正在加载视频...

视频加载失败

加载此视频时出现问题。这可能是由于临时网络问题，或视频可能不可用。

The leading model runnable on a single cloud H100 GPU now fits on a single home GPU! 🔥 We've optimized Gemma 3 27B with QAT so you can run our best-in-class open model on your desktop RTX 3090 or similar. See how easy it is to try via ollama! 👇

Glenn Cameron Jr

9,909 subscribers

81,427 次观看 • 1 年前 •via X (Twitter)

游戏科学技术教育

Anya Rossi• Live Now

Private livecam show

0 条评论

暂无评论

原始帖子的评论将显示在这里

相关视频

Trained a simple world model for my robot arm. It predicts the future over 20000 times faster than real time on a single NVIDIA RTX 3090 GPU (128 batch -> 160x faster each).

Trained a simple world model for my robot arm. It predicts the future over 20000 times faster than real time on a single NVIDIA RTX 3090 GPU (128 batch -> 160x faster each).

Alexander Koch

40,821 次观看 • 2 年前

The new PrismML’s Bonsai 27B model running on iPhone 17 Pro Bonsai 27B is the first 27B-class model to run on a phone and it uses less than 5GB of memory Pushing the limits on what can run a phone

The new PrismML’s Bonsai 27B model running on iPhone 17 Pro Bonsai 27B is the first 27B-class model to run on a phone and it uses less than 5GB of memory Pushing the limits on what can run a phone

Adrien Grondin

67,095 次观看 • 1 个月前

Claude fable 5 is cooked Claude fable 5 Vs Gemma 4 26b a4b qat MTP Gemma 4 26b running locally on my 8GB vram single RTX 4060 built this using three.js in a single session and 3 prompts. no cloud, no subscription 100% private unlimited use. how long until it can catch up completely?

Claude fable 5 is cooked Claude fable 5 Vs Gemma 4 26b a4b qat MTP Gemma 4 26b running locally on my 8GB vram single RTX 4060 built this using three.js in a single session and 3 prompts. no cloud, no subscription 100% private unlimited use. how long until it can catch up completely?

Alok

30,802 次观看 • 2 个月前

it's open source time, with a real leap for world models 🎉 NVIDIA's SANA-WM: a camera-conditioned world model that fits on one GPU. 60s of 720p in 34s on a single 5090 - 2.6B params and Apache 2.0!

it's open source time, with a real leap for world models 🎉 NVIDIA's SANA-WM: a camera-conditioned world model that fits on one GPU. 60s of 720p in 34s on a single 5090 - 2.6B params and Apache 2.0!

Victor M

34,386 次观看 • 3 个月前

Stanford dropped FramePack This AI can run on 6 GB laptop GPU to generate minute long 30fps video from single image No distillation, open source. 10 wild examples & how to try it: 👇

Stanford dropped FramePack This AI can run on 6 GB laptop GPU to generate minute long 30fps video from single image No distillation, open source. 10 wild examples & how to try it: 👇

Min Choi

633,762 次观看 • 1 年前

In what ways are you using #AI on your RTX GPU in your personal or professional life? 🤔 Reply with your answer and use #AIonRTX for a chance to win a MSI Gaming GeForce RTX 4080 GAMING X TRIO GPU. 👇

In what ways are you using #AI on your RTX GPU in your personal or professional life? 🤔 Reply with your answer and use #AIonRTX for a chance to win a MSI Gaming GeForce RTX 4080 GAMING X TRIO GPU. 👇

NVIDIA AI Developer

49,468 次观看 • 2 年前

#RTXRemix can leverage AI on your RTX GPU to help remaster classic games! 🎮 What game would you like to see remastered with #RTXRemix? 🤔 Repost & tell us below with #AIonRTX for a chance to win a GeForce RTX 4090 GPU! 👇

#RTXRemix can leverage AI on your RTX GPU to help remaster classic games! 🎮 What game would you like to see remastered with #RTXRemix? 🤔 Repost & tell us below with #AIonRTX for a chance to win a GeForce RTX 4090 GPU! 👇

NVIDIA Studio

59,305 次观看 • 2 年前

Did you know that AI chatbots like ChatRTX can run on your local PC with a GeForce RTX GPU? 🙌 How would you use ChatRTX to elevate your gaming experinece? Let us know us know below and use #AIonRTX for a chance to WIN an RTX ON Keycap or RTX 4090 poster! 👇

