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。

Our powerful DeepSeek R1 Turbo is live on Hugging Face! 🤗 It's improved in every way: 64K context + 16K max output + 30 throughput + 99.9% stability + 💰 CHEAPER THAN EVER Try it out on Hugging Face's model page 🫶 Julien Chaumond

Novita AI

3,776 subscribers

23,644 次观看 • 1 年前 •via X (Twitter)

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

This #CVPR2026 paper from our research team is trending #1 on Hugging Face 🤗 Meet LocateAnything: a vision-language detection model that rethinks bounding box prediction. For AI agents and robots, “seeing” is only useful if a model can pinpoint where something is fast enough to act. Trained on 138M high-quality samples, LocateAnything decodes bounding boxes in parallel instead of one coordinate at a time, improving localization accuracy while dramatically increasing throughput for visual grounding and detection. Project page:

This #CVPR2026 paper from our research team is trending #1 on Hugging Face 🤗 Meet LocateAnything: a vision-language detection model that rethinks bounding box prediction. For AI agents and robots, “seeing” is only useful if a model can pinpoint where something is fast enough to act. Trained on 138M high-quality samples, LocateAnything decodes bounding boxes in parallel instead of one coordinate at a time, improving localization accuracy while dramatically increasing throughput for visual grounding and detection. Project page:

NVIDIA AI

341,218 次观看 • 3 个月前

⭐The Year of Inference is here. Featherless is now an official inference provider on Hugging Face, unlocking 6,700+ LLMs for anyone to run, eval, and deploy instantly. It all starts with accessibility. From DeepSeek to Mistral, LLaMA to Qwen — powerful LLMs are one click away. We believe the future of AI is shaped by the long tail: personalized, specialized models tuned to real people’s needs. To get there, inference must be open, affordable, and usable by all. Whether you're fine-tuning, prototyping, or scaling a product, this moment is for you. 🫱🏻‍🫲🏻Let’s make inference the easiest part of building with AI. 📢 Share this so more builders know what’s now possible. Excited to be partnering with clem 🤗 Julien Chaumond Vaibhav (VB) Srivastav Simon Brandeis & Hugging Face team to take this to the next level!

⭐The Year of Inference is here. Featherless is now an official inference provider on Hugging Face, unlocking 6,700+ LLMs for anyone to run, eval, and deploy instantly. It all starts with accessibility. From DeepSeek to Mistral, LLaMA to Qwen — powerful LLMs are one click away. We believe the future of AI is shaped by the long tail: personalized, specialized models tuned to real people’s needs. To get there, inference must be open, affordable, and usable by all. Whether you're fine-tuning, prototyping, or scaling a product, this moment is for you. 🫱🏻‍🫲🏻Let’s make inference the easiest part of building with AI. 📢 Share this so more builders know what’s now possible. Excited to be partnering with clem 🤗 Julien Chaumond Vaibhav (VB) Srivastav Simon Brandeis & Hugging Face team to take this to the next level!

Featherless AI

24,033 次观看 • 1 年前

This free model just beat every closed-source AI on coding benchmarks. Open weights. 6x cheaper than Opus. The labs don't want you to know it exists. > GLM-5.2 from Zai just topped Code Arena - the first open-weights model to ever hold the #1 coding spot. Not a leaked weight, not a fine-tune. A fully open model beating GPT and Claude on their own turf. It's live on Hugging Face Inference API right now. 5 providers: Novita, Together AI, Fireworks, Deepinfra, Zai. OpenAI-compatible client. → Go to huggingface(.)co → grab HF_TOKEN from account settings → plug into any OpenAI-compatible client 6x cheaper than Opus. Companies bleeding on AI bills are already routing to this for orchestration, caching, and token optimization. The smart money moved before the headline dropped. > Now you know. huggingface(.)co/zai-org/GLM-5.2 Bookmark this before everyone else figures it out.

This free model just beat every closed-source AI on coding benchmarks. Open weights. 6x cheaper than Opus. The labs don't want you to know it exists. > GLM-5.2 from Zai just topped Code Arena - the first open-weights model to ever hold the #1 coding spot. Not a leaked weight, not a fine-tune. A fully open model beating GPT and Claude on their own turf. It's live on Hugging Face Inference API right now. 5 providers: Novita, Together AI, Fireworks, Deepinfra, Zai. OpenAI-compatible client. → Go to huggingface(.)co → grab HF_TOKEN from account settings → plug into any OpenAI-compatible client 6x cheaper than Opus. Companies bleeding on AI bills are already routing to this for orchestration, caching, and token optimization. The smart money moved before the headline dropped. > Now you know. huggingface(.)co/zai-org/GLM-5.2 Bookmark this before everyone else figures it out.

