Ecosyste.ms: Timeline

Browse the timeline of events for every public repo on GitHub. Data updated hourly from GH Archive.

Betty-J

Betty-J opened an issue on modelscope/ms-swift
ms-swift-3.0微调后推理结果与ms-swift-2.x版本相差较大。
您好,我使用相同的数据集微调同一多模态模型,使用旧版 ms-swift 时可以得到比较理想的结果,更新到 ms-swift-3.0后微调无任何报错,但评测结果会比旧版差很多,以下是我的运行命令: SIZE_FACTOR=8 MAX_PIXELS=602112 CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 NPROC_PER_NODE=8 swift sft ...
Betty-J created a comment on an issue on modelscope/ms-swift
> fixed 您好,我用Qwen2VL 微调依旧报错,下面是我使用的完整命令: `MAX_PIXELS=602112 CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 NPROC_PER_NODE=8 swift sft \ --tuner_backend peft \ --model Qwen2_VL/Qwen2-VL-2B-Instr...

View on GitHub

Betty-J starred FreedomIntelligence/HuatuoGPT
Betty-J created a comment on an issue on QwenLM/Qwen2-VL
> > #190 (comment) > > 图片在训练之前并没有改变resize,而是直接采取原始分辨率作为输入,然后在训练过程中进行resize,resize的原因只是因为不需要高分辨率的图片,减轻训练和推理负担。因此,resize与最终框的偏差没有关系。偏差的原因可以根据[https://github.com/huggingface/transformers/pull/33487...

View on GitHub

Betty-J created a comment on an issue on QwenLM/Qwen2-VL
> #121 (comment) 请问你按照 fix-qwen2vl-no-position_ids(https://github.com/huggingface/transformers/pull/33487)修改后就不再出现box偏差问题了吗。我按照这个修改后,模型的表现并没有太多变化,box偏差也依旧存在。

View on GitHub

Betty-J starred lllyasviel/IC-Light
Betty-J starred lqtrung1998/mwp_ReFT
Betty-J starred taokz/BiomedGPT
Betty-J starred CAMMA-public/UltraSam
Betty-J starred JJGO/shrinkbench
Betty-J starred THUDM/GLM-Edge
Betty-J starred kvcache-ai/Mooncake
Betty-J starred karpathy/nanoGPT
Betty-J starred astral-sh/uv
Betty-J starred modelscope/data-juicer
Betty-J starred Powerlevel9k/powerlevel9k
Betty-J starred ryanoasis/nerd-fonts
Betty-J starred ohmyzsh/ohmyzsh
Betty-J starred OpenBMB/Eurus
Betty-J starred windingwind/zotero-pdf-translate
Betty-J starred lllyasviel/Omost
Betty-J starred VectorSpaceLab/OmniGen
Betty-J created a comment on a pull request on huggingface/transformers
> Yep sorry I approved I don't know why it was lost in tracks! Hi there, when I used `pip install transformers`, I installed version 4.46.2, which still does not include the updates mentioned he...

View on GitHub

Betty-J created a comment on an issue on modelscope/ms-swift
您好,当前执行多模态大模型微调时,设置的多轮对话在执行 infer 后保存的.jsonl文件中,response只包含了最后一轮对话的结果,而history中包含的历史信息是label的,后续可以支持保存多轮对话的全部结果吗

View on GitHub

Betty-J closed an issue on modelscope/ms-swift
微调deepseek-vl-1_3b-chat报错“RuntimeError: Expected to have finished reduction in the prior iteration before starting a new one.”
如题,执行脚本如下: ``` CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 NPROC_PER_NODE=8 swift sft \ --model_type deepseek-vl-1_3b-chat \ --model_id_or_path /Deepseek_VL/deepseek-vl-1.3b-chat \ --da...
Betty-J opened an issue on modelscope/ms-swift
swift2.6.0.dev0执行DPO训练报错KeyError: 'prompt_input_ids'
执行命令 `CUDA_VISIBLE_DEVICES=0,1 \ swift rlhf \ --rlhf_type dpo \ --model_type internvl2-1b \ --beta 0.1 \ --rpo_alpha 0.1 \ --sft_type lora \ --dataset local_train ...
Betty-J starred iancovert/prismatic-vlms
Betty-J starred RLHF-V/RLHF-V
Betty-J starred InternLM/InternLM-XComposer
Betty-J starred THUDM/GLM-4-Voice
Load more