Model reference · open weights

3b-zh-pretrain-research_release

3b-zh-pretrain-research_release is an open-weight audio or speech model from canopylabs. 3b-zh-pretrain-research_release (FP32) weighs 7.6 GB; the smallest configuration that runs it is RTX 3060 12 GB.

• 3b-zh-pretrain-research_release is a 3.8B parameter Llama-based speech model developed by canopylabs for text-to-speech generation.
• It supports Chinese and is designed for tasks including high-quality speech synthesis and zero-shot voice cloning.
• The model is released under the llama3.2 licence.

Summary of the canopylabs/3b-zh-pretrain-research_release model card, 2026-10-01

Hugging Face ↗