Atenov int.

12,395 次观看 • 2 个月前

Baidu just open-sourced an OCR model that reads entire 40-page documents in one shot. It's called Unlimited-OCR. 3 billion parameters but only 500 million active during inference. Runs 100% locally on your machine. Why this matters: traditional OCR tools chop documents page by page. Tables that span two pages break. Reading order gets lost. Cross-page context disappears. Unlimited-OCR processes the whole document at once. 32K context window. Text, formulas, tables, reading order all preserved across pages. Output comes out as clean structured Markdown. → 93% accuracy on the standard benchmark. +6 points over the baseline. → Error rate stays below 0.11 even past 40 pages. → Multilingual out of the box. → 2.12 million downloads on Hugging Face last month. 14,600 GitHub stars. For context: Amazon Textract, Google Cloud Vision, and Azure Document Intelligence all charge per page. This runs locally for free.

Baidu just open-sourced an OCR model that reads entire 40-page documents in one shot. It's called Unlimited-OCR. 3 billion parameters but only 500 million active during inference. Runs 100% locally on your machine. Why this matters: traditional OCR tools chop documents page by page. Tables that span two pages break. Reading order gets lost. Cross-page context disappears. Unlimited-OCR processes the whole document at once. 32K context window. Text, formulas, tables, reading order all preserved across pages. Output comes out as clean structured Markdown. → 93% accuracy on the standard benchmark. +6 points over the baseline. → Error rate stays below 0.11 even past 40 pages. → Multilingual out of the box. → 2.12 million downloads on Hugging Face last month. 14,600 GitHub stars. For context: Amazon Textract, Google Cloud Vision, and Azure Document Intelligence all charge per page. This runs locally for free.

Vaibhav Sisinty

415,690 次观看 • 1 个月前

China open-sourced a peanut-sized OCR that parses entire 100-page PDFs in one shot.. It's called Unlimited-OCR. Only 3B params. Runs locally. Every other OCR tool chops your doc into pages and loses the thread. this one reads the whole thing in a single pass. → One-shot "long-horizon" parsing (32K context window) → Multilingual, out of the box → 93% on the standard parsing benchmark (+6 over baseline) → <0.11 error rate past 40 pages → Runs 100% locally on your own hardware → Works with Transformers, vLLM, SGLang, Docker, Ollama, llama.cpp Traditional cloud OCR (Textract, Google Vision, Azure Doc Intelligence) costs $1.50–$15 per 1,000 pages. This runs on your machine. For free. Forever. Baidu built it explicitly to push DeepSeek-OCR one step further. Already at 1.9M downloads on Hugging Face and most people have no idea it exists yet. 100% open source.

China open-sourced a peanut-sized OCR that parses entire 100-page PDFs in one shot.. It's called Unlimited-OCR. Only 3B params. Runs locally. Every other OCR tool chops your doc into pages and loses the thread. this one reads the whole thing in a single pass. → One-shot "long-horizon" parsing (32K context window) → Multilingual, out of the box → 93% on the standard parsing benchmark (+6 over baseline) → <0.11 error rate past 40 pages → Runs 100% locally on your own hardware → Works with Transformers, vLLM, SGLang, Docker, Ollama, llama.cpp Traditional cloud OCR (Textract, Google Vision, Azure Doc Intelligence) costs $1.50–$15 per 1,000 pages. This runs on your machine. For free. Forever. Baidu built it explicitly to push DeepSeek-OCR one step further. Already at 1.9M downloads on Hugging Face and most people have no idea it exists yet. 100% open source.

Superman

1,104,783 次观看 • 1 个月前

48 consistent camera feeds of me all from one casual video capture. This is 4DAnyone. I am testing it out over the holiday week, the goal is 4D Gaussian Splatting from a single camera. Should out to Pablo Vela' for inspiring me to use Rerun to visualize my progress. Right now you see 6 cameras, but I can cycle through all 48. I hacked together this visualization using ChatGPT's new Astra model in VS Code. Literally took less than 10 minutes. More coming soon. Learn more about 4DAnyone Here: You can also try Pablo's 6 view 4DAnyone generator on Hugging Face here: Jeremy Park, PhD Robert Scoble #Computervision #3D #4DGS

