riaz.somc/transformers
1
0
Fork 0
mirror of https://github.com/huggingface/transformers.git synced 2025-07-24 23:08:57 +06:00
Code Issues Actions 32 Packages Projects Releases Wiki Activity

Fix --bf16 option support for Neuron after PR #22300 (#22307)

Browse Source 
This PR fixes the "RuntimeError: No CUDA GPUs are available"
when running with --bf16 option on Neuron.

Related PRs:
https://github.com/huggingface/transformers/pull/20684
https://github.com/huggingface/transformers/pull/22300
This commit is contained in:
jeffhataws 2023-03-23 09:27:13 -07:00 committed by GitHub
parent aef488c503
commit ec9b18f62d
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 6 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
src/transformers/trainer.py
View File

@ -588,7 +588,12 @@ class Trainer:
if args.fp16 or args.bf16:
if args.half_precision_backend == "auto":
if args.device == torch.device("cpu"):
if is_torch_neuroncore_available():
if args.fp16:
raise ValueError("Tried to use `fp16` but this option is not yet supported on Neuron.")
else:
args.half_precision_backend = "cpu_amp"
elif args.device == torch.device("cpu"):
if args.fp16:
raise ValueError("Tried to use `fp16` but it is not supported on cpu")
elif _is_native_cpu_amp_available: