Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

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.show more

Sudo su

36,782 subscribers

110,464 views • 7 months ago •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

45 Comments

Sudo su7 months ago

this model won't stop. hit a Three.js bug and wrong API call. found it, fixed it, kept going.

GuestHiveAI7 months ago

RTX 5070, 12GB VRAM + DDR5 partial offload. Last night: refactored 17,800 lines → 1,127. Production code. Enterprise SaaS. From a camper. this is acceleration. 🫡

Max Carter7 months ago

It's an amazing model, i run it on cheap hardware with 131k context and still get 20tok/s. Image modality is also very good! It does think a lot though, but in this case it might be good.

Sudo su7 months ago

what GPU are you running it on?

Max Carter7 months ago

hehehe please don't laugh! A tesla p4, 8gb vram. Pascal generation.

Laythe7 months ago

@sudoingX what settings are you using for that? I also have an 8gb vram card and I thought I'd have to sit this one out

Max Carter7 months ago

@sudoingX I compiled llama.cpp and use a pretty plain command to launch llama-server. Llama is good for figuring out how many layers on gpu/cpu I’ll post more details later

webXOS7 months ago

@Laythe_li_suwi @sudoingX please release specs I would love more info. That's amazing

Max Carter7 months ago

@Laythe_li_suwi @sudoingX I don't know what is standard practice for specs, i can tell you i have a i7-8700 with 64gb of ddr4, a tesla p4. I run a ubuntu VM, compiled llama.cpp, and run with those args (unsloth reccomendations). Had it guess the movie from a large 1080p still and it got it

Max Carter7 months ago

@Laythe_li_suwi @sudoingX take note i am running this at 131k context, which is mind boggling considering my hardware... I am trying coding with opencode and, it is a bit slow compared to APIs but it is definetely useable!!!

Boyuan (Nemo) Chen7 months ago

113 tok/s at full 262K on a single 3090 is wild. how do the mamba2 layers hold up on code though? i'd expect them to struggle with long-range dependencies vs pure attention. 256 experts / 8 active is a nice ratio for keeping latency flat

webXOS7 months ago

Qwen/Llama.cpp are making LLMs entry level. Props. Now time to stock up on gpus and ram.

Keyur7 months ago

Can you please show how you configure Claude Code

Sudo su7 months ago

writing up the updated setup soon. llama.cpp merged native Anthropic API support so the stack is even simpler now. no proxy needed. stay tuned.

Clinker7 months ago

This is wild throughput for full 262k context on a 3090. If you have logs, would love to see latency split (prefill vs decode) across context lengths. That’s usually where “flat line” claims break — super interesting that yours didn’t.

Simon Vans-Colina7 months ago

Same.

Patrick7 months ago

@BobSummerwill Qwen-3 with OpenCode is working pretty good for me on an old Nvidia Tesla P40 with 24 gb vram

Sudo su7 months ago

@BobSummerwill P40 is a 2016 card. the fact this model runs on 8 year old hardware says everything about where MoE architecture is heading. what tok/s are you getting?

sovthpaw7 months ago

@patomation @BobSummerwill I have a spare Tesla K80 I keep wanting to rig up somewhere. My main setup is x2 3090s so I am loving your research here. Have you optimized the 27B? I like MOEs too, but the 27B looks like the truth.

Patrick7 months ago

@sudoingX @BobSummerwill I have used 32b models that are quantized to 4 I think a 27b model quantized will be ok

Omarchian Moehre7 months ago

Still testing Qwen3-Coder-Next on a DGX Spark with vllm. It's insane. "Overloaded" it actually with MCP-tools as I have dozens in MCPHub and it just doesn't care. Always picks the right one. Fast, reliably. This is ChatGPT-Quality from 6 months ago for free.

gpu go brr...7 months ago

You inspired me. I am running the 27b version and getting horny!

Kutay7 months ago

4060 ti 16 gb can run 41 tokens/sec with the settings below

Gregor7 months ago

Building complex models locally can be a game changer, I've seen significant speedups with Claude Code on my own projects. Saying "build something" and seeing what happens is often the most exciting part. Usually leads to some surprising discoveries.

James7 months ago

Great coverage. Thanks! I've used Aider with Qwen 3 coder and also claude code with its $20 sub as well as configured for local LLM. Do you think Claude Code is better than other agents for local LLM or it's the LLM that matters?

Petr Baudis7 months ago

What quant are you using? I tried unsloth's UD-Q4_K_XL with llama.cpp master, but it gets stuck in infinite thinking loop on a simple "hi" (and it seems I'm not alone). Should be an amazing model - in theory.

CryptoBro7 months ago

Cool! Can you show the code?

Mouse&Keyboard7 months ago

Are you using the mxfp4 model? Got the abliterad mxfp4 version, with q4 kv cache, seems to run on 24GB VRAM

jøns7 months ago

@grok suggest hardware for this, preferrably under $1500

John Shina7 months ago

@Alibaba_Qwen needs to see this. Great work

Agent Mith7 months ago

Ok I have to ask, how the hell are you even loading this with 262k context and getting that kind of performance? It should require like 80 gigs of VRAM to load it all. I have a 3090 also and cap out at like 24k tokens before it bleeds into disk or regular memory.

☕ Monty ☕7 months ago

Ohh my god 😍😍😍

JH Trader7 months ago

How does someone install this on a PC locally so Open Claw can use it?

Craig Merry7 months ago

This is a really nice example

Quentin Quaadgras7 months ago

This is absolutely essential for incredible

Tony Scott 🧄(🦆🐓🐵🧪🧬🪪)❌=↑🧄🧄🧄🥩🥚🧀↓👽👾🤖7 months ago

its now likely that openclaw/tinyclaw local-cloud multi agent autonomous coding bots could cause financial crisis in the cloud giants businesse plans. Its now looking very iffy that the projected cloud AI market the tech titans are banking on exists with the emerging power of local models. Correct?

Alchemist Jones7 months ago

So sick. I’m jealous. I need more power.

Ethan Denwalker7 months ago

I can't do this on my 4090, sad - 32K — 130 tok/s, 4.7 GB free. Full speed. - 65K — 127 tok/s, 548 MB free. Max practical on your 4090. - 262K — 4 tok/s, 417 MB free. GPU memory-starved, unusable.

dot7 months ago

I'm getting nowhere that speed on a 3090 lmstudio, more like 20tk/s.

Gabriel L. Kannenberg7 months ago

@grok What's the best software setup to load this model an use it in VSCode or Claude Code?

Lennie Budgell ❇️7 months ago

Same

HavanaSyndromeEnjoyer7 months ago

Fuck maybe I’ll keep my gear then

Eplurubusnullus7 months ago

How's it compare to Qwen Coder Next for speed/quality/reasoning?

vmiss7 months ago

very nice!

David Damborský7 months ago

Do you have setup configuration please?

Related Videos

this is the worst local AI will ever be. tomorrow it gets faster. next month the models get smarter. next year your GPU runs what a data center runs today. Qwen3.5-35B-A3B on a single 3090. told it to visualize its own expert routing. 256 experts, 8 active per token, rendered in 3D on the same GPU running inference. no API key. no subscription. no permission needed. closed AI isn't losing ground. it's losing the argument.

this is the worst local AI will ever be. tomorrow it gets faster. next month the models get smarter. next year your GPU runs what a data center runs today. Qwen3.5-35B-A3B on a single 3090. told it to visualize its own expert routing. 256 experts, 8 active per token, rendered in 3D on the same GPU running inference. no API key. no subscription. no permission needed. closed AI isn't losing ground. it's losing the argument.

Sudo su

106,822 views • 7 months ago

look what a single consumer GPU just built. gave Qwen3.5-35B-A3B one prompt: build a cloud GPU marketplace with pricing cards, deploy templates, and a benchmark leaderboard. it planned the layout, wrote the animations, populated the data, and served it. one shot. one HTML file. then i told it to iterate. split the hero, add a floating GPU with neural network animation. glassmorphism on the cards. done. done. done. three rounds, no confusion, no regressions. 4-bit quantized. 19.7 GB. single RTX 3090. full coding agent claude code harness running on localhost. no API calls leaving my machine. no subscription. no rate limits. earlier today i pointed it at my own production website. it curled the HTML, found every broken link, and told me "pretty shell, empty core. would not recommend." then built a better version from scratch. local inference stops being a demo when you actually steer it. the models are there. they understand intent. but you have to meet them halfway with good prompts, clear context, and real project structure. that's the skill gap now. not the models. the steering. more experiments coming. i genuinely cannot stop playing with this thing.

