8000 out of memory when downloading models locally. · Issue #306 · borisdayma/dalle-mini · GitHub
[go: up one dir, main page]

Skip to content

out of memory when downloading models locally. #306

Open

Description

@luluxxxxx

hello. i'm trying to use the models locally to avoid downloading them at each run.
using
DALLE_MODEL = "dalle-mini/dalle-mini/mega-1-fp16:latest"
DALLE_COMMIT_ID = None
works fine and gives me a result but it will download the models at each run.

i've downloaded the models locally using
git lfs install
git clone https://huggingface.co/dalle-mini/dalle-mega
git lfs install
git clone https://huggingface.co/dalle-mini/vqgan_imagenet_f16_16384
under a /model directory
and now use
DALLE_MODEL = "./models/mega/dalle-mega"
DALLE_COMMIT_ID = None
VQGAN_REPO= "./models/vqgan/vqgan_imagenet_f16_16384"
VQGAN_COMMIT_ID= None

but now it gives me a memory error :
Traceback (most recent call last):
File "dalle.py", line 44, in
params = replicate(params)
File "/mnt/shared/foss-20/dalle-mini/dallemini/lib/python3.8/site-packages/flax/jax_utils.py", line 56, in replicate
return jax.device_put_replicated(tree, devices)
File "/mnt/shared/foss-20/dalle-mini/dallemini/lib/python3.8/site-packages/jax/_src/api.py", line 2877, in device_put_replicated
return tree_map(_device_put_replicated, x)
File "/mnt/shared/foss-20/dalle-mini/dallemini/lib/python3.8/site-packages/jax/_src/tree_util.py", line 205, in tree_map
return treedef.unflatten(f(*xs) for xs in zip(*all_leaves))
File "/mnt/shared/foss-20/dalle-mini/dallemini/lib/python3.8/site-packages/jax/_src/tree_util.py", line 205, in
return treedef.unflatten(f(*xs) for xs in zip(*all_leaves))
File "/mnt/shared/foss-20/dalle-mini/dallemini/lib/python3.8/site-packages/jax/_src/api.py", line 2865, in _device_put_replicated
buf, = dispatch.device_put(x, devices[0])
File "/mnt/shared/foss-20/dalle-mini/dallemini/lib/python3.8/site-packages/jax/_src/dispatch.py", line 1024, in device_put
return device_put_handlers[type(x)](x, device)
File "/mnt/shared/foss-20/dalle-mini/dallemini/lib/python3.8/site-packages/jax/_src/dispatch.py", line 1055, in _device_put_device_array
x = _copy_device_array_to_device(x, device)
File "/mnt/shared/foss-20/dalle-mini/dallemini/lib/python3.8/site-packages/jax/_src/dispatch.py", line 1082, in _copy_device_array_to_device
moved_buf = backend.buffer_from_pyval(x.device_buffer.to_py(), device)
jaxlib.xla_extension.XlaRuntimeError: RESOURCE_EXHAUSTED: Out of memory while trying to allocate 402653184 bytes.

i have a Nvidia GTX 1080 with 8192 MB of memory.
but i wonder why it works when downloading the models. is it using a smaller part of it ?
sorry if all this are dumb questions.

thanks in advance

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions

      0