Hojo-ASR-Multi-V1 model card: Qwen3 decoder architecture, Apache 2.0 open weights on HF
rohanpaul_ai · x · 2026-09-04
The Hugging Face model card for Hojo-ASR-Multi-V1 reveals technical details and usage: an Encoder-Adapter-LLM architecture built on the Qwen3 LLM decoder with multi-frame acoustic fusion, optimized via multi-stage modular training and RL, performing well in noisy and conversational conditions across major languages. Apache 2.0 licensed, installable via the hojo-asr pip package for quick inference.
Related event: Hojo-ASR Tops Open Multilingual ASR Leaderboard with 3.54% WER(2 posts)→
More from Multimodal
- Runway joins Claude Marketplace as its first creative partner — tlakomy · 2026-10-21
- Claude Code directed a full music video: 6.5 hours, 765M input tokens — colinmcnamara · 2026-10-21
- Kling 4.0 debuts native 30-second AI video with 10 keyframes, extendable to 2 minutes — lmoroney · 2026-10-21
- Mistral's Voxtral transcribes an 1891 Eiffel phonograph recording at 6.5% WER — IsForAt · 2026-10-21
- Redditor remakes SD clown dance video at 25 steps with motion boost and fixed prompts — roychodraws · 2026-10-21
- Fibo-2, the #1 open image editing model, releases weights on Hugging Face — linoy_tsaban · 2026-10-21