Physics Matrix (DFlash 2 @ n-max 4): - 30k Context: 1,725 t/s prefill | 87.05 t/s decode | 22.2 GB VRAM - 80k Context: 1,789 t/s prefill | 84.20 t/s decode | 23.3 GB VRAM - 110k Context: 1,767 t/s prefill | 83.35 t/s decode | 23.96 GB VRAM (110k context at 83+ tokens a second sitting exactly on the 24GB hardware limit is absolute wizardry). How to compile the PR today: git clone cd llama.cpp git fetch origin pull/27342/head:pr-27342 git switch pr-27342 cmake -B build -DGGML_CUDA=ON && cmake --build build -j Llama.cpp flags for Dflash (110k Context Ceiling): ./build/bin/llama-server -m Qwen3.8-27B-UD-Q4_K_XL.gguf -md Qwen3.8-27B-DFlash2-Q4_K_M.gguf --spec-type draft-dflash --spec-draft-n-max 4 -c 110000 -ngl 99 --port 8080 -ctv q4_0 -ctk q4_0 The fact that the open source community is shipping block diffusion drafters so quickly that run entirely locally on a gaming GPU is unbelievable. If you own a single RTX 3090 or 4090, it is officially time to upgrade to qwen 3.8 27b with dflash 2 and cancel your API subscriptions and let local silicon eat the cloud. This model beats GPT 5.6 Terra, GLM 5.2 DeepSeek V4 Pro, Muse Spark 1.2 and Claude Opus 4.8 on the artificial analysis agentic index (details in the replies) Hugging Face GGUF links (Base + DFlash2) and the full visual VRAM scaling and Dflash2 vs MTP graphs are also in the replies below. are you sticking to native MTP for the 130k context, or sacrificing 20k context to redline your decode speed? How many tokens/sec are you pushing on your current local rig?show more

If you thought the Gemma 4 31B (dense) model... was fast, sit down. I just benched the updated Gemma 4 26B A4B MoE on a single RTX 4090 (24 GB VRAM) 9,200 t/s prefill. 160 t/s decode. 250,000 context window. All on a single consumer RTX 4090. The numbers are completely unhinged. The 31B is a dense behemoth. But the 26B is a Mixture of Experts (MoE), specifically an Active 4 Billion (A4B). It holds 26B parameters of knowledge but only activates 4B per token. Because its inference memory footprint is so light, I didn’t even need KV cache quantization to hit a quarter million context. Compiled the latest llama.cpp from source on Ubuntu 22 (CUDA 13). Fed it a 28k token prompt, and manually cranked the batch sizes (-b 2048 -ub 2048) to absolutely redline the Tensor Cores. Here is the benchmarking breakdown: # 1. The Baseline (No MTP) Even without speculative decoding, the A4B architecture flies. llama.cpp flags: ./build/bin/llama-server -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf -c 250000 -ngl 99 -fa on -b 2048 -ub 2048 --port 8080 -v Context Ceiling: 250,000 tokens (21.5 GB VRAM) Prefill: 9,200 t/s (Absurd) Decode: 124 t/s # 2. The MTP Overdrive Injected the new MTP draft model to enable Speculative Decoding. llama.cpp flags: ./build/bin/llama-server -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf --spec-type draft-mtp --spec-draft-model mtp-gemma-4-26B-A4B-it.gguf --spec-draft-n-max 4 --spec-draft-p-min 0.7 -c 250000 -ngl 99 -fa on -b 2048 -ub 2048 --port 8080 -v Context Ceiling: 250,000 tokens (22.96 GB VRAM) Prefill: 7,054 t/s (MTP draft overhead slightly caps prefill) Decode: 156 t/s # The Agentic Architecture Insight Why does this matter? Because you can now build a killer local agentic loop on a consumer desktop. Use the 31B dense model (from the previous post) as your heavy, deliberate Orchestrator / Verifier / Planner. Pass the actual execution tasks to this 26B MoE. At 160 t/s, this MoE can chew through code generation, tool calling, and massive RAG document retrieval over a 250k context window almost instantly, drastically speeding up your agentic loop. If you own a single RTX 3090 or 4090 and haven't tried this specific stack yet, you need to pull these latest updates and run it. Local inference just leveled up. Hugging Face links to the Unsloth 26B QAT quants and MTP drafters are in the replies. performance graphs also available in the replies.show more

The "I don't have enough VRAM" excuse just died.... I’m running Meta’s new 30B Muse Glimmer Q6_K_XL with a massive 130k context window on just 26GB VRAM FREE compute on Kaggle. Kaggle provides you free 2x Nvidia T4 GPUs. 30 hours usage each week! Yesterday, I showed you the violent throughput of Muse Glimmer on a single RTX 4090. Today, we are securing a Dual NVIDIA T4 GPU cluster with 32GB of total VRAM for exactly $0 and dropping the massive 24.5GB Q6_K_XL GGUF onto it. Here is the exact Kaggle workflow and benchmarking breakdown: # 1. The Storage Bypass & Setup I built a clean cell by cell script in the file. We dynamically fetch the CUDA accelerated llama.cpp binaries and use wget to stream the model directly into Kaggle's /kaggle/tmp scratch storage, which cleanly bypasses their 19.5GB output directory limit. # 2. The Multi GPU Performance With the -ngl 99 flag offloading all model layers across both T4 GPUs (32GB VRAM combined), we pushed a massive 131,072 token context window (-c 131072). The benchmark numbers: Prefill: 265.9 t/s Decode: 9.0 t/s VRAM Total: 26.5 GB # 3. The Architecture Insight The Q6_K_XL model itself is 24.5 GB. Because of Muse Glimmer's aggressive 16:1 GQA, the unquantized KV cache for a massive 130k context window only takes up 2 GB of memory. No heavily degraded Q4 KV quantization required. It just works. No compiling from source. No credit card. No OOM crashes. Zero excuses. If you’re running a single RTX 3090, 4090, or 5090, you need to experience this hyper efficient KV cache right now before the upcoming Qwen 3.8 27B drop completely steals your VRAM tomorrow. pick the Q4 or Q5 quants for 24 GB VRAM rigs. I'm dropping the Unsloth huggingface GGUF links and the free Kaggle notebook link in the replies. spin up your own instance, and show me your multi GPU benchmarks.show more

