Codacus @thecodacus

@Michael90466759 @meloiiy weight has no effect on which one will drop faster..

@mehulmpt it really furstrates me so much, why govt websites are so poorly made. just hire some good devs and put some efforts into it.. just having a website is not the end goal. some of the websites are soo bad that when you visit them you have no idea what to do next..

@CompleteSkeptic yess exactly.. half of the people on internet is confused what it is and how is it different than an llm or just an custom classifier

This looks awesome!! thanks for such contributions

Harsha Gundala @harshagundal

2 days ago

I’m back, with a little something… GPC 1, a new general-purpose classifier enabling: bounding boxes, poses, coordinates, direct mechanical control and so much more, all in milliseconds. Post-trained on millions of datapoints and thousands of unique problem classes learned

21 65 708 42K 777

@Suyash151504 the subsequent queries will be even faster after the 1st query. if you keep the system prompt same just change the context

Half if the people is still confused about JEV. So this is now my favorite way to consume AI news away from Hype 🤣

Prince does AI @princedoesai

3 days ago

Holy crap. Rick and Morty just explained Jev AI to me better than any tech demo could.

253 1K 12K 1.6M 7K

Qwen3.8-27B spent 21 days trying to build its own CUDA inference engine on a single RTX 3090. Mostly unsupervised. No human-written CUDA. Only around 12 human nudges across the entire experiment. The model wasn’t just given one giant prompt and left alone either. DeepSeek Harness managed the whole operation: Subagents. Roles. Handoffs. Context management. Compaction. HyperQwen served Qwen3.8-27B through vLLM while the agents worked on the inference engine. And the project accumulated some serious numbers over those three weeks: 180 subagents 230M tokens processed 1.7B cache-read tokens 699 context compactions 83 hours spent on context compaction The final engine didn’t beat llama.cpp. That’s important. HyperQwen managed roughly 250 prefill tok/s compared with around 700 tok/s for llama.cpp in the reported comparison. So this isn’t a story about an AI suddenly inventing a better CUDA runtime. It’s a story about how far you can push a local 27B model when you give it an engineering environment, persistent state and enough time. The funniest part was what happened when the agent started testing its own work. The RTX 3090 has 24GB of VRAM, and the same GPU needed to run both Qwen3.8-27B through vLLM and the CUDA engine being tested. That meant the workflow had to repeatedly: Stop vLLM. Run the benchmark. Restart vLLM. Continue the project. Then one subagent ignored the handoff protocol and killed vLLM while the orchestrator was still running. It basically shut down its own brain. A genuine AI suicide loop. And yet the larger system recovered and continued the project. That part is more interesting to me than whether the resulting inference engine beat llama.cpp. A quantized Qwen3.8-27B model maintained a real software engineering project for three weeks on consumer hardware. It spawned 180 subagents. Processed 230M tokens. Compacted its own context 699 times. And required only about a dozen human interventions. That’s a very different kind of benchmark. We’re starting to move from: “Can the model write code?” to: “Can the model maintain a software project long enough to actually get somewhere?” The answer here wasn’t “it built a better inference engine.” It didn’t. But it did demonstrate that a local Qwen3.8-27B can participate in a surprisingly long-running engineering workflow without a human sitting there writing the implementation. One clarification because the attached video can be misleading: The video is NOT the CUDA engine produced during the 21-day experiment. It compares HyperQwen, the inference backend used to run Qwen3.8-27B during the experiment, against stock vLLM on the same RTX 3090 using the same prompts. The 21-day experiment and the video comparison are separate things. The engine lost the speed race. The agent endurance experiment is still pretty wild. x.com/jurlycat/statu…

"Can You Run Any LLM in Jev Mode Using llama.cpp?" @thecodacus "Jev answers in milliseconds, picks from your options instead of writing a paragraph... I added the same way of asking to llama.cpp, so it runs on the model already sitting on your drive..." youtu.be/bcGO7xre46o

interesting idea

Aaron @A502404S

5 days ago

#ai anyone tried using connectome and fruit fly neurons mappings with a local LLM like qwen 3.8 27b or qwen 3.6 35B yet? @thecodacus

0 0 1 155 0

Today, we’re announcing Ternary Bonsai 2 27B. Based on Qwen3.8 27B, Bonsai 2 27B is 9x smaller than its full-precision counterpart while retaining 98.2% of its aggregate benchmark performance. Two months after the first Bonsai 27B release, the biggest change is quality. The footprint remains 5.9 GB, but the gap to full precision has narrowed materially, with particularly strong gains in agentic coding, multimodal reasoning, and long-horizon tool use. Ternary Bonsai 2 27B is available today under Apache 2.0.

