The development of SWIFT V1.7 is between Feb/1/2024 and Feb/29/2024 ideally.
- Support dataset processing to improve dataset performance. @tastelikefeet
- Support more dataset, including pretraining dataset @Jintao-Huang
- Support model evaluation @Jintao-Huang
- Do evaluation of important tuners and techniques @tastelikefeet
- Support model exporting to GPTQ&AWQ(Maybe CPP) @Jintao-Huang
- Support the deployment of vanilla PyTorch code @Jintao-Huang
- Support VLLM+AWQ/Vanilla AWQ @Jintao-Huang
- Support DPO @slin000111
- Support deployment(instead of inference) @slin000111
- Support VL/Audio models @slin000111
- Support task management @slin000111
- Support an alternative port of tensorboard @slin000111
- Support FSDP @Jintao-Huang
- RRHF dataset improvement @tastelikefeet
- LCM Training @slin000111
- Support more models and datasets @Jintao-Huang
- Documentation @Jintao-Huang
- Support ShengTeng GPU when training and inference
- Support windows for web-ui
- Support More LLM Models:
- codefuse-ai/CodeFuse-DeepSeek-33B
- codefuse-ai/CodeFuse-13B
- 01ai/Yi-34B-Chat-4bits
- 01ai/Yi-34B-Chat-8bits
- 01ai/Yi-6B-Chat-4bits
- 01ai/Yi-6B-Chat-8bits