Index-Homura
#2 in KI-ÖversettenBilibili · siet 30. September 2026 · 2× · tolest 02. Okt. 2026
100
Momentum
Index-Homura is an AI translation model developed by bilibili's Index LLM team, built on Index-Translate-9B, that adjusts translations toward a specified target syllable count to produce scripts better suited to a dubbing time window. It uses a reinforcement-learning approach (GRPO, building on the HOMURA framework) that jointly optimizes translation quality and adherence to the specified syllable count. It currently supports Chinese input into four target languages and is released as part of the open-source Index-Translate model family (2B and 9B variants) under the Apache-2.0 license.
Momentum-Verloop
04.07.02.10.
Features
| License | Apache-2.0 |
| Platform | Hugging Face, ModelScope, GitHub (inference script syllable_translate.py) |
| Price | Open-source / free (model weights freely available) |
| Release Date | September 30, 2026 |
| Languages | Chinese source language into 4 target languages (incl. English, Japanese, Spanish) |