From 8bf6cf5d9af56afbcf71095b2390a9cf36227517 Mon Sep 17 00:00:00 2001 From: Jintao Huang Date: Fri, 13 Dec 2024 19:09:35 +0800 Subject: [PATCH 1/2] update --- .../\345\277\253\351\200\237\345\274\200\345\247\213.md" | 4 ++-- ...21\275\344\273\244\350\241\214\345\217\202\346\225\260.md" | 2 +- ...16\250\347\220\206\345\222\214\351\203\250\347\275\262.md" | 2 +- docs/source/index.rst | 2 +- docs/source_en/GetStarted/Quick-start.md | 2 +- docs/source_en/Instruction/Command-line-parameters.md | 2 +- docs/source_en/Instruction/Inference-and-deployment.md | 2 +- docs/source_en/index.rst | 2 +- swift/llm/argument/base_args/base_args.py | 2 ++ swift/llm/argument/base_args/data_args.py | 2 ++ swift/llm/dataset/data/dataset_info.json | 1 + 11 files changed, 14 insertions(+), 9 deletions(-) diff --git "a/docs/source/GetStarted/\345\277\253\351\200\237\345\274\200\345\247\213.md" "b/docs/source/GetStarted/\345\277\253\351\200\237\345\274\200\345\247\213.md" index 7878bb5d60..0c5e19e05e 100644 --- "a/docs/source/GetStarted/\345\277\253\351\200\237\345\274\200\345\247\213.md" +++ "b/docs/source/GetStarted/\345\277\253\351\200\237\345\274\200\345\247\213.md" @@ -8,7 +8,7 @@ ms-swift是魔搭社区提供的大模型与多模态大模型训练部署框架 - 轻量训练：支持了LoRA、QLoRA、DoRA、LoRA+、ReFT、RS-LoRA、LLaMAPro、Adapter、GaLore、Q-Galore、LISA、UnSloth、Liger-Kernel等轻量微调方式。 - 分布式训练：支持分布式数据并行（DDP）、device_map简易模型并行、DeepSpeed ZeRO2 ZeRO3、FSDP等分布式训练技术。 - 量化训练：支持对BNB、AWQ、GPTQ、AQLM、HQQ、EETQ量化模型进行训练。 -- RLHF训练：支持纯文本大模型和多模态大模型的DPO、CPO、SimPO、ORPO、KTO等人类对齐训练方法。 +- RLHF训练：支持纯文本大模型和多模态大模型的DPO、CPO、SimPO、ORPO、KTO、RM等人类对齐训练方法。 - 多模态训练：支持对图像、视频和语音不同模态模型进行训练，支持VQA、Caption、OCR、Grounding任务的训练。 - 界面训练：以界面的方式提供训练、推理、评测、量化的能力，完成大模型的全链路。 - 插件化与拓展：支持自定义模型和数据集拓展，支持对loss、metric、trainer、loss-scale、callback、optimizer等组件进行自定义。 @@ -75,4 +75,4 @@ swift infer \ > [!TIP] > 更多例子可以查看：[examples](https://github.com/modelscope/ms-swift/tree/main/examples) > -> 以python方式进行训练和推理的例子可以查看[notebook](https://github.com/modelscope/ms-swift/tree/main/examples/notebook) +> 以python方式进行训练和推理的例子可以查看：[notebook](https://github.com/modelscope/ms-swift/tree/main/examples/notebook) diff --git "a/docs/source/Instruction/\345\221\275\344\273\244\350\241\214\345\217\202\346\225\260.md" "b/docs/source/Instruction/\345\221\275\344\273\244\350\241\214\345\217\202\346\225\260.md" index ce4699fa4d..00d5c59873 100644 --- "a/docs/source/Instruction/\345\221\275\344\273\244\350\241\214\345\217\202\346\225\260.md" +++ "b/docs/source/Instruction/\345\221\275\344\273\244\350\241\214\345\217\202\346\225\260.md" @@ -91,7 +91,7 @@ - 🔥learning_rate: 学习率，全参数默认为1e-5，tuner为1e-4 - lr_scheduler_type: lr_scheduler类型，默认为cosine - lr_scheduler_kwargs: lr_scheduler其他参数 -- 🔥gradient_checkpointing_kwargs: 传入`torch.utils.checkpoint`中的参数. 例如设置为`'{"use_reentrant": false}'` +- 🔥gradient_checkpointing_kwargs: 传入`torch.utils.checkpoint`中的参数. 