Gemma 4 26B A4B MoE - 500+ t/s decode... - Single RTX 4090 (24 GB VRAM) - Llama.cpp concurrency 24 - q8 kv cache How many API users can you simultaneously host on a single RTX 4090 (24 GB VRAM) before it crashes? Yesterday, I proved you can host 14 active users using unquantized memory. Today, I used 8 bit KV Cache Quantization to hack the VRAM footprint. I successfully scaled to 24 concurrent users without a single dropped connection. A 71% server capacity boost for free. By adding the -ctk q8_0 -ctv q8_0 flags to llama.cpp, you compress the KV cache context memory from 16 bit to 8 bit. This unlocks massive concurrency limits on Gemma 4 26B (MoE) on a single 24GB consumer GPU. Here is the exact telemetry from pushing 8 bit quantization to its absolute physical edge: # TEST 1: The 24 User Concurrency Max Server Config: 24 slots (np 24) | 4,096 context per slot | 98,304 Total Context Client Load: 24 simultaneous requests (2,000 token prompt per user) Unquantized KV cache for this load requires 28GB+ VRAM (Instant OOM). Quantized to Q8, it allocated safely at 23.35 GB. The C++ engine crunched the entire batch in 28.5 seconds. Decode Speed: 21 t/s (Per User) | 500 t/s (Agg) # TEST 2: The 48 User Queue Overload What happens to a compressed cache during a traffic spike? Server Config: 24 slots (np 24) | 4,096 context per slot | 98,304 Total Context Client Load: 48 simultaneous requests (2k token prompt per user) Zero queue drops. The scheduler flushed and hot swapped the 8 bit memory flawlessly on the fly, completing all 48 users in 66.0 seconds (a perfect 2.3x queue scaling multiplier). Decode Speed: 18 t/s (Per User) | 430 t/s (Agg) # TEST 3: The 8 User RAG Slam Server Config: 8 slots (np 8) | 60,000 context per slot | 480,000 Total Context Client Load: 8 simultaneous requests (30k token prompt per user) It allocated 23.83 GB VRAM and chewed through ~240,000 prefill tokens in 46 seconds under massive memory pressure. Prefill Speed: 6,200 t/s (Agg) Decode Speed: 22 t/s (Per User) | 175 t/s (Agg) # The Engineering Alpha (The Quantization Tradeoff): You gain a massive 71% increase in server capacity, but what do you lose? Compute latency. Because the cache is stored in 8 bit, the GPU's cores have to dequantize the memory back to 16 bit on the fly during every single prefill step. In my unquantized tests yesterday, single slot prefill was hitting ~1,500+ t/s. Today, under the heavy 48-user Q8 load, prefill dropped as low as ~750 t/s. You trade a few seconds of initial prefill latency to essentially double your API hosting capacity. For production high volume SaaS, this is the ultimate unit economics cheat code. Here is the exact command to run a 24 user Q8 continuous batching server on your own single 4090, single 3090 or any 24gb vram rig: ./build/bin/llama-server -m gemma-4-26B-A4B-it.gguf -c 98304 -np 24 -b 2048 -ub 2048 -ngl 99 -fa on -ctk q8_0 -ctv q8_0 --port 8080 (Note: -c 98304 allocates exactly 4,096 tokens of context per user across 24 slots). Hugging Face links to the Unsloth Gemma 4 26B QAT quants along with performance graphs available in the replies. Would you trade 3 seconds of Time To First Token latency to double your active user capacity?show more

Gemma 4 12B QAT (dense) achieves 1000+ tokens/sec prefill... on 8GB VRAM with 120k context Gemma 4 12B QAT (dense), TurboQuant (Without MTP), RTX 4060 8GB VRAM: Prefill: 1000+ tok/s (42% increase) Decode: 25+ tok/s (25% increase) Context: 120k (150% increase) prefill was 700 tok/sec and decode 20 tok/sec with only 48k context without turbo quant (older test with mtp link in the comments) llama.cpp TurboQuant flags: -m gemma-4-12B-it-qat-UD-Q4_K_XL.gguf -c 120000 --cache-type-k q8_0 --cache-type-v turbo3 -ngl 99 --port 8080 tested with a 27k prompt, 120k context loaded. -ngl 99 here isn't a typo, full 12B dense, every layer on GPU, on an 8GB card. that's the part worth sitting with. The model has vision, audio input, thinking/reasoning and fits your 8GB card. TurboQuant's KV cache savings are what free up the room to do that at 120k context. side by side with yesterday: 26B A4B MoE got 320+ tok/s prefill. this dense 12B is clearing 1000+ rig: RTX 4060 8GB · i7H · 16GB RAM same two flags as yesterday, different model size: --cache-type-k q8_0 --cache-type-v turbo3 thanks to TheTom/llama-cpp-turboquant, TurboQuant fork of llama.cpp by Tom Turney (Tom Turney) to make this work. unsloth's model quant huggingface and the llama.cpp fork github link in the comments Do you prefer a dense or a MoE for your 8GB card?show more

