能够提供个国内下载地址吗？？

[douyu0814]

数据集 和 Checkpoint 国内 都打不开啊 Data] [🤗 Checkpoint]

Training
Referring Expression Comprehension (REC)
GRPO
Download the COCO Train2014 image and unzip it, and we refer to the image dir as <your_image_root>.
Download the RefCOCO/+/g and RefGTA Annotation files and unzip it (RefGTA is used for out-of-domain evaluation).
Write the path of the annotation files in the src/open-r1-multimodal/data_config/rec.yaml file.

[SZhanZ]

Hello, 你好。huggingface的镜像网站可以打开吗 https://hf-mirror.com/omlab/Qwen2.5VL-3B-VLM-R1-REC-500steps

[douyu0814]

谢谢哈！         另外，官方能给说下 跑起来这个模型最低的 电脑配置吗?   估计很多人 都不知道啊

…

----- 原始邮件 ----- 发件人：Haozhan Shen ***@***.***> 收件人：om-ai-lab/VLM-R1 ***@***.***> 抄送人：douyu0814 ***@***.***>, Author ***@***.***> 主题：Re:_[om-ai-lab/VLM-R1]_能够提供个国内下载地址吗？？_(Issue_#132) 日期：2025年03月07日 21点04分 Hello, 你好。huggingface的镜像网站可以打开吗 https://hf-mirror.com/omlab/Qwen2.5VL-3B-VLM-R1-REC-500steps— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: ***@***.***> SZhanZ left a comment (om-ai-lab/VLM-R1#132) Hello, 你好。huggingface的镜像网站可以打开吗 https://hf-mirror.com/omlab/Qwen2.5VL-3B-VLM-R1-REC-500steps — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: ***@***.***>

[snakeztc]

We only have A100-80G GPU so we cannot try out all other resource settings. The general rule of thumb is to start from per_device_train_batch_size=1. Please note that we have updated the trainer so this number is the ACTUAL batch size in a single GPU, and num_generations does not matter here for memory usage.