Fast meets open. 🚀 Qwen3.8-27B is now running on @cerebras with rapid inference. Try it now!

Cerebras @cerebras

2 weeks ago

Qwen3.8-27B is now live at Cerebras speed. The dense, open-weight model from @Alibaba_Qwen scores 34 on the Artificial Analysis Intelligence Index—making it comparable to models such as GPT-5.6 Luna, DeepseekV4 Pro, and Claude Sonnet 4.6.

90 82 1K 291K 372

@mehulmpt I think instead of releasing their smls they should just use open weight models. and once their in house model is comparable to other open weights model at least in the same parameter count class they should release them.. otherwise its just embarrassing

@shuabe but your prefill much be much higher them mine.

@shuabe 3060 actually 🙂

we have seen the video where Anthropic finds emotion directions inside Claude and nudges them to change its behaviour. Thought that was cool, so I tried it on Gemma 4 12B, little modification to the llama.cpp, and here we go

keyboard_arrow_left Previous keyboard_arrow_right Next

"Is Frontier Class Local AI Finally Practical?" Qwen 3.8 Flash Next: "A 177-billion-parameter model, running on a five-year-old RTX 3060 with 12 GB of VRAM. Level with Claude Opus 4.8, going toe to toe with Opus 5 ..." @thecodacus @YouTube youtu.be/IH8XmxiwliQ

@2ndGenAmerica I have not tried fine tuning yet. but unsloth has everything covered on this already i believe

Finally.. hoping its a sub 100b parameter model 🤞

Daniel Han @danielhanchen

a month ago

Qwen announces Qwen3.8-Flash-Next, a new open-weight multimodal MoE model. 💜 The model will be released tomorrow and we are working on @UnslothAI day zero support. Qwen3.8-Flash-Next is built on the new Qwen4 architecture.

29 67 755 63K 149

@acleitao if you are trying multi gpu, them I have not tested my fork with multi gpu setup. just fyi also you can go till 64k context with this command.. 16k is just for testing I used

@acleitao yes mtp works in there, but i had to put it on cpu instead of gpu, because of the limited vram. this is the full command I used for my setup. you need to generate the static expert profile for caching them on gpu. the read me has the details

⚡Meet Qwen3.8-Flash, a multimodal MoE and an early preview of the Qwen4 architecture, now open-weight! The production version Qwen3.8-Flash will be available soon via QwenCloud API at just $ 0.16/1M input tokens and $ 0.47/1M output tokens. 125B parameters + 51B N-gram embeddings, with just 6B activated per token. Unmatched cost-efficiency. What's new: 🥳 - Next architecture: GDN + QSA hybrid attention, Gated Residual, N-gram Embedding & Muon optimizer, serving as a precursor to the architecture used in Qwen4. - Dramatically lower training and inference costs: trained at just 1/9 the cost of Qwen3.7-Plus, while outperforming it across the board with especially strong gains in coding and office tasks. - Strong performance: scoring 58.7 on DeepSWE 1.1, 62.5 on SWE-bench Pro, 73.9 on CoWorkBench, 84.5 on AndroidWorld, and 95.7 on MathVision (with CI). - 262K native context, extensible to 1M with YaRN. We’re also releasing the weights for Qwen3.8-Flash-Next, giving the community an early look at the new architecture we’re exploring for Qwen4.🚀 We can't wait to see what you build with Qwen3.8-Flash!👀👇 - Blog: qwen.ai/blog?id=qwen3.… - Technical Report: github.com/QwenLM/Qwen3.8… - Hugging Face: huggingface.co/Qwen/Qwen3.8-F… - ModelScope: modelscope.cn/models/Qwen/Qw…