English Version

New Features:

1. ReFT support: achieves parameter efficiency that is 15× to 65× greater than LoRA.
2. Multimodal model supports zero3.
3. Supports using environment variables to control parameters such as hd_num, max_num, and video_segments.

New Models:

1. longwriter-glm4-9b, longwriter-llama3_1-8b
2. phi3_5-mini-instruct, phi3_5-moe-instruct, phi3_5-vision-instruct
3. llava-onevision-qwen2-0_5b-ov, llava-onevision-qwen2-7b-ov, llava-onevision-qwen2-72b-ov

New Datasets:

1. longwriter-6k
2. rlaif-v
3. latex-ocr-print, latex-ocr-handwrite

中文版

新功能：

1. 支持ReFT，实现了比 LoRA 高 15 倍到 65 倍的参数效率。
2. 多模态模型支持 zero3。
3. 支持使用环境变量控制模型特有的参数，如 hd_num、max_num 和 video_segments。

新模型：

1. longwriter-glm4-9b, longwriter-llama3_1-8b
2. phi3_5-mini-instruct, phi3_5-moe-instruct, phi3_5-vision-instruct
3. llava-onevision-qwen2-0_5b-ov, llava-onevision-qwen2-7b-ov, llava-onevision-qwen2-72b-ov

新数据集：

1. longwriter-6k
2. rlaif-v
3. latex-ocr-print, latex-ocr-handwrite

What's Changed

• fix imports by @tastelikefeet in https://github.com/modelscope/ms-swift/pull/1748
• compat with torch=1.12/1.13 by @Jintao-Huang in https://github.com/modelscope/ms-swift/pull/1752
• update rlaif-v hf dataset by @hjh0119 in https://github.com/modelscope/ms-swift/pull/1755
• fix lmdeploy: AssertionError: failed to match chat template, please explicit set chat_template_config by @Jintao-Huang in https://github.com/modelscope/ms-swift/pull/1759
• use eager -> sdpa by @Jintao-Huang in https://github.com/modelscope/ms-swift/pull/1764
• Fix GLM4 agent toolcall by @tastelikefeet in https://github.com/modelscope/ms-swift/pull/1767
• Support LongWriter-llama3.1-8b and LongWriter-glm4-9b. by @DaozeZhang in https://github.com/modelscope/ms-swift/pull/1762
• Support llava onevision by @Jintao-Huang in https://github.com/modelscope/ms-swift/pull/1761
• [TorchAcc] fix: fix saving and loading checkpoint for full sft FSDP by @baoleai in https://github.com/modelscope/ms-swift/pull/1765
• Fix deepseek-coder-v2-lite template by @Jintao-Huang in https://github.com/modelscope/ms-swift/pull/1771
• Fix qwen2-audio & zero3 by @Jintao-Huang in https://github.com/modelscope/ms-swift/pull/1774
• Fix zero3 & minicpm-v/internvl2/xcomposer by @Jintao-Huang in https://github.com/modelscope/ms-swift/pull/1772
• fix infer dataset_test_ratio by @Jintao-Huang in https://github.com/modelscope/ms-swift/pull/1779
• fix moe & gradient_checkpointing by @Jintao-Huang in https://github.com/modelscope/ms-swift/pull/1782
• support phi3.5-vision by @Jintao-Huang in https://github.com/modelscope/ms-swift/pull/1780
• ReFT by @tastelikefeet in https://github.com/modelscope/ms-swift/pull/1785
• update doc by @tastelikefeet in https://github.com/modelscope/ms-swift/pull/1789
• support qwen-vl & base64 by @Jintao-Huang in https://github.com/modelscope/ms-swift/pull/1790
• fix yi-vl template by @Jintao-Huang in https://github.com/modelscope/ms-swift/pull/1793
• fix bugs by @Jintao-Huang in https://github.com/modelscope/ms-swift/pull/1794
• fix imports by @tastelikefeet in https://github.com/modelscope/ms-swift/pull/1796
• fix history_roles by @Jintao-Huang in https://github.com/modelscope/ms-swift/pull/1798
• fix mllm rlhf with full sft type by @hjh0119 in https://github.com/modelscope/ms-swift/pull/1800
• fix CI by @tastelikefeet in https://github.com/modelscope/ms-swift/pull/1797
• fix megatron_patch_path by @wning13 in https://github.com/modelscope/ms-swift/pull/1804
• Support hd num by @Jintao-Huang in https://github.com/modelscope/ms-swift/pull/1801
• Support Latex OCR dataset by @Jintao-Huang in https://github.com/modelscope/ms-swift/pull/1810
• fix offline export by @wning13 in https://github.com/modelscope/ms-swift/pull/1805
• fix by @tastelikefeet in https://github.com/modelscope/ms-swift/pull/1811

New Contributors

• @wning13 made their first contribution in https://github.com/modelscope/ms-swift/pull/1804

Full Changelog: https://github.com/modelscope/ms-swift/compare/v2.3.1...v2.3.2