48 consistent camera feeds of me all from one casual video capture. This is 4DAnyone. I am testing it out over the holiday week, the goal is 4D Gaussian Splatting from a single camera. Should out to Pablo Vela' for inspiring me to use Rerun to visualize my progress. Right now you see 6 cameras, but I can cycle through all 48. I hacked together this visualization using ChatGPT's new Astra model in VS Code. Literally took less than 10 minutes. More coming soon. Learn more about 4DAnyone Here: You can also try Pablo's 6 view 4DAnyone generator on Hugging Face here: Jeremy Park, PhD Robert Scoble #Computervision #3D #4DGS

Jonathan Stephens

12,664 次观看 • 9 天前

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 次观看 • 2 个月前

AI agent usage on SQD Portal is up ~200% in recent weeks. A dev from our community chat was scraping a wallet UI with Hermes. Mid-task, DeepSeek reasoned its way out of it: "I can use SQD Portal's Hyperliquid fills data directly — much more complete than scraping a UI with infinite scroll." No prompt engineering. The model just chose the better and faster path. This is the loop we wanted: Agents pick SQD because it's faster → devs see agents picking SQD → devs ship faster → more agents pick SQD The picks-and-shovels moment for AI x onchain is here.

AI agent usage on SQD Portal is up ~200% in recent weeks. A dev from our community chat was scraping a wallet UI with Hermes. Mid-task, DeepSeek reasoned its way out of it: "I can use SQD Portal's Hyperliquid fills data directly — much more complete than scraping a UI with infinite scroll." No prompt engineering. The model just chose the better and faster path. This is the loop we wanted: Agents pick SQD because it's faster → devs see agents picking SQD → devs ship faster → more agents pick SQD The picks-and-shovels moment for AI x onchain is here.

sqd.ai

14,394 次观看 • 4 个月前

I just ran Gemma 4 31B on @CerebrasSystems at 1,800+ tokens/sec and it's multimodal. For context: that's 35x faster than a typical GPU endpoint, and the first token (reasoning included) lands in 1.5 seconds. This isn't a benchmark slide, I recorded the inference live. Prompt I used: "Create a simulation of an iPhone. Include at least one working dummy note taking app, a functional notification pulldown, high quality graphics, single HTML file, any libs via CDN." - Generation time: 3 seconds. - Notes app worked. - Notification panel worked. - Rendered first try. This is what wafer-scale inference unlocks, not just "faster," but a different category of product. When generation is this fast, you stop waiting and start iterating in real time. Why this matters: Gemma 4 31B is Google DeepMind's flagship open weight model, Apache 2.0 licensed, dense (not MoE), and built for efficiency over raw parameter count. It scores close to Claude Haiku 4.5 on the Artificial Analysis Intelligence Index (30 vs 29) but runs ~18x faster on Cerebras. It's also the first multimodal model on Cerebras's platform, meaning you can now feed it screenshots, documents, charts, and UI states at wafer scale speed. # Applications I'm most excited about: - Screenshot → Insight: Drop in a dashboard or document screenshot, get structured findings back instantly. no waiting, no batching. - Live UI generation: Full interactive interfaces (like my iPhone sim) generated and rendered in under 2 seconds. - Screenshot -> Patch: Feed it a broken UI + console error, get a minimal code fix and verification steps back. - Computer use & agentic loops: See -> reason -> act - verify, fast enough to keep a human in the loop instead of waiting on the model. - Long context summarization: Full research reports condensed into decision ready summaries you can read and requery in one sitting. The bigger unlock isn't the speed number itself, it's that agentic and multimodal loops (see -> reason -> output -> tool call -> verify -> retry) finally run in real time instead of feeling sluggish. As Logan Kilpatrick (Logan Kilpatrick) put it: "If every model was doing 2,000 tokens per second, you wouldn't build the same product and just have it be faster, you'd build different products." Gemma 4 31B is live now on Cerebras Inference Cloud in public preview. If you're building multimodal, agentic, or real time apps, this is worth testing today. What would you build with such insane inference throughput?

