Closed Catiks closed 1 year ago
Got the same error while running prediction with --cpu-only flag.
This is due to the ESM2 LM trunk being loaded in fp16 by default, and CPU not supporting fp16. @nikitos9000 maybe we can simply modify the inference script on L133 to move the LM to fp32?
Also see the huggingface notebook - open in colab.
Should be fixed now in #360 Will close for now, please let us know if any further issue!
I'm probably missing something obvious but I'm still getting this error, even after uninstalling and reinstalling with
pip install git+https://github.com/facebookresearch/esm.git
esm reports version 2.0.1
Is scripts/esmfold_inference.py
still your entrypoint? Could you please check if you have this line as in PR https://github.com/facebookresearch/esm/pull/360/files#diff-5c27367cac97a707aa8cedecbf7726c1e817e644fe4a999548d52ed96dffb483R129 ?
Is
scripts/esmfold_inference.py
still your entrypoint? Could you please check if you have this line as in PR https://github.com/facebookresearch/esm/pull/360/files#diff-5c27367cac97a707aa8cedecbf7726c1e817e644fe4a999548d52ed96dffb483R129 ?
ah removing everything then using pip again and copying the scripts/esmfold_inference.py resolved the issue, I am just bad at git. thanks for the quick response.
I want to run esmfold on only cpu, but i got the following error: RuntimeError: "LayerNormKernelImpl" not implemented for 'Half'