例如设置为`--gradient_checkpointing_kwargs '{"use_reentrant": false}'` - report_to: 默认值为`tensorboard` - remove_unused_columns: 默认值False - logging_first_step: 是否记录第一个step的打印，默认值True diff --git "a/docs/source/Instruction/\346\216\250\347\220\206\345\222\214\351\203\250\347\275\262.md" "b/docs/source/Instruction/\346\216\250\347\220\206\345\222\214\351\203\250\347\275\262.md" index d257d71011..2e10580d56 100644 --- "a/docs/source/Instruction/\346\216\250\347\220\206\345\222\214\351\203\250\347\275\262.md" +++ "b/docs/source/Instruction/\346\216\250\347\220\206\345\222\214\351\203\250\347\275\262.md" @@ -3,7 +3,7 @@ SWIFT支持以命令行、Python代码和界面方式进行推理和部署： - 使用`engine.infer`或者`engine.infer_async`进行python的方式推理. 参考[这里](https://github.com/modelscope/ms-swift/blob/main/examples/infer/demo.py). - 使用`swift infer`使用命令行的方式进行推理. 参考[这里](https://github.com/modelscope/ms-swift/blob/main/examples/infer/cli_demo.sh). -- 使用`swift deploy`进行服务部署，并使用openai API或者`client.infer`的方式推理. 参考[这里](https://github.com/modelscope/ms-swift/tree/main/examples/deploy/client) +- 使用`swift deploy`进行服务部署，并使用openai API或者`client.infer`的方式推理. 服务端参考[这里](https://github.com/modelscope/ms-swift/tree/main/examples/deploy/server), 客户端参考[这里](https://github.com/modelscope/ms-swift/tree/main/examples/deploy/client). - 使用`swift web-ui`部署模型进行界面推理, 可以查看[这里](../GetStarted/界面使用.md) diff --git a/docs/source/index.rst b/docs/source/index.rst index 4ea240c330..71dfe47585 100644 --- a/docs/source/index.rst +++ b/docs/source/index.rst @@ -17,12 +17,12 @@ Swift DOCUMENTATION :maxdepth: 2 :caption: Instruction + Instruction/命令行参数.md Instruction/预训练及微调.md Instruction/人类对齐.md Instruction/推理和部署.md Instruction/评测.md Instruction/导出.md - Instruction/命令行参数.md Instruction/支持的模型和数据集.md Instruction/使用tuners.md Instruction/智能体的支持.md diff --git a/docs/source_en/GetStarted/Quick-start.md b/docs/source_en/GetStarted/Quick-start.md index d23844c9ac..7e00536d18 100644 --- a/docs/source_en/GetStarted/Quick-start.md +++ b/docs/source_en/GetStarted/Quick-start.md @@ -8,7 +8,7 @@ ms-swift is a comprehensive training and deployment framework for large language - Lightweight Training: Supports lightweight fine-tuning methods like LoRA, QLoRA, DoRA, LoRA+, ReFT, RS-LoRA, LLaMAPro, Adapter, GaLore, Q-Galore, LISA, UnSloth, Liger-Kernel, and more. - Distributed Training: Supports distributed data parallel (DDP), simple model parallelism via device_map, DeepSpeed ZeRO2 ZeRO3, FSDP, and other distributed training technologies. - Quantization Training: Provides training for quantized models like BNB, AWQ, GPTQ, AQLM, HQQ, EETQ. -- RLHF Training: Supports human alignment training methods like DPO, CPO, SimPO, ORPO, KTO for both text-based and multimodal large models. +- RLHF Training: Supports human alignment training methods like DPO, CPO, SimPO, ORPO, KTO, RM for both text-based and multimodal large models. - Multimodal Training: Capable of training models for different modalities such as images, videos, and audios; supports tasks like VQA (Visual Question Answering), Captioning, OCR (Optical Character Recognition), and Grounding. - Interface-driven Training: Offers training, inference, evaluation, and quantization capabilities through an interface, enabling a complete workflow for large models. - Plugins and Extensions: Allows customization and extension of models and datasets, and supports customizations for components like loss, metric, trainer, loss-scale, callback, optimizer, etc. diff --git