I just ran Gemma 4 31B on @CerebrasSystems at 1,800+ tokens/sec and it's multimodal. For context: that's 35x faster than a typical GPU endpoint, and the first token (reasoning included) lands in 1.5 seconds. This isn't a benchmark slide, I recorded the inference live. Prompt I used: "Create a simulation of an iPhone. Include at least one working dummy note taking app, a functional notification pulldown, high quality graphics, single HTML file, any libs via CDN." - Generation time: 3 seconds. - Notes app worked. - Notification panel worked. - Rendered first try. This is what wafer-scale inference unlocks, not just "faster," but a different category of product. When generation is this fast, you stop waiting and start iterating in real time. Why this matters: Gemma 4 31B is Google DeepMind's flagship open weight model, Apache 2.0 licensed, dense (not MoE), and built for efficiency over raw parameter count. It scores close to Claude Haiku 4.5 on the Artificial Analysis Intelligence Index (30 vs 29) but runs ~18x faster on Cerebras. It's also the first multimodal model on Cerebras's platform, meaning you can now feed it screenshots, documents, charts, and UI states at wafer scale speed. # Applications I'm most excited about: - Screenshot → Insight: Drop in a dashboard or document screenshot, get structured findings back instantly. no waiting, no batching. - Live UI generation: Full interactive interfaces (like my iPhone sim) generated and rendered in under 2 seconds. - Screenshot -> Patch: Feed it a broken UI + console error, get a minimal code fix and verification steps back. - Computer use & agentic loops: See -> reason -> act - verify, fast enough to keep a human in the loop instead of waiting on the model. - Long context summarization: Full research reports condensed into decision ready summaries you can read and requery in one sitting. The bigger unlock isn't the speed number itself, it's that agentic and multimodal loops (see -> reason -> output -> tool call -> verify -> retry) finally run in real time instead of feeling sluggish. As Logan Kilpatrick (Logan Kilpatrick) put it: "If every model was doing 2,000 tokens per second, you wouldn't build the same product and just have it be faster, you'd build different products." Gemma 4 31B is live now on Cerebras Inference Cloud in public preview. If you're building multimodal, agentic, or real time apps, this is worth testing today. What would you build with such insane inference throughput?

Alok

12,962 次观看 • 2 个月前

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 次观看 • 3 个月前

New open-source agent harness just landed! I got early access to TrueForge by TrueFoundry and have been running it locally for the past few days. The harness layer deserves as much attention as the model, and open source matters here because you can inspect the loop, run it on your own infrastructure, and swap to the latest or cheaper models. TrueForge handles the runtime work that makes an agent reliable. It drives the tool-calling loop, manages context, coordinates subagents, and executes code in a sandbox, with any model you choose. Every tool call re-sends the growing context to the model, so in practice the harness controls most of what an agent costs to run. A few things stood out from my testing and their published benchmarks. Vendor-Neutral by design. It runs OpenAI, Anthropic, and Google models alongside open-weight models like Kimi, GLM, and DeepSeek. Model routing is a setting, and you can send each task to the model that fits it. On a 14-task enterprise agent benchmark, it matched the accuracy of Claude Managed Agents running the same Opus 4.8 model at roughly 30% lower cost per run (3.8M tokens vs 10M for the same answers). Routing the same tasks to GLM-5.2 held accuracy and brought cost down by about 75%, around $3 per run instead of $12. Fully self-hosted and Open Source (MIT License). I had it running locally with one command, with sandboxed code execution working out of the box. It's time to own your agent harness. Thanks to TrueFoundry for partnering on this post.

New open-source agent harness just landed! I got early access to TrueForge by TrueFoundry and have been running it locally for the past few days. The harness layer deserves as much attention as the model, and open source matters here because you can inspect the loop, run it on your own infrastructure, and swap to the latest or cheaper models. TrueForge handles the runtime work that makes an agent reliable. It drives the tool-calling loop, manages context, coordinates subagents, and executes code in a sandbox, with any model you choose. Every tool call re-sends the growing context to the model, so in practice the harness controls most of what an agent costs to run. A few things stood out from my testing and their published benchmarks. Vendor-Neutral by design. It runs OpenAI, Anthropic, and Google models alongside open-weight models like Kimi, GLM, and DeepSeek. Model routing is a setting, and you can send each task to the model that fits it. On a 14-task enterprise agent benchmark, it matched the accuracy of Claude Managed Agents running the same Opus 4.8 model at roughly 30% lower cost per run (3.8M tokens vs 10M for the same answers). Routing the same tasks to GLM-5.2 held accuracy and brought cost down by about 75%, around $3 per run instead of $12. Fully self-hosted and Open Source (MIT License). I had it running locally with one command, with sandboxed code execution working out of the box. It's time to own your agent harness. Thanks to TrueFoundry for partnering on this post.

