Zhenyu Wang (Luffy)

Need Help for Usage: how to do resume/continuing training from checkpoint

6

### Reminder - [x] I have read the above rules and searched the existing issues. ### System Info - llamafactory version: 0.9.3.dev0 - Platform: Linux-5.4.0-162-generic-x86_64-with-glibc2.31 - Python version: 3.12.9 -...

Hanging problem when LoRA finte-tune Qwen2.5Omni with multi-turn video-audio samples with deepspeedz3

41

### Reminder - [x] I have read the above rules and searched the existing issues. ### System Info - `llamafactory` version: 0.9.3.dev0 - Platform: Linux-5.4.0-162-generic-x86_64-with-glibc2.31 - Python version: 3.12.9 -...

fit the cases of mmdata in sys_msg

14

# What does this PR do? Fit the cases of mmdata in sys_msg In some tasks like Multimodal Roleplaying, we may need to give the role image in system message....

support new processor arg video_maxlen_ttl

4

# What does this PR do? Fixes #7693 Replace #7700 (too many changes in it to check, so open this PR with a new branch) ## Before submitting - [x]...

### Reminder - [x] I have read the above rules and searched the existing issues. ### Description Multi-video cases need a param to better control `video_maxlen_per_video` in a refined way...

**Describe the bug** We‘ve discussed a lot in llamafactory community: https://github.com/hiyouga/LLaMA-Factory/issues/7767 And it's weird that this issue only happens when we use **ds-z3** or **ds-z3-offload**. **To Reproduce** 1. Env preparation:...

[BUG] AttributeError: 'Linear' object has no attribute 'ds_grads_remaining'

15

**Describe the bug** When I use llamafactory to lora sft on qwen2.5vl-7b with ds-z3, I met: ``` Traceback (most recent call last): File "/home/mas-wang.zhenyu/anaconda3/envs/vllm/bin/torchrun", line 8, in sys.exit(main()) ^^^^^^ File...

How to incrementally update memories on an on-going conversation?

1

During an on-going conversation, the messages of it can be incrementally updated. An example is like: ``` # first round messages = [ {"role":"user", "content": "Hello!"}, {"role":"assistant", "content": "Hi there,...