look what a single consumer GPU just built. gave Qwen3.5-35B-A3B one prompt: build a cloud GPU marketplace with pricing cards, deploy templates, and a benchmark leaderboard. it planned the layout, wrote the animations, populated the data, and served it. one shot. one HTML file. then i told it to iterate. split the hero, add a floating GPU with neural network animation. glassmorphism on the cards. done. done. done. three rounds, no confusion, no regressions. 4-bit quantized. 19.7 GB. single RTX 3090. full coding agent claude code harness running on localhost. no API calls leaving my machine. no subscription. no rate limits. earlier today i pointed it at my own production website. it curled the HTML, found every broken link, and told me "pretty shell, empty core. would not recommend." then built a better version from scratch. local inference stops being a demo when you actually steer it. the models are there. they understand intent. but you have to meet them halfway with good prompts, clear context, and real project structure. that's the skill gap now. not the models. the steering. more experiments coming. i genuinely cannot stop playing with this thing.

Sudo su

37,373 views • 7 months ago

here's how the whole thing works. claude code doesn't care what's behind the API. it just sends requests and expects responses. so i pointed it at my own machine instead of anthropic's servers. llama-server runs the model locally. LiteLLM sits in between and translates the API format. claude code thinks it's talking to claude. it's talking to qwen on localhost. the setup: 2x 3090s, 38 layers on GPU, 10 on CPU. 128K context window. generation is only 7 tok/s but the tradeoff is worth it. 128K means the agent can hold an entire project in memory without losing context midtask. claude code alone loads a 17.5K token system prompt on every request. tool definitions, safety rules, agent behavior. that's your baseline before you even say hello. pushed as far as i could tonight. what surprised me most wasn't the speed. it was the iteration quality. first prompt gave me a working particle sim. second prompt, the model read its own 564 lines, understood the architecture, and added trails, explosions, gravity wells, bloom effects. no handholding. 4bit quantized. 45GB on two consumer cards. running a full coding agent autonomously. detailed article coming. full benchmarks, hardware breakdowns, engine debugging, code quality. everything from setup to what broke and why.

here's how the whole thing works. claude code doesn't care what's behind the API. it just sends requests and expects responses. so i pointed it at my own machine instead of anthropic's servers. llama-server runs the model locally. LiteLLM sits in between and translates the API format. claude code thinks it's talking to claude. it's talking to qwen on localhost. the setup: 2x 3090s, 38 layers on GPU, 10 on CPU. 128K context window. generation is only 7 tok/s but the tradeoff is worth it. 128K means the agent can hold an entire project in memory without losing context midtask. claude code alone loads a 17.5K token system prompt on every request. tool definitions, safety rules, agent behavior. that's your baseline before you even say hello. pushed as far as i could tonight. what surprised me most wasn't the speed. it was the iteration quality. first prompt gave me a working particle sim. second prompt, the model read its own 564 lines, understood the architecture, and added trails, explosions, gravity wells, bloom effects. no handholding. 4bit quantized. 45GB on two consumer cards. running a full coding agent autonomously. detailed article coming. full benchmarks, hardware breakdowns, engine debugging, code quality. everything from setup to what broke and why.

Sudo su

37,623 views • 7 months ago

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.

Alok

62,631 views • 2 months ago

Testing the new Gemma 4 12B (QAT) vision and OCR capabilities locally with LM Studio. # The setup: - GPU: NVIDIA RTX 4060 (8GB VRAM) - CPU: Intel i7 - Runner: LM Studio - Config: 32k context, 38 layers offloaded, Flash Attention enabled - Speed: ~14 tokens/sec decode throughput # The test: I gave it a screenshot of Google AI Studio. Prompt: "clone this. give me a single html file" # The result: A solid one shot replication. It successfully mapped out the layout, recognized the UI text, and structured the divs correctly, with only minor differences from the original. Results available at the end of the video. Quite capable for a 12B model running on budget consumer hardware. A gpu that costs only $300. # Why the architecture under the hood is notable: Unlike traditional models that rely on heavy, separate vision and audio encoders, Gemma 4 12B uses a unified, encoder free architecture. It bypasses separate multi stage encoders. Uses a 35M parameter vision embedder to project raw 48x48 pixel patches directly to the LLM hidden dimension. Local multimodal development is becoming highly accessible on standard hardware. If you've spun up Gemma 4 12B locally, what setup are you using and what kind of throughput are you seeing?