Open source AI is actually moving at an unhinged... pace right now. I literally hadn't even finished typing up my last Gemma 4 12b benchmark notes before Google went ahead and dropped the official Quantization Aware Training (QAT) checkpoints on Hugging Face. If you missed the news, QAT basically bakes the compression directly into the training process. Instead of standard post training quantization degrading the model's reasoning capabilities, QAT trains the model with compression in mind. Unsloth is reporting near original performance at 4-bit with ~72% lower memory footprint. Details in the comments. Naturally, had to instantly pull the new GGUFs to see what a single RTX 4090 card (24 GB VRAM, Cuda 12.8, ubuntu 22) could do. i fired up llama.cpp engine again Look at these numbers: 1. Unsloth Gemma 4 26B-A4B IT (QAT Q4_K_XL) flags: ./build/bin/llama-cli -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf -cnv -ngl 99 -c 250000 -fa on -v VRAM Used: 19.5 GB context: 250,000 tokens decode throughput: 193 tps 2. Unsloth Gemma 4 31B IT (QAT Q4_K_XL) flags: Command: ./build/bin/llama-cli -m gemma-4-31B-it-qat-UD-Q4_K_XL.gguf -cnv -ngl 99 -c 60000 -fa on -v - VRAM Used: 23 GB (Tight, but zero system RAM spillover) - context: 60,000 tokens - decode throughput: 47 tps We are essentially watching hardware bottlenecks evaporate in real time. An update literally drops before you can finish benchmarking the previous one. What a time to be running local hardware. If you have a single rtx 3090, rtx 4090, these are the latest gemma models to try this week.show more

the 24gb vram tier is enough for most builder... work in 2026. gemma 4 31b dense on my rog scar 18 just autonomously built a production hero section in one prompt, one html file and 5 minutes end to end. hardware: rog scar 18, rtx 5090 laptop 24gb vram. model: google gemma 4 31b dense at q4_k_m quant, using 22.8 of 24gb. engine: llama.cpp built for blackwell (sm_120). harness: hermes agent with native tool parsing. speed: 15 tok/s sustained, 94 watts, 50c. flags i used: ./build/bin/llama-server -m ~/models/gemma4-31b/google_gemma-4-31B-it-Q4_K_M.gguf -ngl 99 -c 131072 -np 1 -fa on --cache-type-k q4_0 --cache-type-v q4_0 --jinja --host 127.0.0.1 --port 8080 if you own 24gb vram in 2026, you have enough for most ui work, most agentic coding, most autonomous builds. no subscription, no one logging your prompts. a dense open model on consumer hardware shipping real software on your desk. this was the warmup. full page next on same hardware, then the octopus invaders final multifile autonomous challenge.show more

I told you to claim your free 16GB NVIDIA... GPU for learning Local LLMs. Now I’m going to show you how to double its inference speed without touching the hardware. Google Colab gives you an enterprise grade NVIDIA Tesla T4 GPU for free, roughly 4 hours every single day. It is the absolute perfect sandbox for learning AI engineering, testing inference flags, and pushing massive context windows. The local AI timeline is moving way too fast. If you aren't using Multi Token Prediction (MTP) yet, you are leaving massive performance on the table. I just pushed DeepMind’s Gemma 4 26B to 64.9 t/s on this exact free tier. Let's look at the raw benchmark data running on an Ubuntu Linux environment with the latest compiled llama.cpp binaries and quantized GGUFs from Unsloth via HuggingFace: # Qwen 3.5 9B (Dense): Base: [ Prompt: 626.7 t/s | Generation: 21.0 t/s ] With MTP: [ Prompt: 539.1 t/s | Generation: 24.8 t/s ] # Gemma 4 26B QAT (MoE): Base: [ Prompt: 634.2 t/s | Generation: 48.3 t/s ] With MTP: [ Prompt: 572.1 t/s | Generation: 64.9 t/s ] If you are paying attention, this single Colab notebook reveals 3 massive observations about the current state of local LLMs: # 1. The MTP Speedup (Software Overclocking) Standard autoregressive decoding guesses one token at a time. MTP acts like a highly optimized, built in speculative decoder. It predicts multiple future tokens at once and the main model verifies them in parallel. The result? Zero accuracy loss and a massive throughput increase. Gemma jumped from 48 to 65 t/s just by flipping a flag. # 2. The MoE Paradox (Bigger is Faster) How does a 26B parameter model absolutely destroy a 9B model in raw speed on the exact same hardware? Architecture. Qwen 3.5 9B is a dense model. it activates all 9 billion parameters for every single token. Gemma 4 26B is a Mixture of Experts (MoE) model. It routes data efficiently, activating only 4B parameters per token. You get the reasoning capabilities of a 26B model with the compute cost of a 4B model. 3. Thinking Efficiency When I ran the exact same complex prompt on both models, the larger MoE spent significantly fewer "thinking" tokens to arrive at the correct answer. A smarter model doesn't just give better answers; it gets to the point faster, saving you compute cycles and preserving your context window. # Want to run this yourself? Here are the exact llama.cpp CLI commands. For Qwen (MTP is baked into the main model): ./llama-cli -m Qwen3.5-9B-UD-Q4_K_XL.gguf -p "Explain quantum computing." -n 2000 -c 8000 -ngl 99 -fa on --spec-type draft-mtp --spec-draft-n-max 4 --spec-draft-p-min 0.7 For Gemma (Using a separate lightweight draft model): ./llama-cli -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf --model-draft mtp-gemma-4-26B-A4B-it.gguf -p "Explain quantum computing." -n 2000 -c 8000 -ngl 99 -fa on --spec-type draft-mtp --spec-draft-n-max 4 --spec-draft-p-min 0.7 Stop waiting for a $3,000 rig. Boot up Colab, pull these models, and start building your stack. I’ve put together a completely free, cell by cell Google Colab notebook that automates this entire workflow so you can test it yourself in 5 minutes and learn. Link to the notebook is in the comments below. Experiemt with different MTP parameters, context windows and post your results in the comments.show more

