- 7B and 20B: these models are really lightweight, how it applies to he field of semantic segmantation or the perception of Autonomous driving is an interesting topic, cause current models are difficult for the product implementation
- Code ability similar to chatgpt, it should be gpt3.5, as for is the code generation ability and code interpretation ability compared with github copilot should wait for the follow-up use of experience
- It can call tools to achieve retrieval, and its functions are richer than chatgpt(not plus). Maybe it can help me read paper
- It is necessary to investigate whether it can be applied in the field of perception in autonomous driving
- fine tune
- support flash attention for acceleration
- 可以关注一下
- provide a full-flow solution for deploying large models on GPU(nice work!)