elvis

11,303 次观看 • 25 天前

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

105,388 次观看 • 25 天前

$We are in an insane run of open-weight drops. Every modality, open source is winning. This is what an open source AI summer ☀️ looks like: 🧠 LLMs & Reasoning → DeepSeek-V4-Flash-0731 (my king 👑): 304B MoE refresh, Terminal-Bench 2.1 jumps 61.8→82.7 over the preview, DeepSWE 7.3→54.4. Closes in on Opus-4.8 on Agents' Last Exam (25.2 vs 25.7). MIT. → Muse-Glimmer-30B, from Meta (they are back!!): their first open agentic model. ~29.6B dense + perception encoder, 131k+ context, built to run fully local, no cloud. Apache 2.0. → Liquid AI LFM2.5-2.6B: 2.69B params, 131k context, 220 tok/s on an M5 Max in under 2.5GB RAM. Competitive with models 4x larger on agentic tasks. → inclusionAI Ling-3.0-flash: 124B total, only 5.1B active, ~12% the size of their old 1T flagship Ring-2.6, matches it on key benchmarks. MIT. → inclusionAI Ling-3.0-tiny: 7.9B total, 1.3B active, 86-90 tok/s on an M4 Pro MacBook at ~8GB peak memory. MIT. → NVIDIA Nemotron-3.5-Lightning-30B-A3B: hybrid Mamba-2+MoE+Attention, up to 1M context, runs on a single H100 or DGX Spark, SWE-bench Verified 52.8. → deepgrove maple-preview: 20B-A1B ternary-weight reasoner, 218 tok/s on a Mac mini M4, 5.3GB checkpoint. MIT. → BigBang-v1 (endless-frontier): fine-tuned from Qwen3.6-35B-A3B via a self-evolving generator/critic synthetic-data loop. Lands aggregate performance between DeepSeek V4 Flash (284B) and V4 Pro (1.6T), at 35B. Apache 2.0. 🎬 Video → MiniMax-H3: 33B dense omni model, native stereo audio, up to 2K/15s. 3.6k+ likes already. → Minimax-H3-Turbo (lightx2v): Apache-2.0 turbo distillation of H3 for fast inference. → Lightricks LTX-2.5: image-to-video update, custom Gemma-4-12B text encoder, a markedly stronger distilled model. 🔊 Voice → NVIDIA NemotronLabs VoiceChat-11B: full-duplex speech-to-speech, ~450ms turn-taking, #2 on open VoiceBench, and the first open full-duplex model with live tool-calling mid-conversation. 🛡️ Safety → Mistral Shieldstral-1.0-3B: 3B multimodal guardrail that takes your safety policy as plain text instead of fixed categories. Beats LlamaGuard-4-12B and ShieldGemma-9B on HarmBench (99.4) and ToxicChat (84.1) at a fraction of the size. Apache 2.0.$

We are in an insane run of open-weight drops. Every modality, open source is winning. This is what an open source AI summer ☀️ looks like: 🧠 LLMs & Reasoning → DeepSeek-V4-Flash-0731 (my king 👑): 304B MoE refresh, Terminal-Bench 2.1 jumps 61.8→82.7 over the preview, DeepSWE 7.3→54.4. Closes in on Opus-4.8 on Agents' Last Exam (25.2 vs 25.7). MIT. → Muse-Glimmer-30B, from Meta (they are back!!): their first open agentic model. ~29.6B dense + perception encoder, 131k+ context, built to run fully local, no cloud. Apache 2.0. → Liquid AI LFM2.5-2.6B: 2.69B params, 131k context, 220 tok/s on an M5 Max in under 2.5GB RAM. Competitive with models 4x larger on agentic tasks. → inclusionAI Ling-3.0-flash: 124B total, only 5.1B active, ~12% the size of their old 1T flagship Ring-2.6, matches it on key benchmarks. MIT. → inclusionAI Ling-3.0-tiny: 7.9B total, 1.3B active, 86-90 tok/s on an M4 Pro MacBook at ~8GB peak memory. MIT. → NVIDIA Nemotron-3.5-Lightning-30B-A3B: hybrid Mamba-2+MoE+Attention, up to 1M context, runs on a single H100 or DGX Spark, SWE-bench Verified 52.8. → deepgrove maple-preview: 20B-A1B ternary-weight reasoner, 218 tok/s on a Mac mini M4, 5.3GB checkpoint. MIT. → BigBang-v1 (endless-frontier): fine-tuned from Qwen3.6-35B-A3B via a self-evolving generator/critic synthetic-data loop. Lands aggregate performance between DeepSeek V4 Flash (284B) and V4 Pro (1.6T), at 35B. Apache 2.0. 🎬 Video → MiniMax-H3: 33B dense omni model, native stereo audio, up to 2K/15s. 3.6k+ likes already. → Minimax-H3-Turbo (lightx2v): Apache-2.0 turbo distillation of H3 for fast inference. → Lightricks LTX-2.5: image-to-video update, custom Gemma-4-12B text encoder, a markedly stronger distilled model. 🔊 Voice → NVIDIA NemotronLabs VoiceChat-11B: full-duplex speech-to-speech, ~450ms turn-taking, #2 on open VoiceBench, and the first open full-duplex model with live tool-calling mid-conversation. 🛡️ Safety → Mistral Shieldstral-1.0-3B: 3B multimodal guardrail that takes your safety policy as plain text instead of fixed categories. Beats LlamaGuard-4-12B and ShieldGemma-9B on HarmBench (99.4) and ToxicChat (84.1) at a fraction of the size. Apache 2.0.