you're paying $20/mo for something your $500 GPU can... already do. Gemma 4 26B A4B QAT MoE + Hermes Agent running on a single RTX 4060 (8GB VRAM). Built a vision capable, 100% free, 100% local, private AI assistant that lives in my Chrome browser. No API keys. No cloud. No subscriptions. 100% vibe coded. 0% handholding. It has full context of whatever's on my screen can answer questions, summarize pages, extract data, and see images. Same local model handles everything, no external calls, ever. keep reading for the model and hermes agent tips i learnt while building this locally. Here's the exact setup for anyone running local LLMs on 6-8 GB VRAM: llama.cpp server flags (on my NVIDIA RTX 4060 8gb VRAM): -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf --cache-type-k q8_0 --cache-type-v q8_0 -c 150000 --port 8080 Throughput with quantization: Prefill: 200-250 tokens/sec Decode: 20-25 tokens/sec reduce context if oom on 6 gb vram card. Key learnings: - Quantize KV cache to q8 for faster prefill/decode. Prefill goes from 100-150 (unquantized) to 200-250 tok/s (q8). - But watch out, once actual context grows past ~50k tokens on high entropy workloads, q8 KV quantization can cause hallucinations. Low entropy workloads are mostly unaffected. If you see it happening, drop the quantization. This is common across all local models. - In Hermes Agent settings -> Memory & Context, bump compression threshold from default 0.5 to 0.7. Default triggers way too frequent context compression and eats time. Up next: add persistent memory, web search, tool calling, streaming output and whatever you suggest. Running a 26B MoE with vision + 150k context window on 8GB VRAM would've sounded impossible 6 months ago. Works the same on the NVIDIA RTX 3060 Ti, 3070, 4060 Ti, 5060, 2080, or any 8GB card. VRAM is the only requirement. Local AI agents are closer than people think. You just need to know where the knobs are. Model's Unsloth quant hugging face link in the comments. Have you tried Hermes agent by Nous Research yet? What are you building with local LLMs? Drop it below, let's see what this community is shipping.show more

Run Gemma 4 26b MTP on 8 GB VRAM... GPUs at 25+ tokens/second. Flags included! local llm space is moving at terminal velocity. only 3 days ago google released gemma 4 26b a4b qat quants. more efficient than before, ran on 8gb vram at 20 tok/sec. and now just a few hours ago, mainline llama.cpp merged a massive update and we just shattered our own record. decode throughput went 25-40% up on the same 8 GB VRAM setup! Before MTP: 20 tps -> After MTP: 28 tps! llama.cpp just officially merged PR #23398 ("add Gemma4 MTP"), bringing native Multi-Token Prediction (MTP) support to Gemma 4 models. By running speculative drafting on the same 8GB VRAM RTX 4060 setup, my decode throughput on a 64k context instantly leaped to a blistering 25–27 tokens/sec thats 25-30% increase with the same hardware. Here is the architectural catch you need to know: Unlike the Qwen 3.5 and 3.6 series, which bake the MTP heads directly into the base GGUF, the Gemma 4 MTP head is not built in. You must download a separate, specialized MTP drafter GGUF (the assistant model) to act as the speculator. (I've dropped the download link in the replies). copy and try the exact flags: -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf --spec-type draft-mtp --spec-draft-n-max 6 --spec-draft-p-min 0.7 --spec-draft-model gemma-4-26b-A4B-it-assistant-Q4_0.gguf -c 64000 -v n-max 4 and p-min 0.7 is also worth checking out. benchmark on your setup and workflow. if you have a single 8 gb vram nvidia rtx 4060, 3060, 3070, 2080, 2070, grab the MTP drafter GGUF link in the comments and try it yourself. Check it out even if you have asmaller or a larger gpu, such as a single rtx 3090, 4090, 3060, 2060. MTP works for all gemma 4 sizes such as gemma 4 12b, gemma 4 31b etc. but remember to grab the correct mtp draft assistant models respectively. what are you benchmarking todayshow more

