swift icon indicating copy to clipboard operation
swift copied to clipboard

ms-swift: Use PEFT or Full-parameter to finetune 250+ LLMs or 25+ MLLMs

Results 206 swift issues
Sort by recently updated
recently updated
newest added

**Describe the bug** Eval 模块使用时 C-Eval 进行测试时,会大量重复出现 [INFO:modelscope] dataset_type: 1 [INFO:modelscope] dataset_type: 1 [INFO:modelscope] dataset_type: 1 ![image](https://github.com/modelscope/swift/assets/143787055/e53531c2-9cfc-4d8e-bf23-a3b32317ad69) 这让我怀疑程序是否卡死,直到想起C-Eval拥有大量推理科目时才明白程序在做什么 希望能优化这个问题

bug

目前测试数据上下文比较长,且无法缩短。请教一下有没有方法可以支持一下更长的上下文 ![image](https://github.com/modelscope/swift/assets/47747764/8391f136-2b8c-4e96-a00a-89f0e56dc6fd)

enhancement

**Describe the bug** SftArguments 数据参数如下设置,其他无关均省略 dataset=[f'{DatasetName.alpaca_zh}#100', f'{DatasetName.alpaca_en}#50', f'{DatasetName.self_cognition}#250'] custom_train_dataset_path=["./data/faq_train.jsonl"], custom_val_dataset_path=["./data/faq_valid.jsonl"], dataset_test_ratio 默认为0.01,这导致了 custom_train_dataset_path和custom_val_dataset_path传入路径时,均按照0.01进行了切分,且把custom_train_dataset_path和custom_val_dataset_path里面99%的内容作为训练,1%的内容作为验证集,不合符逻辑设定。 swift版本:commitId:845ac0ce46c4b904a809e3570e8dfb830f9b4e00

目前npu都支持哪些训练和哪些模型?在使用npu微调72b_chat模型的时候总是报错

# PR type - [ ] Bug Fix - [ ] New Feature - [ ] Document Updates - [x] More Models or Datasets Support # PR information https://github.com/modelscope/swift/issues/908

有关InternVL-Chat 1.5最佳实践中所提到的微调过程,想问下可以对Visual Grounding任务进行微调么?以及如果可以的话,prompt模板是什么样子的? ``` { "id": "n0167", "image": "xxxxx", "conversations": [ { "from": "human", "value": "\nPlease provide the bounding box coordinate of the region this sentence describes: xxx" }, {...

@hjh0119 在1.5版本微调训练结束后,按照教程里面的推理命令,并加载了本地的权重,我使用的infer命令如下: ``` CUDA_VISIBLE_DEVICES=0,1 swift infer --ckpt_dir output/internvl-chat-v1_5/v0-20240512-191616/checkpoint-25 --load_dataset_config true --dtype bf16 --model_id_or_path xxxxx/InternVL/pretrained/InternVL-Chat-V1-5 ``` 但在加载后出现了报错情况,不知之前测试时有没有发生 ``` Traceback (most recent call last): File "/data2/renyw/PythonWorkspace/FM-LLM/swift/swift/cli/infer.py", line 5, in infer_main() File "/data2/renyw/PythonWorkspace/FM-LLM/swift/swift/utils/run_utils.py",...

**Describe the bug** What the bug is, and how to reproduce, better with screenshots(描述bug以及复现过程,最好有截图) 使用UI 生成的命令行使用 `--self_cognition_sample` 参数(用这个参数是可以微调成功的),但是文档里又说该参数已经废弃,请使用`--dataset` , 使用新参数是报错的。 ![image](https://github.com/modelscope/swift/assets/37682426/552a8324-d0e9-40e2-946b-8103e05e4723) **Your hardware and system info** Write your system info...

**Describe the bug** 自从支持lisa以后,8个v100可以全参数微调32b了 我一直是用main分支源码安装使用swift的,最近几天发现同样入参的微调命令,以前可以运行,现在会报显存不足的问题 ``` CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 NPROC_PER_NODE=1 \ swift sft --sft_type full \ --model_type qwen1half-32b-chat \ --dataset ms-bench \ --train_dataset_sample 5000 \ --self_cognition_sample 1000 \ --logging_steps 5 \ --max_length...

Current code only supports single or batch inference for multi-modal models (Llava1.6, cogvlm etc) due to lack of vllm support. Any plans to add feature support to enable API support...

enhancement