World Tour Finals 2026 Algorithm: Human vs. AI Exhibition Matchatcoder.jp2 points·by limoce·3 दिन पहले·0 comments
China's DeepSeek developing its own AI chip, sources sayreuters.com11 points·by limoce·5 दिन पहले·0 comments
China Blocks Meta's $2B Acquisition of AI Firm Manusbloomberg.com19 points·by limoce·3 माह पहले·9 comments
Declining to review for NeurIPS due to policy on sanctioned institutionspastebin.com5 points·by limoce·4 माह पहले·0 comments
Classical Chinese Jailbreak Prompt Optimization via Bio-Inspired Searcharxiv.org3 points·by limoce·4 माह पहले·0 comments
Cursor's 'Composer 2' model is apparently just Kimi K2.5 with RL fine-tuningold.reddit.com1 points·by limoce·4 माह पहले·1 comments
SteptronOss: Lightweight, AI-native training framework for large language modelsgithub.com1 points·by limoce·4 माह पहले·0 comments
Reduce Complexity of Matmul to O(N^2logN) by Asymptotically Optimal Quantum Algoarxiv.org4 points·by limoce·5 माह पहले·1 comments
Learning from context is harder than we thoughthy.tencent.com245 points·by limoce·5 माह पहले·136 comments
LongCat-Flash-Lite 100B A3B Technical Report [pdf]huggingface.co2 points·by limoce·5 माह पहले·0 comments
Jinja.cpp: A single-header C++11 Jinja2 template engine for LLM chat templatesgithub.com1 points·by limoce·7 माह पहले·0 comments
limoce·4 माह पहले·discussThe title should be "Adaptive Test-time Learning and Autonomous Specialization".
limoce·5 माह पहले·discuss> This site is an independent informational resource and is not officially affiliated with Zhipu AI.
limoce·6 माह पहले·discussI think separating thinking tokens from "representing" tokens might be a better approach, like what those thinking models does
limoce·9 माह पहले·discuss> ollama gpt-oss 120b mxfp4 1 94.67 11.66This is insanely slow given its 200+GB/s memory bandwidth. As a comparison, I've tested GPT OSS 120B on Strix Halo and it obtains 420tps prefill and >40tps decode.