Open whybfq opened 4 months ago
Is it possible to do a single image inference of stabilityai/stable-diffusion-2-1
model on your GPU?
Most of my experiments are conducted on A5000/A6000.
For the case you are testing, it is a simple SD 2.1 inference with modified prompts, you can perhaps change some hyperparameters here to see if the code runs.
Thanks for your suggestion, while I set im_batch=1, num_inference_steps=5, nbatches set 4, still have this error
I first generated the pictures using the diff_inference.py
python diff_inference.py -nb 4000 --dataset laion --capstyle instancelevel_blip --rand_augs rand_numb_add
while I met File "/home/anaconda3/envs/diffrep/lib/python3.9/site-packages/diffusers/models/cross_attention.py", line 314, in call attention_probs = attn.get_attention_scores(query, key, attention_mask) File "/home/anaconda3/envs/diffrep/lib/python3.9/site-packages/diffusers/models/cross_attention.py", line 253, in get_attention_scores attention_probs = attention_scores.softmax(dim=-1) torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 3.16 GiB (GPU 0; 15.46 GiB total capacity; 11.31 GiB already allocated; 2.48 GiB free; 11.39 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONFwhile I still have a lot of gpu, thanks for your suggestion