Did you know that AI chatbots like ChatRTX can run on your local PC with a GeForce RTX GPU? 🙌 How would you use ChatRTX to elevate your gaming experinece? Let us know us know below and use #AIonRTX for a chance to WIN an RTX ON Keycap or RTX 4090 poster! 👇

NVIDIA GeForce

66,749 次观看 • 2 年前

Fine-tune DeepSeek-OCR on your own language! (100% local) DeepSeek-OCR is a 3B-parameter vision model that achieves 97% precision while using 10× fewer vision tokens than text-based LLMs. It handles tables, papers, and handwriting without killing your GPU or budget. Why it matters: Most vision models treat documents as massive sequences of tokens, making long-context processing expensive and slow. DeepSeek-OCR uses context optical compression to convert 2D layouts into vision tokens, enabling efficient processing of complex documents. The best part? You can easily fine-tune it for your specific use case on a single GPU. I used Unsloth to run this experiment on Persian text and saw an 88.26% improvement in character error rate. ↳ Base model: 149% character error rate (CER) ↳ Fine-tuned model: 60% CER (57% more accurate) ↳ Training time: 60 steps on a single GPU Persian was just the test case. You can swap in your own dataset for any language, document type, or specific domain you're working with. I've shared the complete guide in the next tweet - all the code, notebooks, and environment setup ready to run with a single click. Everything is 100% open-source!

Fine-tune DeepSeek-OCR on your own language! (100% local) DeepSeek-OCR is a 3B-parameter vision model that achieves 97% precision while using 10× fewer vision tokens than text-based LLMs. It handles tables, papers, and handwriting without killing your GPU or budget. Why it matters: Most vision models treat documents as massive sequences of tokens, making long-context processing expensive and slow. DeepSeek-OCR uses context optical compression to convert 2D layouts into vision tokens, enabling efficient processing of complex documents. The best part? You can easily fine-tune it for your specific use case on a single GPU. I used Unsloth to run this experiment on Persian text and saw an 88.26% improvement in character error rate. ↳ Base model: 149% character error rate (CER) ↳ Fine-tuned model: 60% CER (57% more accurate) ↳ Training time: 60 steps on a single GPU Persian was just the test case. You can swap in your own dataset for any language, document type, or specific domain you're working with. I've shared the complete guide in the next tweet - all the code, notebooks, and environment setup ready to run with a single click. Everything is 100% open-source!

Akshay 🚀

126,122 次观看 • 9 个月前

NVIDIA open-sourced a 600M model that transcribes 40 languages in real-time at 80ms latency and it costs $0. that's faster than you can blink. across mandarin, arabic, hindi, portuguese, tagalog, whatever,from a SINGLE checkpoint. → 17x more concurrent streams than buffered ASR on the same H100. → punctuation + capitalization built-in. no post-processing. → runs on your own GPU. no API bill 100% Open Source.

NVIDIA open-sourced a 600M model that transcribes 40 languages in real-time at 80ms latency and it costs $0. that's faster than you can blink. across mandarin, arabic, hindi, portuguese, tagalog, whatever,from a SINGLE checkpoint. → 17x more concurrent streams than buffered ASR on the same H100. → punctuation + capitalization built-in. no post-processing. → runs on your own GPU. no API bill 100% Open Source.

Superman

104,663 次观看 • 1 个月前

i found a way to make UNCENSORED AI AGENT on a RTX 4090 GPU (!!!) with LOCAL 30B model weights this is GLM-4.7-Flash with abliteration, need 24GB VRAM, safety alignment surgically removed from the weights, the model has native tool calling, it actually executes bash, edits files, runs git (1) use ollama to pull weights of GLM > ollama pull huihui_ai/glm-4.7-flash-abliterated:q4_K (2) proxy it to any coding agent via ollama > ollama launch claude --model huihui_ai/glm-4.7-flash-abliterated:q4_K > ollama launch codex --model huihui_ai/glm-4.7-flash-abliterated:q4_K > ollama launch opencode --model huihui_ai/glm-4.7-flash-abliterated:q4_K (3) have fun