a/docs/source_en/Instruction/Command-line-parameters.md b/docs/source_en/Instruction/Command-line-parameters.md index fdcf25d468..0015c8c6b0 100644 --- a/docs/source_en/Instruction/Command-line-parameters.md +++ b/docs/source_en/Instruction/Command-line-parameters.md @@ -92,7 +92,7 @@ This parameter list inherits from transformers `Seq2SeqTrainingArguments`, with - 🔥learning_rate: Learning rate, default is 1e-5 for all parameters, and 1e-4 for the tuner. - lr_scheduler_type: LR scheduler type, default is cosine. - lr_scheduler_kwargs: Other parameters for the LR scheduler. -- 🔥gradient_checkpointing_kwargs: Parameters passed to `torch.utils.checkpoint`. For example, set to `{"use_reentrant": false}`. +- 🔥gradient_checkpointing_kwargs: Parameters passed to `torch.utils.checkpoint`. For example, set to `--gradient_checkpointing_kwargs '{"use_reentrant": false}'`. - report_to: Default is `tensorboard`. - remove_unused_columns: Default is False. - logging_first_step: Whether to log the first step print, default is True. diff --git a/docs/source_en/Instruction/Inference-and-deployment.md b/docs/source_en/Instruction/Inference-and-deployment.md index 56b80c02aa..69787d5add 100644 --- a/docs/source_en/Instruction/Inference-and-deployment.md +++ b/docs/source_en/Instruction/Inference-and-deployment.md @@ -3,7 +3,7 @@ SWIFT supports inference and deployment through command line, Python code, and interface methods: - Use `engine.infer` or `engine.infer_async` for Python-based inference. See [here](https://github.com/modelscope/ms-swift/blob/main/examples/infer/demo.py) for reference. - Use `swift infer` for command-line-based inference. See [here](https://github.com/modelscope/ms-swift/blob/main/examples/infer/cli_demo.sh) for reference. -- Use `swift deploy` for service deployment and perform inference using the OpenAI API or `client.infer`. Refer to [here](https://github.com/modelscope/ms-swift/tree/main/examples/deploy/client) for more information. +- Use `swift deploy` for service deployment and perform inference using the OpenAI API or `client.infer`. Refer to the server guidelines [here](https://github.com/modelscope/ms-swift/tree/main/examples/deploy/server) and the client guidelines [here](https://github.com/modelscope/ms-swift/tree/main/examples/deploy/client). - Deploy the model with `swift web-ui` for web-based inference. You can check [here](../GetStarted/Interface-usage.md) for details. diff --git a/docs/source_en/index.rst b/docs/source_en/index.rst index 85587e7ea6..6131e05d61 100644 --- a/docs/source_en/index.rst +++ b/docs/source_en/index.rst @@ -17,12 +17,12 @@ Swift DOCUMENTATION :maxdepth: 2 :caption: Instruction + Instruction/Command-line-parameters.md Instruction/Pre-training-and-Fine-tuning.md Instruction/RLHF.md Instruction/Inference-and-deployment.md Instruction/Evaluation.md Instruction/Export.md - Instruction/Command-line-parameters.md Instruction/Supported-models-and-datasets.md Instruction/Use-tuners.md Instruction/Agent-support.md diff --git a/swift/llm/argument/base_args/base_args.py b/swift/llm/argument/base_args/base_args.py index 07dcea06e6..cc26e1fb29 100644 --- a/swift/llm/argument/base_args/base_args.py +++ b/swift/llm/argument/base_args/base_args.py @@ -63,6 +63,8 @@ class BaseArguments(GenerationArguments, QuantizeArguments, DataArguments, Templ def _init_custom_register(self) -> None: """Register custom .py file to datasets""" + if isinstance(self.custom_register_path, str): + self.custom_register_path = [self.custom_register_path] self.custom_register_path = to_abspath(self.custom_register_path, True) for path in self.custom_register_path: folder, fname = os.path.split(path) diff --git a/swift/llm/argument/base_args/data_args.py b/swift/llm/argument/base_args/data_args.py index d6d06fb8c2..3afb3683d5 100644 --- a/swift/llm/argument/base_args/data_args.py +++ b/swift/llm/argument/base_args/data_args.py @@ -50,6 +50,8 @@ class DataArguments: def _init_custom_dataset_info(self): """register custom dataset_info.json to datasets""" + if isinstance(self.custom_dataset_info, str): + self.custom_dataset_info = [self.custom_dataset_info] for path in self.custom_dataset_info: register_dataset_info(path) diff --git a/swift/llm/dataset/data/dataset_info.json b/swift/llm/dataset/data/dataset_info.json index b011c15eab..d24bebf57e 100644 --- a/swift/llm/dataset/data/dataset_info.json +++ b/swift/llm/dataset/data/dataset_info.json @@ -309,6 +309,7 @@ }, { "ms_dataset_id": "swift/Infinity-Instruct", + "subsets": ["3M", "7M", "0625", "Gen", "7M_domains"], "hf_dataset_id": "BAAI/Infinity-Instruct", "tags": ["qa", "quality", "multi-task"], "huge_dataset": true From bca52c87757b25b993450b5506951906c0076d24 Mon Sep 17 00:00:00 2001 From: Jintao Huang Date: Sat, 14 Dec 2024 13:36:36 +0800 Subject: [PATCH 2/2] update --- swift/llm/dataset/data/dataset_info.json | 3 +++ tests/tuners/test_extra_state_dict.py | 2 +- 2 files changed, 4 insertions(+), 1 deletion(-) diff --git a/swift/llm/dataset/data/dataset_info.json b/swift/llm/dataset/data/dataset_info.json index d24bebf57e..ab95cb5a36 100644 --- a/swift/llm/dataset/data/dataset_info.json +++ b/swift/llm/dataset/data/dataset_info.json @@ -310,6 +310,9 @@ { "ms_dataset_id": "swift/Infinity-Instruct", "subsets": ["3M", "7M", "0625", "Gen", "7M_domains"], + "columns": { + "label": "_" + }, "hf_dataset_id": "BAAI/Infinity-Instruct", "tags": ["qa", "quality", "multi-task"], "huge_dataset": true diff --git a/tests/tuners/test_extra_state_dict.py b/tests/tuners/test_extra_state_dict.py index f3e7e26520..9293da5458 100644 --- a/tests/tuners/test_extra_state_dict.py +++ b/tests/tuners/test_extra_state_dict.py @@ -34,7 +34,7 @@ def test_swift_extra_state_dict(self): with open(os.path.join(self.tmp_dir, 'extra_states', 'adapter_model.bin'), 'wb') as f: torch.save(state_dict, f) model = Model.from_pretrained('damo/nlp_structbert_sentence-similarity_chinese-base') - model = Swift.from_pretrained(model, self.tmp_dir) + model = Swift.from_pretrained(model, self.tmp_dir, inference_mode=False) names = [name for name, value in model.named_parameters() if value.requires_grad] self.assertTrue(any('classifier' in name for name in names)) self.assertTrue(torch.allclose(state_dict['classifier.weight'], model.base_model.classifier.weight))