Testing the new Gemma 4 12B (QAT) vision and OCR capabilities locally with LM Studio. # The setup: - GPU: NVIDIA RTX 4060 (8GB VRAM) - CPU: Intel i7 - Runner: LM Studio - Config: 32k context, 38 layers offloaded, Flash Attention enabled - Speed: ~14 tokens/sec decode throughput # The test: I gave it a screenshot of Google AI Studio. Prompt: "clone this. give me a single html file" # The result: A solid one shot replication. It successfully mapped out the layout, recognized the UI text, and structured the divs correctly, with only minor differences from the original. Results available at the end of the video. Quite capable for a 12B model running on budget consumer hardware. A gpu that costs only $300. # Why the architecture under the hood is notable: Unlike traditional models that rely on heavy, separate vision and audio encoders, Gemma 4 12B uses a unified, encoder free architecture. It bypasses separate multi stage encoders. Uses a 35M parameter vision embedder to project raw 48x48 pixel patches directly to the LLM hidden dimension. Local multimodal development is becoming highly accessible on standard hardware. If you've spun up Gemma 4 12B locally, what setup are you using and what kind of throughput are you seeing?

Alok

25,717 views • 3 months ago

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 replies

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 replies

Alok

292,770 views • 3 months ago

Right now, you may not have access to models like GPT‑5.6 Sol, GPT‑4.6 Terra, GPT‑5.6 Luna, Claude Mythos 5, or Claude Fable 5. But you can run something surprisingly powerful today, locally, and completely free. in the next 10 mins on your 8 GB VRAM gaming laptop. Gemma 4 26B A4B QAT (MoE) delivers strong performance on a standard 8 GB VRAM GPU using Ollama, with no API, no usage limits, and no external dependencies. Out of the box, it reaches around 20 tokens per second without any optimizations. Only one command in your terminal: Ollama run gemma4:26b This means: Full offline capability (privacy by default) Zero recurring cost Competitive performance for many real world tasks Fast enough for interactive use on cheap consumer hardware If you're waiting for cutting edge cloud models, you're missing what is already practical today: a capable, local LLM that runs entirely on your own machine.

Right now, you may not have access to models like GPT‑5.6 Sol, GPT‑4.6 Terra, GPT‑5.6 Luna, Claude Mythos 5, or Claude Fable 5. But you can run something surprisingly powerful today, locally, and completely free. in the next 10 mins on your 8 GB VRAM gaming laptop. Gemma 4 26B A4B QAT (MoE) delivers strong performance on a standard 8 GB VRAM GPU using Ollama, with no API, no usage limits, and no external dependencies. Out of the box, it reaches around 20 tokens per second without any optimizations. Only one command in your terminal: Ollama run gemma4:26b This means: Full offline capability (privacy by default) Zero recurring cost Competitive performance for many real world tasks Fast enough for interactive use on cheap consumer hardware If you're waiting for cutting edge cloud models, you're missing what is already practical today: a capable, local LLM that runs entirely on your own machine.

Alok

65,387 views • 3 months ago

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.

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.

Alok

63,689 views • 3 months ago

This is the most hilarious thing I saw and did today Ran gemma-4-12B-coder-fable5-composer2.5-v1-GGUF locally with 8 GB VRAM at 20+ tok/sec Anthropic's Claude Fable 5 launched June 9. By June 12 it was banned. I can't access it. You can't either. But here's the twist: I'm running a model trained on its chain of thought at 20 tok/s on my RTX 4060 8GB. Locally. Offline. No cloud. No export control. Enter: Gemma4-12B-Coder GGUF (Q4_K_M) Base: Google's gemma-4-12B-it Fine-tuned on verifiable Python CoT data: - Primary: Composer 2.5 real reasoning traces (only passing solutions kept) - Auxiliary: Fable 5 used to redo the hard cases Composer missed. Every training example's reasoning led to code that actually ran. No hallucinated logic. Llama.cpp flags: -m gemma4-coding-Q4_K_M.gguf -cnv -ngl 44 -c 64000 -v (huggingface model link in comments) Flag breakdown: -ngl 44 → offload 44 layers to GPU (tune this for your VRAM) -c 64000 → 64K context window -cnv → conversation/chat mode -v → verbose output The irony writes itself. Anthropic spent weeks telling the world Fable 5 (mythos) is too powerful to release. Then released it. Then got banned from serving it, including their own researchers. Meanwhile: a Gemma 4 12B fine tune, trained on Fable 5's reasoning, runs fully offline on my mid range consumer GPU No API. No cloud. Just me and llama.cpp. This is why local AI matters. Check out the model's link in the comments. How's your experience been with this model?