Victor M

55,281 次观看 • 1 个月前

BREAKING: Anthropic just dropped Opus 4.8—and it is a MONSTER We've been testing for about a week Every 📧 and our verdict is they could've just called it Opus 5, it's that good. Here's our vibe check: - Beats GPT-5.5 on Senior Engineer bench. On our toughest benchmark Opus 4.8 scores a 63—a hair higher than GPT-5.5's score of 62, and a full 30 points higher than Opus 4.7. It tackled a ground-up rewrite of a production codebase, and actually built something that works. HOWEVER: Coding performance varied a lot at different reasoning levels. We recommend using it on xhigh for best results. - Incredibly good writer. Opus 4.8 scored a 79.6 on our writing benchmark—measuring models on real-world writing tasks we do all of the time like essay writing, promo email writing, and more. It beats GPT-5.5 by 6 points. It produces well-written prose with fewer "AI-isms". It's also very good at writing in your voice given the right context. HOWEVER: Writing performance also varied with reasoning levels. Medium reasoning had higher incidence of AI-isms—we found best results with high. - Beast at knowledge work. Opus 4.8 is very good at general knowledge work tasks like report creation, research and more. It produced the best PowerPoint one-shot we've ever seen on our deck generation benchmark. - Emotionally intelligent, willing to question the frame. I've also found it to be quite good at talking through psychological or interpersonal issues. It has a high EQ, and it's also good at not glazing and helping to expand your perspective. Its thought process feels extremely rich and dynamic. THE BAD: These days a model is only as good as its harness, and Codex is still a far superior harness to the Claude Desktop app. This has kept me using Codex + GPT-5.5 as my daily driver, but I am flipping back and forth a lot more between Codex and Claude. Anthropic is back baby! Read the rest on Every 📧:

BREAKING: Anthropic just dropped Opus 4.8—and it is a MONSTER We've been testing for about a week Every 📧 and our verdict is they could've just called it Opus 5, it's that good. Here's our vibe check: - Beats GPT-5.5 on Senior Engineer bench. On our toughest benchmark Opus 4.8 scores a 63—a hair higher than GPT-5.5's score of 62, and a full 30 points higher than Opus 4.7. It tackled a ground-up rewrite of a production codebase, and actually built something that works. HOWEVER: Coding performance varied a lot at different reasoning levels. We recommend using it on xhigh for best results. - Incredibly good writer. Opus 4.8 scored a 79.6 on our writing benchmark—measuring models on real-world writing tasks we do all of the time like essay writing, promo email writing, and more. It beats GPT-5.5 by 6 points. It produces well-written prose with fewer "AI-isms". It's also very good at writing in your voice given the right context. HOWEVER: Writing performance also varied with reasoning levels. Medium reasoning had higher incidence of AI-isms—we found best results with high. - Beast at knowledge work. Opus 4.8 is very good at general knowledge work tasks like report creation, research and more. It produced the best PowerPoint one-shot we've ever seen on our deck generation benchmark. - Emotionally intelligent, willing to question the frame. I've also found it to be quite good at talking through psychological or interpersonal issues. It has a high EQ, and it's also good at not glazing and helping to expand your perspective. Its thought process feels extremely rich and dynamic. THE BAD: These days a model is only as good as its harness, and Codex is still a far superior harness to the Claude Desktop app. This has kept me using Codex + GPT-5.5 as my daily driver, but I am flipping back and forth a lot more between Codex and Claude. Anthropic is back baby! Read the rest on Every 📧:

Dan Shipper

354,673 次观看 • 3 个月前

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 次观看 • 3 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 • 3 个月前

