Junjie

Results 34 comments of Junjie

Add the option of saving in parquet instead of arrow

Thanks! That's good!

when will the code be released

thanks, have fun :)

[Performance]: Added request take too much time, and the model will not run untill all the request are added into the cache

same problem. any other suggestions?

支持GME微调么

感谢您的工作！ 请问 InfoNCE 数据格式这里如何支持多张图片呢？ ```python # LLM {"query": "sentence1", "response": "sentence2"} # MLLM {"query": "", "response": "sentence", "images": "/some/images.jpg"} {"query": "sentence1", "response": "sentence2", "images": "/some/images.jpg"} ```

支持GME微调么

> 支持多图的，比如： > > {"query": "sentence1", "response": "sentence2", "images": ['img1', 'img2']} 感谢回复。请问这样如何标识图片是来自 query 还是 response 呢？

支持GME微调么

哦哦。这样 images 列表里是 query 和 response 里按照 `` 顺序读吗？e.g. ```python {"query": "sentence1", "response": "sentence2", "images": ['/path/to/img1', '/path/to/img2', '/path/to/img3']} ---> {"query": "sentence1", "response": "sentence2"} ``` 也就是先 query 图片后 response 图片？因为 VLM...

支持GME微调么

> response里面带有图片的格式我现在正在调试，尽快支持 okok 辛苦

ActorUnavailableError when using multi node training

single node multi gpus encounter the same error. how to debug ?

使用qwen2.5_VL fine-tune后评测出现错误，是怎么回事

``` FileNotFoundError: [Errno 2] No such file or directory: 'workspace/qwen25vl_3b_sft/T20250402_G1774d8cd/qwen25vl_3b_sft_DocVQA.xlsx' ``` combination 没有成功，为啥还是会读这个 `xxx_DocVQA.xlsx ` 文件呢？这里应该是读 `xxx_DocVQA_TEST.xlsx`。 麻烦帮忙看下。

Can OMG method be transferred to flux model ?

can the spatial relation be preserved in mm-dit ?