openbmb

/

MiniCPM3-4B

like 394

Follow

OpenBMB 448

@neoz
Hi, Zhi,
We are trying to enable minicpm3 in intel Habana Gaudi accelerator, and since it's a remote code model, we rely on model_type field to recognize the model and optimize it to peak performance, so we propose to add model_type field in config, it's a common practice in hugging face models like llama(https://huggingface.co./meta-llama/Meta-Llama-3.1-8B/blob/main/config.json#L15), decilm(https://huggingface.co./Deci/DeciLM-7B/blob/main/config.json#L16), deepseek-v2(https://huggingface.co./deepseek-ai/DeepSeek-V2-Chat/blob/main/config.json#L22) etc.

Thx very much for review.

LGTM. Thank you for your contribution.