Koin
Koin
I made my version by imitating the coding style in "run_webcam.py" Couldn't wait to see your version int the future
我指的参数列表和tutorial这样的，在api的主页上没有找到
请问下正负样本的构造必须要加`\n` 前缀吗？ 关于这块儿有没有其他说明呢 @czczup
by the way, i used 32 GPUs(V100 32g) with pytorch as launcher
i put my start script here: ```shell python internvl/train/internvl_chat_finetune.py \ --model_name_or_path "/xxx/InternVL/pretrained/InternVL-Chat-Chinese-V1-2-Plus" \ --conv_style "Hermes-2" \ --output_dir ${OUTPUT_DIR} \ --meta_path "internvl_chat/shell/data/zw_data.json" \ --overwrite_output_dir True \ --force_image_size 448 \ --down_sample_ratio 0.5...
> 噢噢，可能需要在shell上面加一行： export LAUNCHER=pytorch 麻烦您试试看 I still encounter this problem even if i change the scale from 448 into 224 ```groovy torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 280.00...
> Sorry, perhaps adjustments to the deepspeed zero3 configuration are needed (my default configuration may be more suitable for 80G A100 GPUs, keeping more models per GPU to reduce communication),...