Reminder
System Info
llamafactory version: 0.9.3.dev0- Platform: Linux-6.8.0-55-generic-x86_64-with-glibc2.39
- Python version: 3.12.9
- PyTorch version: 2.5.1+cu124 (GPU)
- Transformers version: 4.50.0
- Datasets version: 3.3.2
- Accelerate version: 1.4.0
- PEFT version: 0.12.0
- TRL version: 0.9.6
- GPU type: NVIDIA A100-SXM4-80GB
- GPU number: 8
- GPU memory: 79.25GB
- DeepSpeed version: 0.16.4
- vLLM version: 0.7.3
- Git commit: 76cb0a9fc305ee42e26472d5afe219caffc36bd7
Reproduction
Referring to Issues #7385, #7325, #7320, and #7291, several users have reported the problem Processor was not found, please check and update your processor config when training with Gemma 3. After inspecting the source code, I found that in this linehttps://github.com/hiyouga/LLaMA-Factory/blob/165d3ed084b093accb2aa1d1209d1903183245e1/src/llamafactory/model/loader.py#L96, when loading the processor, Gemma 1B fails to find the processor because it's a pure text model. This then causes an error later during the process_messages function call herehttps://github.com/hiyouga/LLaMA-Factory/blob/ef5f1c1def3da62ee2d5e6ba933f9d7d6aab4340/src/llamafactory/data/mm_plugin.py#L401, since this process_messages function seems to be written with a multimodal architecture in mind. It might be necessary to add some additional conditional handling specifically for Gemma 1 to address this.
Others
No response
Reminder
System Info
llamafactoryversion: 0.9.3.dev0Reproduction
Referring to Issues #7385, #7325, #7320, and #7291, several users have reported the problem
Processor was not found, please check and update your processor configwhen training with Gemma 3. After inspecting the source code, I found that in this linehttps://github.com/hiyouga/LLaMA-Factory/blob/165d3ed084b093accb2aa1d1209d1903183245e1/src/llamafactory/model/loader.py#L96, when loading the processor, Gemma 1B fails to find the processor because it's a pure text model. This then causes an error later during theprocess_messagesfunction call herehttps://github.com/hiyouga/LLaMA-Factory/blob/ef5f1c1def3da62ee2d5e6ba933f9d7d6aab4340/src/llamafactory/data/mm_plugin.py#L401, since thisprocess_messagesfunction seems to be written with a multimodal architecture in mind. It might be necessary to add some additional conditional handling specifically for Gemma 1 to address this.Others
No response