tencent/Hy-SimulMT-30B-A3B-FP8
AI model · https://huggingface.co/tencent/Hy-SimulMT-30B-A3B-FP8
First seen by Oriel on .
AI summary
Hy-SimulMT is Tencent Hunyuan’s Chinese-English simultaneous speech translation model, designed to translate streaming audio with low lag and optionally clone the speaker’s voice.
- On the RealSI long-form benchmark, it achieved approximately 3 seconds StreamLAAL, the lowest among systems shown, and a GEMBA score above compared open-source models and four of five commercial APIs.
- Across four short-form benchmarks, Hy-SimulMT scored highest among listed systems on RealSI-Sentence and outperformed other open-source models on FLEURS, ACL6060, and RealSI-Sentence.
- A small comparison using 18 recordings and six professional interpreters found translation quality on par with expert interpreters, often at lower latency; the authors caution that this does not constitute a comprehensive assessment.
- Inference is provided through SimulInfer, with vLLM deployment instructions and optional TTS; GGUF support for llama.cpp is described as unmerged, with server support still planned. Voice cloning requires a configured TTS service.
#speech translation #simultaneous interpretation #vllm #llama.cpp
Generated by gpt-5.6-sol. Check the original for details.
Metric trends
Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained
Likes
Total likes reported by the source.
- Hugging Face Models: 457, latest observation ; +343 over observed period since
Downloads (30d)
Rolling 30-day downloads. A change in this value is not new downloads during the observed period.
- Hugging Face Models: 396, latest observation ; +303 over observed period since
Aliases
| Source | External ID | URL |
|---|---|---|
| Hugging Face Models | tencent/Hy-SimulMT-30B-A3B-FP8 | https://huggingface.co/tencent/Hy-SimulMT-30B-A3B-FP8 Read discussion |