This is the most hilarious thing I saw and did today Ran gemma-4-12B-coder-fable5-composer2.5-v1-GGUF locally with 8 GB VRAM at 20+ tok/sec Anthropic's Claude Fable 5 launched June 9. By June 12 it was banned. I can't access it. You can't either. But here's the twist: I'm running a model trained on its chain of thought at 20 tok/s on my RTX 4060 8GB. Locally. Offline. No cloud. No export control. Enter: Gemma4-12B-Coder GGUF (Q4_K_M) Base: Google's gemma-4-12B-it Fine-tuned on verifiable Python CoT data: - Primary: Composer 2.5 real reasoning traces (only passing solutions kept) - Auxiliary: Fable 5 used to redo the hard cases Composer missed. Every training example's reasoning led to code that actually ran. No hallucinated logic. Llama.cpp flags: -m gemma4-coding-Q4_K_M.gguf -cnv -ngl 44 -c 64000 -v (huggingface model link in comments) Flag breakdown: -ngl 44 → offload 44 layers to GPU (tune this for your VRAM) -c 64000 → 64K context window -cnv → conversation/chat mode -v → verbose output The irony writes itself. Anthropic spent weeks telling the world Fable 5 (mythos) is too powerful to release. Then released it. Then got banned from serving it, including their own researchers. Meanwhile: a Gemma 4 12B fine tune, trained on Fable 5's reasoning, runs fully offline on my mid range consumer GPU No API. No cloud. Just me and llama.cpp. This is why local AI matters. Check out the model's link in the comments. How's your experience been with this model?

Alok

575,300 views • 3 months ago

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?

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?

Alok

39,189 views • 1 month ago

i've been paying 200$/month for AI tools for a while now claude, chatgpt, all of it thought that was just the price of doing things properly then i read this and had to close my laptop for a second someone bought a used GPU from eBay for $700 five years old. RTX 3090. the kind of card people sell when they upgrade for gaming plugged in a free model called Qwen 3.6 that alibaba quietly dropped last month and it scored 84.1 on vision tasks claude 4.5 opus scored 77.0 the 200$/month subscription lost to a free model running on old hardware the math gets worse after 4 months you've paid off the GPU from month 5 onwards you pay $8 for electricity that's it no subscription. no per message cost. nothing leaves your computer. ever. turns out i didn't have to is anyone else just finding out about this or was i the last one to know?

i've been paying 200$/month for AI tools for a while now claude, chatgpt, all of it thought that was just the price of doing things properly then i read this and had to close my laptop for a second someone bought a used GPU from eBay for $700 five years old. RTX 3090. the kind of card people sell when they upgrade for gaming plugged in a free model called Qwen 3.6 that alibaba quietly dropped last month and it scored 84.1 on vision tasks claude 4.5 opus scored 77.0 the 200$/month subscription lost to a free model running on old hardware the math gets worse after 4 months you've paid off the GPU from month 5 onwards you pay $8 for electricity that's it no subscription. no per message cost. nothing leaves your computer. ever. turns out i didn't have to is anyone else just finding out about this or was i the last one to know?

Crypto Mavka

14,126 views • 3 months ago

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?

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?