Run Gemma 4 26B MoE on 8GB VRAM with... 250k context at 20+ tokens/sec If you own any 8GB VRAM graphics card, stop what you are doing. Local AI just had its absolute "Holy Shit" moment for budget hardware. Yesterday, I benchmarked Unsloth Gemma 4 12B Q4_K_XL on an 8GB card. The community went wild but immediately demanded more: "Can we run a 25B+ model on budget GPUs?" Today, I’m delivering exactly that. I am running a massive 26B parameter Mixture of Experts (MoE) model locally on a standard 8GB VRAM setup with 250k full native context!. If you own an RTX 3060, 3070, 4060, or any budget GPU with 8GB of VRAM, the local AI paradigm has completely changed. The performance metrics are astonishing: - 20 tokens/sec flat decode throughput. - Stable, flat decode speed even with massive prompts. - I threw a 60k token prompt at it, and it still clocked in at 20 TPS without dropping a single frame. # What about prefill? Yes, Time To First Token (TTFT) is slightly high when swallowing massive contexts. But with a solid 200 tokens/sec prefill speed, the wait is barely noticeable and highly usable. And this is running completely without Multi Token Prediction (MTP) active. How is this possible? It’s the magic of Google's new QAT (Quantization Aware Training) quants for Gemma 4. The model weight file (unsloth gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf) is only 13.2 GB, making it the ultimate local powerhouse. # The Test Setup: CPU: Intel Core i7 RAM: 16GB System RAM GPU: NVIDIA GeForce RTX 4060 Laptop GPU (8GB VRAM) # The Secret Sauce (The -cmoe Flag) To make this work properly on any 8GB card, you must use the -cmoe (CPU MoE) flag in llama.cpp. This flag isolates the heavy MoE expert weights directly to system memory (CPU/RAM) while letting your GPU focus strictly on the Attention layers and the KV Cache. It prevents VRAM spillage and holds the throughput rock solid. # The flags: -m "gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf" -cmoe -c 248000 -v Once running, just open the UI on localhost and toggle the new reasoning lightbulb icon in the text input box to watch the model perform multi step thinking. Are you still running smaller models, or are you ready to scale up your budget local setups? Let's discuss in the repliesshow more

my 8 GB VRAM gaming laptop is absolutely going... to hate me for this. but I still did it. ran a 31b dense model (Gemma 4 31b Q4) with only 8 GB VRAM last week I ran Gemma 4 26B A4B a mixture of experts model on my RTX 4060 and hit 25–28 tokens/sec using llama.cpp's new MTP support. smooth. snappy. but MoE has a secret: it only activates 4B parameters per token despite having 26B total. that's why it flies. so the real question started haunting me. what if I throw a full, no tricks, every parameter fires on every token, 31B DENSE model at the same machine? # Hardware: GPU: NVIDIA RTX 4060, 8 GB VRAM RAM: 16 GB CPU: Intel Core i7 H Laptop. Gaming. Modest. The model: gemma-4-31B-it-qat-UD-Q4_K_XL.gguf (model's unsloth huggingface link in the comments) This is Google DeepMind's flagship dense model in the Gemma 4 family that can run on single consumer GPU. It packs a hybrid attention architecture, supports up to 256K context natively, and is QAT (Quantization Aware Training) optimized, meaning it retains far more quality than standard post training quants at the same bit depth. This is NOT the MoE. This is 31 BILLION dense parameters, every single one of them loaded. # the flags I used: -m gemma-4-31B-it-qat-UD-Q4_K_XL.gguf -cnv --spec-type draft-mtp --spec-draft-model mtp-gemma-4-31B-it.gguf --spec-draft-n-max 8 --spec-draft-p-min 0.6 -c 6000 -v Multi Token Prediction (MTP) is still active here. Separate draft GGUF required, same as the 26B setup. # Results: → Decode: ~3 tokens/sec → Prefill: ~2 tokens/sec → Context: 6000 tokens → Hardware crying quietly in the corner: yes so is 3 tps actually usable? For real time back and forth chat? Not ideal. You're not having a fluid conversation at 3 tps. but slow ≠ useless. And this is where it gets genuinely interesting. think about how senior devs actually work in a real team. But when something is architectural, deeply complex, or needs serious reasoning? they walk down the hall and escalate to the senior. That's exactly the local AI agent architecture this unlocks: → Fast orchestrator model (Gemma 4 26B MoE at 25+ tps) handles routing, simple queries, tool calls, memory. The junior dev. → Gemma 4 31B dense is the senior, called only when the fast model genuinely hits a wall. Hard multi step reasoning. Complex code generation. Deep architectural decisions. The agentic loop stays fast. Only the hard hops touch the 31B. That's a legitimate production grade local AI architecture on a budget hardware. (requires 2 8gb gpus) other workflows where 3 tps is completely fine: - overnight batch jobs. summarize documents, extract structured data, review code. Fire it off. Sleep. wake up to results. - One shot deep reasoning - Silent code audit loops, you write and test, the 31B reviews diffs and flags issues in the background between your sprints - Any workflow where output quality > output speed A few weeks ago, nobody was running a 30B+ dense model on a single consumer GPU with 8 GB VRAM. At all. Now we're doing it on an Intel i7-H gaming laptop with a NVIDIA RTX 4060, thanks to llama.cpp + QAT quants + MTP speculative drafting. Google DeepMind said the Gemma 4 31B targets "consumer GPUs and workstations." They were not exaggerating. The hardware bar to run serious frontier class models locally keeps dropping. the tools are here. the models are here. you just have to be willing to abuse your laptop a little. what workflows would you actually run on a local 3 tps 31B dense model? genuinely curious. drop it below.show more

