Model request: Hermes-3-Llama-3.2-3B

[yyjhao]

https://huggingface.co/NousResearch/Hermes-3-Llama-3.2-3B

There's Hermes-3-Llama-3.1 but 3.2 is better

[FriskyFennecFox]

It seems that there is a distinct lack of 3B models. There's Llama-3.2-3B-Instruct which is notorious for its refusals, making it unsuitable for the cases that require the model to always return expected generations. Other than that, there's only Qwen2.5-3B-Instruct and Qwen2.5-Coder-3B-Instruct, with the latter being a more niche-focused model.

The proposed model should fill this gap well. NousResearch is a well-known player in the LLM field and the Hermes 3 series of models have been researched and developed with enough care to be considered production-ready.

For a project with a mission to deliver in-browser LLMs, I think it's a must to add more official support for the smaller models.

[CharlieFRuan]

Thanks for the request, added in npm 0.2.78:

• [Version] Bump version to 0.2.78 #653

[CharlieFRuan]

It is also now live on https://chat.webllm.ai/