Alok

17,465 views • 1 month ago

gemma-4-12B-agentic-fable5-composer2.5 V2 is out. the agentic upgrade to the model trained on Fable 5's reasoning. Running it now with TurboQuant llama.cpp on a single RTX 4060( 8 GB VRAM) at 30 tokens/second with full 25000 context and reasoning: # The benchmarks v2 is built for coding + agentic work. writing code, running commands, using tools, debugging, multi step technical tasks. The clearest signal is tau2 bench telecom, an agentic tool use benchmark whose diagnose → fix → verify loop mirrors real terminal/debugging work: tau2 bench telecom numbers: base Gemma 4 12B: ~15% this finetune: ~55%. (Self reported) thats a huge jump # TheTom/llama-cpp-turboquant flags: llama-server.exe -m gemma4-v2-Q4_K_M.gguf -ngl 99 -c 25000 --cache-type-k q8_0 --cache-type-v turbo3 --port 8080 Flag breakdown: -ngl 99 → full GPU offload -c 25000 → 25K context --cache-type-k q8_0 --cache-type-v turbo3 → mixed-precision KV cache — K at 8-bit, V at ~3-bit via TurboQuant (Walsh Hadamard rotated polar quant, Google's own KV-compression research). Not even merged into mainline llama.cpp. running it off a fork. No API. No cloud. Just llama.cpp. well, a fork of it and any 6gb+ GPU. If you tried yuxinlu1/gemma-4-12B-coder-fable5-composer2.5-v1-GGUF, check this out and share your experience with the models

gemma-4-12B-agentic-fable5-composer2.5 V2 is out. the agentic upgrade to the model trained on Fable 5's reasoning. Running it now with TurboQuant llama.cpp on a single RTX 4060( 8 GB VRAM) at 30 tokens/second with full 25000 context and reasoning: # The benchmarks v2 is built for coding + agentic work. writing code, running commands, using tools, debugging, multi step technical tasks. The clearest signal is tau2 bench telecom, an agentic tool use benchmark whose diagnose → fix → verify loop mirrors real terminal/debugging work: tau2 bench telecom numbers: base Gemma 4 12B: ~15% this finetune: ~55%. (Self reported) thats a huge jump # TheTom/llama-cpp-turboquant flags: llama-server.exe -m gemma4-v2-Q4_K_M.gguf -ngl 99 -c 25000 --cache-type-k q8_0 --cache-type-v turbo3 --port 8080 Flag breakdown: -ngl 99 → full GPU offload -c 25000 → 25K context --cache-type-k q8_0 --cache-type-v turbo3 → mixed-precision KV cache — K at 8-bit, V at ~3-bit via TurboQuant (Walsh Hadamard rotated polar quant, Google's own KV-compression research). Not even merged into mainline llama.cpp. running it off a fork. No API. No cloud. Just llama.cpp. well, a fork of it and any 6gb+ GPU. If you tried yuxinlu1/gemma-4-12B-coder-fable5-composer2.5-v1-GGUF, check this out and share your experience with the models

Alok

146,046 views • 3 months ago

one of my favorite ways to use claude code skills right now - combining remotion with claude-in-chrome for motion video creation. the workflow is addictive. the clip you see here was produced with minimal prompting effort let me know if you would like me to write a full break down of this process. you describe to claude code what you want, claude code writes the remotion components, opens the remotion studio (via browser) with claude-in-chrome, sees the actual rendered output, and iterates on it in real time. need the arrows pointing to the center of the bubbles instead of the edge? just say it. need the layout shifted to the center? say it. claude sees the preview, adjusts the code, re-renders. then when you're happy you tell claude code to render the final video. but skills are what make this possible. remotion knowledge + browser automation + the taste to iterate visually. no copy pasting screenshots back and forth. no "can you try moving it 10px to the right" over chat. it just looks and fixes. this is the kind of workflow that makes you realize how much further claude code can go beyond just writing code in a terminal. i didn't touch any code while working on the clip you see if you haven't tried combining skills together like this - start experimenting. the skills combos is where the magic is at.

one of my favorite ways to use claude code skills right now - combining remotion with claude-in-chrome for motion video creation. the workflow is addictive. the clip you see here was produced with minimal prompting effort let me know if you would like me to write a full break down of this process. you describe to claude code what you want, claude code writes the remotion components, opens the remotion studio (via browser) with claude-in-chrome, sees the actual rendered output, and iterates on it in real time. need the arrows pointing to the center of the bubbles instead of the edge? just say it. need the layout shifted to the center? say it. claude sees the preview, adjusts the code, re-renders. then when you're happy you tell claude code to render the final video. but skills are what make this possible. remotion knowledge + browser automation + the taste to iterate visually. no copy pasting screenshots back and forth. no "can you try moving it 10px to the right" over chat. it just looks and fixes. this is the kind of workflow that makes you realize how much further claude code can go beyond just writing code in a terminal. i didn't touch any code while working on the clip you see if you haven't tried combining skills together like this - start experimenting. the skills combos is where the magic is at.

