Hunyuan-A13B Pretrain
Hunyuan-A13B Pretrain has 80B total parameters and activates 13B per token. This is the base training checkpoint in the A13B family.
Model design and access
The model uses fine-grained expert routing to reduce active computation. Tencent provides inference instructions for TensorRT-LLM, vLLM and SGLang. The full set of experts still determines weight storage. The pretrained and instruction-tuned checkpoints should remain separately identified.Hunyuan-A13B Pretrain official model card
Practical use
Practical guidance. Select this release for continued pretraining or custom post-training. Confirm that the adaptation improves the target task without removing useful general language behavior.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| License | Tencent Hunyuan Community License; excludes the EU, UK and South KoreaHunyuan A13B community license |
| Context | 256K tokensHunyuan-A13B Pretrain official model card |
| Attention | Grouped-query attentionHunyuan-A13B Pretrain official model card |
Related models
Hy3 is a 295B mixture-of-experts language model with 21B active parameters and a separate 3.8B multi-token prediction layer.
Launch date not confirmedDownloadable · API- Hunyuan-A13B Instruct
Tencent · Hunyuan-A13B
An instruction-tuned mixture-of-experts language model with 80B total and 13B active parameters.
Launch date not confirmedDownloadable