i found a way to make UNCENSORED AI AGENT on a RTX 4090 GPU (!!!) with LOCAL 30B model weights this is GLM-4.7-Flash with abliteration, need 24GB VRAM, safety alignment surgically removed from the weights, the model has native tool calling, it actually executes bash, edits files, runs git (1) use ollama to pull weights of GLM > ollama pull huihui_ai/glm-4.7-flash-abliterated:q4_K (2) proxy it to any coding agent via ollama > ollama launch claude --model huihui_ai/glm-4.7-flash-abliterated:q4_K > ollama launch codex --model huihui_ai/glm-4.7-flash-abliterated:q4_K > ollama launch opencode --model huihui_ai/glm-4.7-flash-abliterated:q4_K (3) have fun

chiefofautism

341,710 次观看 • 6 个月前

Bonsai 27B running locally on an iPhone in Atomic Chat! Bonsai is the first 27B-class model that fits on a phone. PrismML built it on Qwen3.6 27B with 1-bit weights. It takes 3.9GB instead of 54GB and keeps ~90% of the benchmark scores. Available now on iPhone and Android

Bonsai 27B running locally on an iPhone in Atomic Chat! Bonsai is the first 27B-class model that fits on a phone. PrismML built it on Qwen3.6 27B with 1-bit weights. It takes 3.9GB instead of 54GB and keeps ~90% of the benchmark scores. Available now on iPhone and Android

atomic.chat

54,808 次观看 • 1 个月前

watch anon. a 27b model thinking out loud on gtx 1660 super, i asked bonsai what model it is and it reasoned through the whole answer before it spoke. 20 tokens a second, gpu pinned at 125 watts, 4.25 of 6 gigs used. bonsai is a 1bit quant of qwen 3.6 27b, the king of the 3090 crushed to 3.5gb. turns out the hardware was in your drawer the whole time. you just needed the quant to catch up.

watch anon. a 27b model thinking out loud on gtx 1660 super, i asked bonsai what model it is and it reasoned through the whole answer before it spoke. 20 tokens a second, gpu pinned at 125 watts, 4.25 of 6 gigs used. bonsai is a 1bit quant of qwen 3.6 27b, the king of the 3090 crushed to 3.5gb. turns out the hardware was in your drawer the whole time. you just needed the quant to catch up.

Sudo su

26,133 次观看 • 1 个月前

NVIDIA Nemotron 3 Nano Omni, a new multimodal reasoning model, is now live on Jetson AI Lab and unifies vision, audio, and language into a single reasoning loop. 🙌 Power your NemoClaws by running this model with Ollama, vLLM and other inference frameworks on NVIDIA Jetson hardware. Try it ➡️

NVIDIA Nemotron 3 Nano Omni, a new multimodal reasoning model, is now live on Jetson AI Lab and unifies vision, audio, and language into a single reasoning loop. 🙌 Power your NemoClaws by running this model with Ollama, vLLM and other inference frameworks on NVIDIA Jetson hardware. Try it ➡️

NVIDIA Robotics

16,031 次观看 • 3 个月前

ABot World 0.5B is out! we have now world models at home, running in real-time on consumer GPU feed it an initial image, steer it with with your keyboard 🌎 you can run it locally or play now on spaces ▶️

ABot World 0.5B is out! we have now world models at home, running in real-time on consumer GPU feed it an initial image, steer it with with your keyboard 🌎 you can run it locally or play now on spaces ▶️

Hugging Apps

140,763 次观看 • 21 天前

BOOM! STANFORD LAUNCHES FRAMEPACK A FREE OPEN SOURCE AI THAT CAN RUN ON 6 GB LAPTOP GPU TO GENERATE MINUTE LONG 30FPS VIDEO FROM SINGLE IMAGE. It is game changing…

BOOM! STANFORD LAUNCHES FRAMEPACK A FREE OPEN SOURCE AI THAT CAN RUN ON 6 GB LAPTOP GPU TO GENERATE MINUTE LONG 30FPS VIDEO FROM SINGLE IMAGE. It is game changing…

Brian Roemmele

534,742 次观看 • 1 年前

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 次观看 • 2 个月前

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.

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.

Alok

26,841 次观看 • 2 个月前

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 次观看 • 2 个月前