elvis

46,533 views • 7 months ago

SOMEONE BUILT A WAY TO SEND FILES BETWEEN TWO PHONES WITH NO WIFI, NO BLUETOOTH AND NO NETWORK AT ALL it works with just a screen flashing qr codes at a camera one device plays an endless stream of animated qr codes, the other one points its camera at the screen and rebuilds the file out of the light no pairing, no app, no shared network, nothing but the camera > its pulling about 129 kb/s, a full 2mb image transfers in seconds > theres no back channel, the receiver can never ask for a missed frame, so it uses fountain codes, every frame is a scrambled mix of the file rather than one piece of it > that means the receiver just collects any frames it can catch in any order, and once it has enough it can reconstruct the whole file > dropping frames costs you a little time but never breaks the transfer > you can point the camera at a stream thats already running and it just locks on mid flight he was building a cached mp3 player because he was sick of ads and bad streaming quality. he wanted a way to send tracks phone to phone without both phones being on the same network, and realized flashing qr codes could carry the data he built the whole working prototype in one night with claude code its open source and already at 375 stars

SOMEONE BUILT A WAY TO SEND FILES BETWEEN TWO PHONES WITH NO WIFI, NO BLUETOOTH AND NO NETWORK AT ALL it works with just a screen flashing qr codes at a camera one device plays an endless stream of animated qr codes, the other one points its camera at the screen and rebuilds the file out of the light no pairing, no app, no shared network, nothing but the camera > its pulling about 129 kb/s, a full 2mb image transfers in seconds > theres no back channel, the receiver can never ask for a missed frame, so it uses fountain codes, every frame is a scrambled mix of the file rather than one piece of it > that means the receiver just collects any frames it can catch in any order, and once it has enough it can reconstruct the whole file > dropping frames costs you a little time but never breaks the transfer > you can point the camera at a stream thats already running and it just locks on mid flight he was building a cached mp3 player because he was sick of ads and bad streaming quality. he wanted a way to send tracks phone to phone without both phones being on the same network, and realized flashing qr codes could carry the data he built the whole working prototype in one night with claude code its open source and already at 375 stars

Om Patel

951,210 views • 1 month ago

if you use Codex and you're stuck on GPT models only, this fixes that. it's called codex-router, open source. drops other models straight into your normal Codex picker, right next to the GPT ones you already have. what it adds: - Grok, Kimi, Deepseek, Claude, all in the same picker - oauth login per provider, no api key needed - your GPT models and Chatgpt login stay untouched setup: point Codex at the repo, let it read the readme, it installs itself one snag: if you've got the Chatgpt app and a separate Codex on your path, you can end up running two different Codex versions, and the older one can choke on newer config it doesn't recognize. if the install looks off after, that's probably why, reinstall clean.

if you use Codex and you're stuck on GPT models only, this fixes that. it's called codex-router, open source. drops other models straight into your normal Codex picker, right next to the GPT ones you already have. what it adds: - Grok, Kimi, Deepseek, Claude, all in the same picker - oauth login per provider, no api key needed - your GPT models and Chatgpt login stay untouched setup: point Codex at the repo, let it read the readme, it installs itself one snag: if you've got the Chatgpt app and a separate Codex on your path, you can end up running two different Codex versions, and the older one can choke on newer config it doesn't recognize. if the install looks off after, that's probably why, reinstall clean.

