no 7B model size?

[yiyepiaoling0715]

what is the reason for no avaliable model less than 7B?
which model size in the prod env ?

[yiyepiaoling0715]

infer latency?

[twelveand0]

I'm not quite clear on your question yet. Could you provide more details, including the exception stack, startup configuration, and training configuration? MFTCoder does not have specific size restrictions for models that have already been adapted. The currently supported models include llama2, codellama, baichuan, qwen1, chatglm2, among others. Notably, the codeqwen_competition branch can support qwen2, which includes both qwen-1.8b and 72b.