search

InternLM / InternLM-XComposer

InternLM-XComposer-2.5: A Versatile Large Vision Language Model Supporting Long-Contextual Input and Output
2.06k stars 128 forks source link

How much VRAM required for Share Captioner? #232

Open deadpipe opened 3 months ago

deadpipe commented 3 months ago

Hi

1) How much VRAM is required for Share Captioner?

2) Is there a way to use Multiple GPUs for loading Share Captioner?

3) Are there Quantization methods (4bit, 8bit) available for Share Captioner?

I have tried to run Share Captioner but I have got CUDA Out of Memory error on a Rtx 3090:

torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 172.00 MiB. GPU 0 has a total capacty of 24.00 GiB of which 0 bytes is free. Including non-PyTorch memory, this process has 17179869184.00 GiB memory in use. Of the allocated memory 23.10 GiB is allocated by PyTorch, and 3.24 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

Han230104 commented 1 month ago

I am also having this question. Does anybody have a answer ?

Han230104 commented 4 weeks ago

I am able to run the share-captioner with A6000 (48G). 4090 (24G) is not enough.

deadpipe commented 4 weeks ago

@Han230104 So how much VRAM is actually being used in your A6000?