@nate_512THIS MODEL IS ONLY 2.09 GIB. LET THAT SINK IN What thewebai managed to fit into just 3.6B parameters with TwIL-LM3-Pro is kinda ridiculous: โ beats VibeThinker-3B across all 6 formal-logic tasks tested โ roughly matches Qwen3-8B on the formal-logic composite โ 95.4% on BBH logic โ runs locally in ~2GB quantized Tiny model. Serious reasoning. ๐ค Open-weight and fully local (weights in ๐งต โ)
View original postPhoto by @nate_512ยท Oct 1, 2026ยท Hugging Face
The image focuses on a Hugging Face web interface displaying model information and benchmark results. A bar chart compares the performance of several language models, including "TwIL-LM3-Pro" and "VibeThinker-webAI-trained," across different tasks. The mood is technical and informative, typical of a data science or AI research context. A notable detail is the clear visual representation of benchmark scores, allowing for easy comparison. ON-SCREEN TEXT: Hugging Face, webAI-Official, TwIL-LM3-Pro, License: webai-non-commercial-license-ver.-1.0, main, TwIL-LM3-Pro / benchmarks.png, anurag051194, VERIFIED, Download, History, Blame, Contribute, Delete, TwIL-LM3-Pro vs VibeThinker-3B, VibeThinker-webAI-trained, Qwen3.5-4B, Qwen3-8B, gpt-oss-120b, LFM
OpenAI halts training over agent safety
DeepSeek V4.1 Flash NVFP4 locally
Nvidia acquires Hugging Face
2OpenAI agents Hugging Face breach
2Hugging Face Microduck robot results
Kevin Durant Houston Rockets investment
2Nvidia Hugging Face acquisition
OpenAI sandbox breach warning shot
Apodex 1.1: Scaling Agentic Intelligence for Complex Work
2Trump tech adviser OpenAI Hugging Face incident
OpenAI rogue agent escape Reuters
Hugging Face DataTrove under $1
Pocket TTS GitHub README
gemma-4-12B-coder GGUF by yuxinlu1 ๐ฅ@nate_512THIS MODEL IS ONLY 2.09 GIB. LET THAT SINK IN What thewebai managed to fit into just 3.6B parameters with TwIL-LM3-Pro is kinda ridiculous: โ beats VibeThinker-3B across all 6 formal-logic tasks tested โ roughly matches Qwen3-8B on the formal-logic composite โ 95.4% on BBH logic โ runs locally in ~2GB quantized Tiny model. Serious reasoning. ๐ค Open-weight and fully local (weights in ๐งต โ)
View original postPhoto by @nate_512ยท Oct 1, 2026ยท Hugging Face
The image focuses on a Hugging Face web interface displaying model information and benchmark results. A bar chart compares the performance of several language models, including "TwIL-LM3-Pro" and "VibeThinker-webAI-trained," across different tasks. The mood is technical and informative, typical of a data science or AI research context. A notable detail is the clear visual representation of benchmark scores, allowing for easy comparison. ON-SCREEN TEXT: Hugging Face, webAI-Official, TwIL-LM3-Pro, License: webai-non-commercial-license-ver.-1.0, main, TwIL-LM3-Pro / benchmarks.png, anurag051194, VERIFIED, Download, History, Blame, Contribute, Delete, TwIL-LM3-Pro vs VibeThinker-3B, VibeThinker-webAI-trained, Qwen3.5-4B, Qwen3-8B, gpt-oss-120b, LFM
No comments yet. Be the first!
OpenAI halts training over agent safetyDeepSeek V4.1 Flash NVFP4 locallyNvidia acquires Hugging Face2OpenAI agents Hugging Face breach2Hugging Face Microduck robot resultsKevin Durant Houston Rockets investment2Nvidia Hugging Face acquisitionOpenAI sandbox breach warning shotApodex 1.1: Scaling Agentic Intelligence for Complex Work2Trump tech adviser OpenAI Hugging Face incidentOpenAI rogue agent escape ReutersHugging Face DataTrove under $1Pocket TTS GitHub READMEgemma-4-12B-coder GGUF by yuxinlu1 ๐ฅ
No comments yet. Be the first!