90% of "AI developers" just download pre packaged GGUF... files from Hugging Face, hit run, and call it a day. The top 10% know how to pull the raw safetensors, run the math, and quantize massive models into Q4_K_M themselves. If you think llama.cpp can only execute models, you’re missing the best part of the open source ecosystem. It’s a high performance optimization suite. Manually stripping 69% of the VRAM footprint off a brand new model architecture is where real infrastructure value is made. If you want to actually master local inference and deploy models like Google’s massive Gemma 4 12B it on consumer NVIDIA hardware using llama.cpp, you need to learn this pipeline. Let's build it. I just took the raw 22.7 GB Gemma 4 baseline and manually compressed it down to a 7.02 GB Q4_K_M GGUF artifact using llama.cpp. That is a 69% reduction in footprint. No quality loss. No VRAM bottlenecks. Just native, hardware accelerated C++ inference running a full 2,50,000 token context window on a dual NVIDIA Tesla T4 setup. Stop melting your VRAM on unoptimized weights and stop relying on other people's pipelines. Own your stack. I mapped this entire architecture from dynamic binary fetching to raw quantization and real time GPU streaming into a single, bulletproof notebook. Notebook link is in the comments below. Bookmark this blueprint for your next deployment and tell me which quantization works best for your workflow and model.show more

If you have an RTX 3090 or 4090, Mia... just shipped you a free massive upgrade in both speed and intelligence. I will explain to you why this will make your Qwen 3.8 27B on your card, even better, and my flags for running it. Qwen3.8-27B, EXL3 3.5bpw, DFlash2 speculative decode, RTX 4090. Single stream. The kit is from MiaAI-Lab, EXL3 is turboderp's format. I re-measured everything on my own card because the my first benchmarks seemed off. It turns out it really does run much faster. WHY EXL3 IS A DIFFERENT ANIMAL The old way (Q4_K_M) rounds each weight to the nearest 4-bit value independently. Every weight introduces its own rounding error. Those errors accumulate across millions of weights and causes drift (Slightly dumber). EXL3 is a fundamentally different compression algorithm. Instead of rounding each weight on its own, it encodes the entire weight vector as a path through a constrained codebook and spreads the rounding error across dimensions using a Hadamard transform. The result is that at the same bits per weight, more of the original model's intelligence is preserved. The important part is this CAN ACTUALLY BE MEASURED. The cleanest way to see that is KL divergence against a high-precision teacher. Lower means the quantized model thinks more like the original. On the malaiwah independent teacher-logit panel for GLM-5.3-Flash: EXL3 4bpw: 0.0246 nats Official FP8: 0.0206 nats NVFP4: 0.0605 nats EXL3 sits 0.004 nats behind native FP8 at half the size. NVFP4 at higher bit width is 2.5x further from the teacher. That panel is GLM-5.3-Flash, not Qwen 3.8. Cited as the mechanism, not as this run's data. But the point stands: EXL3 is not just smaller, it is smarter per bit than the formats most people are running. WHAT I MEASURED I first measure 108 tok/s from a single run. After that number looked too good to be true. I reran it. It looks like after a warm up, the numbers are even better. Basically, like people long thought, the RTX 3090 and RTX 4090 are actually superb AI computer cards. Hence why NVIDIA stopped shipping them with NVLINK since the 4090. Short context ceiling (~2k in, 1016-token output, TTFT-separated): 135, 138, 153, 174, 133, 129 tok/s across 6 runs. Sustained longform (2040-token essay): 105.4, 94.5, 98.4 tok/s Short answer (504 tokens): 93.2 tok/s The honest shape: ~130-150 tok/s at short context is the ceiling, ~94-105 sustained on longform. The ceiling matters because that is what people feel in chat. The old dense Q4_K_M on llama.cpp ran ~37 tok/s on this same card. (No MTP), with MTP about 60 tok/s Sustained is roughly 2.5-3x. Ceiling is closer to 4x. Same model, different quantization and engine. The multiplier comes from EXL3, the ExLlamaV2 engine, and DFlash2 together. CONCURRENCY IS A RTX 4090 LANE. Just like the old config on the 4090, the 24gb vram, means a long context can only hold one stream, and running concurrency requires to lower context length, because it runs fast it sort of makes up for it by being faster than slower GPU chips. CONTEXT LADDER The recipe doc measured prefill. I re-ran it with TTFT separated from decode, because decode is what you actually feel after the first token. ~5k in: decode 140 tok/s (TTFT 0.5s), needle HIT ~18k in: decode 85 tok/s (TTFT 0.3s*), needle HIT ~73k in: decode 28 tok/s (TTFT 1.8s), needle HIT ~146k in: decode 16 tok/s (TTFT 2.3s), needle HIT (*0.3s at 18k is a prefix-cache hit from the paired pass. Cold prefill for reference: ~2,020 tok/s at 17k falling to ~508 at 153k.) Needle hit at every depth, mine and the original 7/7. Retrieval is intact at max context. Speed is not: decode falls ~9x from short to max. Past ~50k tokens this stops being a chat tool and becomes a batch tool. At 146k it works, but nobody is typing interactively against 16 tok/s. WHERE IT BROKE The model's native context is 262k. The README says DFlash2 fits ~220k on a 24GB card. My 200,704-token attempt failed with insufficient VRAM. Dropped to 168,960 and it booted. The real ceiling is somewhere between 168,960 and 200,704 and I never tested that gap. I jumped to a value that worked and called it done. That is ~32k tokens of context I left on the table. One thing the numbers taught me: JSON tokenizes at ~1.5 chars/token, prose at ~3.9. "150k tokens of JSON" needs ~2.7x more filler than the same estimate in prose. Size by real tokens, not estimates. THE UPGRADE If you own a 4090 and you are running Q4_K_M on llama.cpp, you are leaving a good bit of speed and measurable intelligence on the table. The same model, on the same card, with a better quantization and engine, goes from 60 tok/s to 130-150 at short context and 94-105 sustained. The model also thinks closer to the original because EXL3 preserves more of the output distribution per bit than the old rounding method. The recipe is in the first reply. Everything above came from one 4090 and one afternoon of re-measuring. The decode ladder especially needs independent numbers. If your card gets different falloff, that is worth knowing. Recipe and flags/ findings in reply 👇show more