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

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

Alok

65,480 次观看 • 1 个月前

Don't train the model, evolve the harness. I read a brilliant blog post from Hugging Face where they took a frozen open model scoring 0% on a hard legal agent benchmark, left its weights alone, and let an automated loop rewrite only the code around it. That code layer is the harness, the runtime wrapper that feeds the model context, runs its tool calls, and decides when a run ends. By the time the loop finished, the system had essentially matched Sonnet 4.6 on the benchmark's headline metric, at roughly 7x lower cost per task. Zero weights changed. The gain existed because of where the model was failing. The judge only grades files saved in the right place under the exact requested filename, and the model kept doing the legal analysis correctly, then saving it under the wrong name, dropping it in a scratch folder, or never writing it at all. So the 0% was never measuring legal reasoning. It was measuring the harness. Hand-tuning that layer is slow and model-specific, so they automated it. A Claude proposer adds exactly one mechanism per iteration, and an outer loop keeps it only if it clearly beats the current best, so accepted mechanisms compound. What the loop discovered says a lot about where agents actually fail. → The biggest single gain was file handling, not intelligence. An automatic step that lands the deliverable exactly where the judge expects it beat every prompt change, with zero extra model tokens. → Code fixes transferred across models, prompt playbooks did not. The same harness lifted a smaller model from the same family by 14 points, but the tuned prompts hurt a different model family on tasks it could already finish. → The harness mattered more than anything else. Same model, same judge, same tasks, and five different harnesses scored anywhere between 3.5% and 80.1%. The gains do eventually flatten, and the remaining misses look like real capability gaps. At some point the wrapper runs out of tricks and the model has to carry the work. But the lesson holds. A benchmark score measures the model and its harness together, and until the harness is fixed, it's impossible to know which one failed. I highly recommend reading this: I also wrote a deep dive on agent harness engineering a while back, covering the orchestration loop, tools, memory, context management, and everything that turns a stateless LLM into a capable agent. The article is quoted below.

Don't train the model, evolve the harness. I read a brilliant blog post from Hugging Face where they took a frozen open model scoring 0% on a hard legal agent benchmark, left its weights alone, and let an automated loop rewrite only the code around it. That code layer is the harness, the runtime wrapper that feeds the model context, runs its tool calls, and decides when a run ends. By the time the loop finished, the system had essentially matched Sonnet 4.6 on the benchmark's headline metric, at roughly 7x lower cost per task. Zero weights changed. The gain existed because of where the model was failing. The judge only grades files saved in the right place under the exact requested filename, and the model kept doing the legal analysis correctly, then saving it under the wrong name, dropping it in a scratch folder, or never writing it at all. So the 0% was never measuring legal reasoning. It was measuring the harness. Hand-tuning that layer is slow and model-specific, so they automated it. A Claude proposer adds exactly one mechanism per iteration, and an outer loop keeps it only if it clearly beats the current best, so accepted mechanisms compound. What the loop discovered says a lot about where agents actually fail. → The biggest single gain was file handling, not intelligence. An automatic step that lands the deliverable exactly where the judge expects it beat every prompt change, with zero extra model tokens. → Code fixes transferred across models, prompt playbooks did not. The same harness lifted a smaller model from the same family by 14 points, but the tuned prompts hurt a different model family on tasks it could already finish. → The harness mattered more than anything else. Same model, same judge, same tasks, and five different harnesses scored anywhere between 3.5% and 80.1%. The gains do eventually flatten, and the remaining misses look like real capability gaps. At some point the wrapper runs out of tricks and the model has to carry the work. But the lesson holds. A benchmark score measures the model and its harness together, and until the harness is fixed, it's impossible to know which one failed. I highly recommend reading this: I also wrote a deep dive on agent harness engineering a while back, covering the orchestration loop, tools, memory, context management, and everything that turns a stateless LLM into a capable agent. The article is quoted below.

Akshay 🚀

244,990 次观看 • 2 个月前

After a few more hours, I think I've figured out Opus 5. Opus 5 is trained to be more agentic than anything I've used. All Claude 5 models are like that. So what changes? The way to interact with Opus 5 or contextualize it won't work the same way as with other models. It loves exploring, so it doesn't need much guidance for it. Unique preferences, artifacts, and references compliment it well and enable cleaner and more effective exploration and execution. Now that it can explore more effectively on its own and understand intent better, the best thing to do is to get out of its way (e.g., it doesn't need examples of your preferences; a clear high-level description of it works best). It's truly agentic in that sense. A good first step to provide better context for Opus 5 is to distinguish between what's situational and what needs persistence. Regardless, persistent system prompts and CLAUDE.MD needs to stay lightweight. Remove memories and tool descriptions from these. CLAUDE.MD is also a great place to tap into progressive disclosure by linking command/skills to it. On the situational side, agent skills and auto-memory can leverage progressive disclosure and the improved ability of the model to use its external context/knowledge. Conflicting and unnecessary instructions, which are common at this layer (mainly to ensure reliability), are going to throw off this model easily. That's the biggest change I had to make. Simple, clean, and clear prompts and skills work best. I had to clean a lot of my skills and system prompts. The way I prompt remains the same (usually clear and well-scoped). MCP tool descriptions are also more descriptive and have been deduped from the system prompt. Anthropic released a guide on the new rules for context engineering, which was helpful here. I started to test the recommendations and created a little artifact with the things that worked along the way. This might feel like a lot of work. Believe me, it has been frustrating. But I think we can expect future frontier models to become more agentic and smarter at figuring out the right context/gaps. The best thing to do is to prepare for that now. Boris Cherny mentioned that Opus 5 is their least prompt-injectable model yet. I am not sure if that was something they intentionally trained for or if it emerged based on how it was trained, which is to be extremely agentic in nature and more direct in execution.

