YiJunSachs

训练qwen-vl-2,5-3b:出现oom错误，但我的image质量确实比较高，调低cutofflength又会报shape mismatch

4

### Reminder - [x] I have read the above rules and searched the existing issues. ### System Info ```# YAML 配置文件对应于提供的 llamafactory-cli train 命令 stage: sft do_train: true model_name_or_path: "Qwen/Qwen2.5-VL-3B-Instruct"...

### Reminder - [x] I have read the above rules and searched the existing issues. ### System Info [2025-10-14 22:24:04,861] [INFO] [real_accelerator.py:222:get_accelerator] Setting ds_accelerator to cuda (auto detect) Traceback (most...

### What happened? my dataset is relatively large ### Versions python =3.10 os=ubuntu22.04 chromadb=1.1.0 ### Relevant log output ```shell ```

全量微调vl-3.5-1B后，使用vllm推理报错：ValueError: pixel_values dim[2] expected 448, got 384

2

但是我lora去调1b，然后再去推理并没有报这个错啊

在部署的过程中，出现api和network error，我怎么知道是哪个服务出现错误了呢

3

报错信息如下： Error: Attempt 7 failed with an API or network error: Error code: 500 Retrying in 30.00 seconds... Error: Attempt 7 failed with an API or network error: Connection error.