Qwen3.8-Flash-Next now reaches ~43 tok/s after a 122,902-token prompt... on ONE DGX Spark. ⚡🚀 MTP k=2 won my draft-depth sweep, with +42.5% mean decode over no draft. The PLE table stays fully on-device. I promised the deeper MTP tests. Here are the results, and now you can explore them in an interactive benchmark page too. 𝗧𝗪𝗢 𝗗𝗥𝗔𝗙𝗧 𝗧𝗢𝗞𝗘𝗡𝗦 𝗪𝗢𝗡 Mean single-request decode with 32K context configured: MTP k=2: 39.21 tok/s MTP k=3: 36.42 tok/s MTP k=1: 35.18 tok/s No draft: 27.51 tok/s k=2 also produced the fastest individual sweep run: 41.34 tok/s. Four runs each for no draft, k=1 and k=2. Seven for k=3. Decode excludes time to first token. Here, k means speculative draft depth, not quantization bits. k=3 produced more tokens per step, but the extra drafting work did not pay off in throughput. k=2 is my current pick for this setup. 𝗧𝗛𝗘 𝟭𝟮𝟯𝗞-𝗧𝗢𝗞𝗘𝗡 𝗣𝗥𝗢𝗠𝗣𝗧 𝗧𝗘𝗦𝗧 I then ran a separate long-prompt comparison: Actual input: 122,902 tokens Configured context: 262,144 Requested output: 128 tokens One request at a time MTP k=2: ~43 tok/s No draft: 26.2 tok/s Time to first token: 110.6 seconds with MTP 107.0 seconds without it The win here is generation speed, not faster prefill. To keep the scope clear: 256K was the configured limit. This was a real ~123K input, not a completely filled 256K window or a full k sweep at that depth. 𝗣𝗟𝗘 𝗦𝗧𝗔𝗬𝗦 𝗢𝗡 𝗧𝗛𝗘 𝗦𝗣𝗔𝗥𝗞 Whole model on-device: 78.57 GiB Packed 5-bit PLE table: 30.4 GiB, included in that total No NVMe PLE offload in this build. This is still turboderp’s 3.05bpw_h5_ng5 EXL3 pack, served through my vllm-exl3 integration. My work here is the serving integration and testing. These are preliminary performance measurements, not a quality evaluation or a claim of bit-exact full-output parity. 𝗘𝗫𝗣𝗟𝗢𝗥𝗘 𝗧𝗛𝗘 𝗥𝗘𝗦𝗨𝗟𝗧𝗦 The benchmark page has the individual sweep values, long-prompt comparison, and measurement scope. You can play the animation, export the charts, or download the HTML and data to render them yourself. No Spark needed to view the results. Credit to turboderp / ExLlamaV3 for the pack and kernels, vLLM for the serving engine, and Qwen Qwen Developers for the model. Recipe + reproduction: Interactive benchmark:show more

I just got Gemma 4 26B A4B MoE model... running fully locally with Hermes agent on an 8GB RTX 4060 and it's now backtesting trading strategies end to end, no hand holding. If you’re a trader or work on Wall Street, you don’t want to miss this. Yes. fully automated. No cloud. No APIs beyond market data. # Here's what I did: Setup: - Model: Gemma 4 26B-A4B QAT (MoE), Q4_K_XL Unsloth's quant (link in the comments) - Inference: llama.cpp (turboquant fork by Tom Turney link in the comments) - Hardware: RTX 4060, 8GB VRAM + 16GB RAM only (with 50 other chrome tabs open) - Context: 64K llama.cpp turboquant flags: -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf -c 64000 --cache-type-k q8_0 --cache-type-v turbo3 --port 8080 turboquant helps achieve high prefill and decode throughput for interactive sessions. throughput with Hermes agent: decode: 25+ tokens/sec prefill: 250+ tokens/sec # Then I gave the agent one task: Backtest a strategy: - Buy when RSI crosses above 30 - Sell at +2% profit or -1% stoploss - No overlapping positions - Use Google stock via yfinance - Generate a full HTML report with candlestick charts + signals What happened next was wild. It didn't just write code, it ran the entire workflow itself: Audited the environment (pip list, dependency check) Hit a ModuleNotFoundError, multiple Python installs were conflicting Ran where python to map every interpreter on the system Manually selected the correct Python 3.13 path and re ran the script Wrote a clean statevmachine backtester (strict no overlapping trades logic) Patched a yfinance MultiIndex quirk that would've crashed the script Built Plotly candlestick + RSI charts with buy/sell markers Calculated win rate, PnL, and summary stats Exported a polished single file HTML report. check the report at the end of the video or in the comments. Biggest takeaway: local LLMs aren't just "chat assistants" anymore. They debug their own environment, write production code, and ship a finished deliverable on consumer hardware, for $0 in API costs. If you're still calling local models "toys," you're already behind. This is just the beginning. Hermes agent just surpassed 1 trillion tokens in a single day on OpenRouter. Think about the scale of total token generation happening right now. Disclaimer: This is not financial advice. Consult a professional before making any trading decisions.show more

