English Version

New Features:

1. RLHF reconstruction, supporting all integrated multimodal models, compatible with DeepSpeed Zero2/Zero3, and supports lazy_tokenize.
2. Using infer_backend vllm, inference deployment of multimodal large models supports multiple images.

New Models:

1. Qwen2.5 series, Qwen2-vl-72b series (base/instruct/gptq-int4/gptq-int8/awq)
2. Qwen2.5-math, Qwen2.5-coder series (base/instruct)
3. Deepseek-v2.5

New Datasets:

1. longwriter-6k-filtered

中文版

新特性：

1. RLHF重构，支持所有已接入的多模态模型，兼容deepspeed zero2/zero3，支持lazy_tokenize
2. 使用infer_backend vllm，推理部署多模态大模型支持多图.

新模型：

1. qwen2.5系列、qwen2-vl-72b系列（base/instruct/gptq-int4/gptq-int8/awq）
2. qwen2.5-math, qwen2.5-coder系列（base/instruct）
3. deepseek-v2.5

新数据集：

1. longwriter-6k-filtered

What's Changed

• fix model_mapping by @Jintao-Huang in https://github.com/modelscope/ms-swift/pull/1982
• fix patch by @Jintao-Huang in https://github.com/modelscope/ms-swift/pull/1997
• fix by @tastelikefeet in https://github.com/modelscope/ms-swift/pull/1995
• Support Deepseek 2.5 by @DaozeZhang in https://github.com/modelscope/ms-swift/pull/1992
• fix EngineGenerationConfig importError of lmdeploy by @irexyc in https://github.com/modelscope/ms-swift/pull/1990
• compat lmdeploy==0.6 by @Jintao-Huang in https://github.com/modelscope/ms-swift/pull/2001
• Fix rlhf ref model by @Jintao-Huang in https://github.com/modelscope/ms-swift/pull/2003
• Support llava1.6-llama3.1-8b-instruct by @DaozeZhang in https://github.com/modelscope/ms-swift/pull/2005
• fix lmdeploy qwen_vl by @Jintao-Huang in https://github.com/modelscope/ms-swift/pull/2009
• Add FAQ Document by @slin000111 in https://github.com/modelscope/ms-swift/pull/2013
• Florence use _post_encode & template support encoder-decoder by @Jintao-Huang in https://github.com/modelscope/ms-swift/pull/2019
• refactor rlhf by @Jintao-Huang in https://github.com/modelscope/ms-swift/pull/1975
• update code by @Jintao-Huang in https://github.com/modelscope/ms-swift/pull/2028
• fix deploy eval kill by @Jintao-Huang in https://github.com/modelscope/ms-swift/pull/2029
• Fix olora and pissa saving files which will cause the second saving failed by @tastelikefeet in https://github.com/modelscope/ms-swift/pull/2032
• fix rlhf & zero3 by @Jintao-Huang in https://github.com/modelscope/ms-swift/pull/2034
• Add longwriter filtered dataset by @wangxingjun778 in https://github.com/modelscope/ms-swift/pull/2037
• fix mplug-owl3 by @Jintao-Huang in https://github.com/modelscope/ms-swift/pull/2042
• support multi bbox grounding by @tastelikefeet in https://github.com/modelscope/ms-swift/pull/2045
• Fix multi coordinate grounding by @tastelikefeet in https://github.com/modelscope/ms-swift/pull/2047
• llama3 tool calling by @tastelikefeet in https://github.com/modelscope/ms-swift/pull/2048
• update docs by @Jintao-Huang in https://github.com/modelscope/ms-swift/pull/2050
• fix qwen2vl position_ids by @Jintao-Huang in https://github.com/modelscope/ms-swift/pull/2051
• support qwen2-vl-base by @Jintao-Huang in https://github.com/modelscope/ms-swift/pull/2052
• Support qwen2.5 by @Jintao-Huang in https://github.com/modelscope/ms-swift/pull/2054
• support qwen2-vl -72b/qwen2.5-math/qwen2.5-coder by @Jintao-Huang in https://github.com/modelscope/ms-swift/pull/2056
• vllm support mutli image by @Jintao-Huang in https://github.com/modelscope/ms-swift/pull/2059
• support qwen2.5-coder by @Jintao-Huang in https://github.com/modelscope/ms-swift/pull/2061
• fix notebook gradio by @Jintao-Huang in https://github.com/modelscope/ms-swift/pull/2062
• update qwen2-vl docs by @Jintao-Huang in https://github.com/modelscope/ms-swift/pull/2063

New Contributors

• @irexyc made their first contribution in https://github.com/modelscope/ms-swift/pull/1990
• @wangxingjun778 made their first contribution in https://github.com/modelscope/ms-swift/pull/2037

Full Changelog: https://github.com/modelscope/ms-swift/compare/v2.4.1...v2.4.2