bilibili/Index-Translate
Bilibili released a Qwen3.5-based translation model family spanning text, speech, syllable-controlled output, and long-document translation.
The text models cover 150 languages and come in 2B, 9B, and 35B-A3B preview checkpoints. The repo also lists Index-Echo for translated subtitles or voice-conditioned speech, Index-Homura for target syllable counts, and Index-NativeLong for full-document context. Official quantized builds are available in GGUF, FP8, and Blackwell-optimized NVFP4 across the family. Four related benchmarks were released with datasets or metadata on Hugging Face and evaluation scripts on GitHub. GitHub · LLM repos' note
The text models cover 150 languages and come in 2B, 9B, and 35B-A3B preview checkpoints. The repo also lists Index-Echo for translated subtitles or voice-conditioned speech, Index-Homura for target syllable counts, and Index-NativeLong for full-document context. Official quantized builds are available in GGUF, FP8, and Blackwell-optimized NVFP4 across the family. Four related benchmarks were released with datasets or metadata on Hugging Face and evaluation scripts on GitHub. GitHub · LLM repos' note
score 6