cogvlm2 official demo

[wsxwd]

here is the official demo, http://cogvlm2-online.cogviewai.cn:7861. I deploy the "THUDM/cogvlm2-llama3-chat-19B" model on my machine, but the result is worse than the online demo. Can you provide requirements and generation parameters in your web demo? requirements on my machine: requirements.txt

generation parameters: gen_kwargs = { "top_p": 0.4, "temperature": 0.8, "top_k": 1, "max_new_tokens": 2048, "pad_token_id": 128002, }

[zRzRzRzRzRzRzR]

gen_kwargs = { "max_new_tokens": 2048, "pad_token_id": 128002, "top_k": 1, }

[wsxwd]

gen_kwargs = { "max_new_tokens": 2048, "pad_token_id": 128002, "top_k": 1, }

Great thanks! Can you share requirements.txt in your web demo?

[FurkanGozukara]

@zRzRzRzRzRzRzR so we dont provide temperature and top_p or they are default set to 1?