Free NVIDIA GPU with 16 GB VRAM GPU for... Running Local LLMs! If you want to master local LLMs but you're waiting until you can afford a $1,500 GPU, you're honestly not going to make it. The open source AI ecosystem is moving way too fast for you to wait on your budget to catch up. Especially when you can build a bleeding edge inference engine from scratch right now, completely for free. You don't need a heavy local rig to start. Google is literally letting you use an enterprise grade NVIDIA Tesla T4 GPU for $0/hour. At standard cloud computing rates (~$0.20/hr), Google Colab’s 4 hour daily free tier hands you roughly $24 worth of data center tier GPU compute every single month. And most people just waste it. Let’s talk about the hardware you get access to for free. The NVIDIA Tesla T4 is an absolute workhorse: - Architecture: NVIDIA Turing (TU104) - VRAM: 16GB GDDR6 (320 GB/s bandwidth) - Compute: 320 Tensor Cores | 2560 CUDA Cores - Performance: 130 TOPS INT8 | 8.1 TFLOPS FP32 - Power: Sipping energy at a max 70W TDP This is the exact same hardware I used to run DeepMind's Gemma 4 26B A4B QAT MoE at a 250,000 context window without a single Out Of Memory (OOM) crash. If you have a web browser and 10 minutes, you have everything you need. I’ve put together a fully documented, cell by cell Google Colab notebook that teaches you exactly how to do this. Here is what the notebook actually teaches you: - How to provision an Ubuntu Linux environment with CUDA 13.0 and verify your driver stack. - How to pull the source code and compile the latest llama.cpp C++ binaries from scratch, specifically optimizing the build for your exact GPU using the -DCMAKE_CUDA_ARCHITECTURES=native flag. - How to directly download quantized local LLMs (GGUF format) straight from HuggingFace using the CLI. - How to manage 16GB VRAM limits, offload neural network layers to the GPU, and push massive context windows. Compile raw llama.cpp, ollama run a model, or spin up the LM Studio CLI. Pick whatever stack you are comfortable with. just start building. No hardware. No credit card. No excuses. Bookmark this post right now so you don't lose the tutorial. Even if you don't have time to run it today, you are going to want this workflow in your engineering toolkit. The link to the free Colab Notebook is in the comments below. Lemme know if you need more tutorials like this.show more

llama.cpp isn't just for text LLMs anymore. Pure C++... zero shot voice cloning just officially landed in mainline. Text generation was only step one. If you’re building autonomous local AI agents, real time voice assistants, or edge workflows, instant low latency audio is the missing piece. Thanks to PR #26254, Alibaba’s state of the art Qwen3 TTS model family is now natively supported directly inside the llama.cpp repository under the multimodal (mtmd) framework. No Python bloat. No massive PyTorch CUDA overhead. Just raw, hyper optimized C++ running GGUF voice weights. Here is why this native update is a massive deal for the open source local AI stack: # Multimodal Architecture (.gguf + mmproj) Qwen3-TTS splits the workload between the base language model backbone and a multimodal projection adapter. llama.cpp handles this using the llama-tts binary, mapping the text model alongside its --mmproj projector to process audio tokens seamlessly. # Zero Shot Voice Cloning in Seconds You don't need fine tuning or massive dataset training. Feed the C++ engine a single 5 to 10 second .wav audio sample using the --tts-speaker-file flag, and it accurately clones the exact timbre, tone, and accent on the fly. # Real World T4 GPU Benchmark & Resource FootprintRunning the 1.7B Base model in 8-bit quantization (Q8_0): - VRAM Footprint: ~7 GB peak VRAM during active zero-shot cloning. - Audio Quality: Studio grade, natural-sounding voice output in seconds. • - Execution: Direct execution via native compiled binaries or sub process calls. # Coming Next to llama-server (PR #26603) Beyond CLI execution, a native POST /tts HTTP endpoint is currently being added to llama-server, which will soon allow you to trigger voice generation directly via standard REST API requests! # quick note on Colab compilation: Because this code was merged into mainline very recently, pre-built third-party binaries haven't fully caught up yet. Compiling llama-tts directly from source on Google Colab's free CPU instance can take about 1 hour (or ~1-2 minutes if targeting single GPU arch like -DCMAKE_CUDA_ARCHITECTURES=75). Be patient during the build step, or compile it locally on your own rig for instant execution! To test this out yourself, I built a zero config Google Colab notebook that compiles llama.cpp, downloads the Q8_0 GGUF files from HuggingFace, and spins up an interactive Gradio Studio UI so you can record/upload 3 second clips and clone voices in real time. Stop sleeping on native C++ audio. The era of bulky Python audio pipelines is officially over. Links to the free Google Colab notebook and the official ggml org GGUF HuggingFace model repository are in the replies below! available in q4 and q8 both variants, 1 GB and 1.85 GBs respectively (requires additional ~500MB mmproj gguf) Are you building local voice agents yet? What does your current audio stack look like? Drop your setups below!show more