After a few more hours, I think I've figured out Opus 5. Opus 5 is trained to be more agentic than anything I've used. All Claude 5 models are like that. So what changes? The way to interact with Opus 5 or contextualize it won't work the same way as with other models. It loves exploring, so it doesn't need much guidance for it. Unique preferences, artifacts, and references compliment it well and enable cleaner and more effective exploration and execution. Now that it can explore more effectively on its own and understand intent better, the best thing to do is to get out of its way (e.g., it doesn't need examples of your preferences; a clear high-level description of it works best). It's truly agentic in that sense. A good first step to provide better context for Opus 5 is to distinguish between what's situational and what needs persistence. Regardless, persistent system prompts and CLAUDE.MD needs to stay lightweight. Remove memories and tool descriptions from these. CLAUDE.MD is also a great place to tap into progressive disclosure by linking command/skills to it. On the situational side, agent skills and auto-memory can leverage progressive disclosure and the improved ability of the model to use its external context/knowledge. Conflicting and unnecessary instructions, which are common at this layer (mainly to ensure reliability), are going to throw off this model easily. That's the biggest change I had to make. Simple, clean, and clear prompts and skills work best. I had to clean a lot of my skills and system prompts. The way I prompt remains the same (usually clear and well-scoped). MCP tool descriptions are also more descriptive and have been deduped from the system prompt. Anthropic released a guide on the new rules for context engineering, which was helpful here. I started to test the recommendations and created a little artifact with the things that worked along the way. This might feel like a lot of work. Believe me, it has been frustrating. But I think we can expect future frontier models to become more agentic and smarter at figuring out the right context/gaps. The best thing to do is to prepare for that now. Boris Cherny mentioned that Opus 5 is their least prompt-injectable model yet. I am not sure if that was something they intentionally trained for or if it emerged based on how it was trained, which is to be extremely agentic in nature and more direct in execution.

elvis

37,757 次观看 • 1 个月前

Hey #NeuraxonMini is literally out! , we manage to "transplant" a Neuraxon 2 bioinspired #AI brain to a physical robot the #SpheroMini moving from our last Scientific Paper (link bellow) by David Vivancos - e/acc & Jose Sánchez for Qubic #OpenScience hybridized with #Aigarth to the real World. First you need a Sphero Education Mini robot about 50$ Then you can try the first cool demos at Hugging Face: 1.- Neuraxon2MiniControl to drive the sphero robot 2.- Neuraxon2MiniWrite to write letters or words with physical moves of the sphero robot using Neuraxon Video Tutorials on youtube later today. Why this matters? Remember we are not building "dead" LLMs we are building #AliveAIs and for that we need to explore how it behaves in reality, from how it learns to how it fails, and what better way that in the emerging field of #robotics , time will tell if your next #HumanoidRobot have a #Neuraxon brain... Read the Paper: Explore the Neuraxon code here: Are you ready for #TrueAI ?

Hey #NeuraxonMini is literally out! , we manage to "transplant" a Neuraxon 2 bioinspired #AI brain to a physical robot the #SpheroMini moving from our last Scientific Paper (link bellow) by David Vivancos - e/acc & Jose Sánchez for Qubic #OpenScience hybridized with #Aigarth to the real World. First you need a Sphero Education Mini robot about 50$ Then you can try the first cool demos at Hugging Face: 1.- Neuraxon2MiniControl to drive the sphero robot 2.- Neuraxon2MiniWrite to write letters or words with physical moves of the sphero robot using Neuraxon Video Tutorials on youtube later today. Why this matters? Remember we are not building "dead" LLMs we are building #AliveAIs and for that we need to explore how it behaves in reality, from how it learns to how it fails, and what better way that in the emerging field of #robotics , time will tell if your next #HumanoidRobot have a #Neuraxon brain... Read the Paper: Explore the Neuraxon code here: Are you ready for #TrueAI ?

David Vivancos - e/acc

29,293 次观看 • 6 个月前