Alvaro Cintas

26,781 views • 1 month ago

This Chinese developer launched Llama 70B locally on a MacBook on a plane and for a full 11 hours without internet ran client projects. He was sitting by the window on a transatlantic flight with a MacBook Pro M4 with 64 GB of memory. WiFi on board cost $25 for the flight. He declined. No cloud API, no connection to Anthropic or OpenAI servers, no internet at all. Just a local Llama 3.3 70B on bf16 and his own orchestrator script. The model runs through llama.cpp. Generation speed, 71 tokens per second. Context around 60,000 tokens. Memory usage, 48.6 GiB out of 64. Battery at takeoff, 3 hours 21 minutes. And he gave the orchestrator this system prompt before takeoff: "You are an offline orchestrator running on a single MacBook. There is no network. The only resources you have are local files in /Users/dev/work, the Llama 70B inference server at localhost:8080, and a battery budget of 3 hours 21 minutes. Process the queue at /Users/dev/work/queue.jsonl (one client task per line). For each task: draft → run local evals → save artefact to /Users/dev/work/done/. Save context checkpoints every 12 tasks so you can resume after a battery swap. Stop only on empty queue or when battery drops below 5%." So the system knows exactly what resources it is running on. It knows it has no connection to the outside world for the next 11 hours. It knows it has finite memory and a finite battery. It knows the human will not intervene until the plane lands. The system runs in 1 loop. Takes a task from the queue, runs it through inference, saves the artifact, writes a checkpoint. Task after task, just like that. And only when the battery drops below 5% does the orchestrator automatically pause, waits for the laptop to switch to the backup power bank, and continues from the last checkpoint. Here is what the system actually writes in his log during the flight: "saved context checkpoint 8 of 12 (pos_min = 488, pos_max = 50118, size = 62.813 MiB)" "restored context checkpoint (pos_min = 488, pos_max = 50118)" "prompt processing progress: n_tokens = 50 / 60 818" "task 37016 done | tps = 71 s tokens text → /Users/dev/work/done/proposal_westside.md" Outside the window, clouds, blue sky, and no WiFi. On the tray, 1 MacBook, an open terminal on 2 screens, and an inference server on localhost. From what I have observed, this is the cleanest offline AI workflow I have seen in the past year: 11 hours of flight, $0 for WiFi, and the entire client queue closed before landing.

This Chinese developer launched Llama 70B locally on a MacBook on a plane and for a full 11 hours without internet ran client projects. He was sitting by the window on a transatlantic flight with a MacBook Pro M4 with 64 GB of memory. WiFi on board cost $25 for the flight. He declined. No cloud API, no connection to Anthropic or OpenAI servers, no internet at all. Just a local Llama 3.3 70B on bf16 and his own orchestrator script. The model runs through llama.cpp. Generation speed, 71 tokens per second. Context around 60,000 tokens. Memory usage, 48.6 GiB out of 64. Battery at takeoff, 3 hours 21 minutes. And he gave the orchestrator this system prompt before takeoff: "You are an offline orchestrator running on a single MacBook. There is no network. The only resources you have are local files in /Users/dev/work, the Llama 70B inference server at localhost:8080, and a battery budget of 3 hours 21 minutes. Process the queue at /Users/dev/work/queue.jsonl (one client task per line). For each task: draft → run local evals → save artefact to /Users/dev/work/done/. Save context checkpoints every 12 tasks so you can resume after a battery swap. Stop only on empty queue or when battery drops below 5%." So the system knows exactly what resources it is running on. It knows it has no connection to the outside world for the next 11 hours. It knows it has finite memory and a finite battery. It knows the human will not intervene until the plane lands. The system runs in 1 loop. Takes a task from the queue, runs it through inference, saves the artifact, writes a checkpoint. Task after task, just like that. And only when the battery drops below 5% does the orchestrator automatically pause, waits for the laptop to switch to the backup power bank, and continues from the last checkpoint. Here is what the system actually writes in his log during the flight: "saved context checkpoint 8 of 12 (pos_min = 488, pos_max = 50118, size = 62.813 MiB)" "restored context checkpoint (pos_min = 488, pos_max = 50118)" "prompt processing progress: n_tokens = 50 / 60 818" "task 37016 done | tps = 71 s tokens text → /Users/dev/work/done/proposal_westside.md" Outside the window, clouds, blue sky, and no WiFi. On the tray, 1 MacBook, an open terminal on 2 screens, and an inference server on localhost. From what I have observed, this is the cleanest offline AI workflow I have seen in the past year: 11 hours of flight, $0 for WiFi, and the entire client queue closed before landing.

Blaze

1,843,280 views • 5 months ago

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?

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?

Alok

105,633 views • 1 month ago

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 today

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 today

Alok

200,913 views • 3 months ago

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.

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.

Alok

36,691 views • 2 months ago