Hunyuan 4B Pretrain
Hunyuan 4B Pretrain is the intermediate size in Tencent’s dense language-model series. It is the pretrained checkpoint for further adaptation, distinct from the instruction-tuned release.
Model design and access
The family uses training strategies related to Hunyuan-A13B. Its 4B size is a dense parameter count, not the active count of a larger expert network. Instruction-following behavior must be supplied through subsequent training; a chat template alone does not add that behavior.Hunyuan 4B Pretrain official model card
Practical use
Practical guidance. Use this checkpoint when the training objective needs control over the model’s response style. Establish a held-out task review before and after adaptation, including examples outside the training domain.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| License | Tencent Hunyuan Community License; excludes the EU, UK and South KoreaHunyuan dense model community license |
| Architecture | Dense transformer with grouped-query attentionHunyuan 4B Pretrain official model card |
| Context | 256K tokensHunyuan 4B Pretrain official model card |
Related models
- Hunyuan 4B Instruct
Tencent · Hunyuan dense
A 4B dense instruction-tuned language model with fast and deliberate reasoning modes.
Launch date not confirmedDownloadable