Nemotron 3.5 Transcribe Collection Open-weight offline ASR model supporting 40 languages, 1-hr audio, and 64k context. Hybrid MoE (~4.3B active). Available in BF16, FP8, and NVFP4. • 7 items • Updated 1 day ago • 11
UBTree: Parallel Tree Drafting via Unigram and Bigram Models for Speculative Decoding Paper • 2609.39972 • Published Sep 30 • 41
deepseek-ai/DeepSeek-V4.1-Flash-1017 Image-Text-to-Text • 763B • Updated 14 days ago • 146k • • 4.02k
Selecting Diverse SFT Traces Improves Post-RL Generalization Paper • 2609.33780 • Published Sep 27 • 38
Running 181 HF Viewer · Model Architecture Explorer 🟩 181 Interactive architecture graph for any HF model
DeepSeek Elastic Compute (DSec): A Sandbox Infrastructure for Effective Agentic Training at Scale Paper • 2609.22978 • Published Sep 19 • 4