zhangyiss/mlx-examples
1
0
Fork 0
mirror of https://github.com/ml-explore/mlx-examples.git synced 2025-09-01 12:49:50 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

QWEN: Fix unsupported ScalarType BFloat16

Browse Source 
Fix unsupported ScalarType BFloat16.

Env: Mac M1 Ultra
torch: torch-2.0.0, metal
Apple clang version 15.0.0 (clang-1500.0.40.1)
Target: arm64-apple-darwin23.1.0
Thread model: posix
InstalledDir: /Library/Developer/CommandLineTools/usr/bin

```
Traceback (most recent call last):
  File "/Volumes/v1/models/mlx-examples-main/llms/qwen/convert.py", line 110, in <module>
    convert(args)
  File "/Volumes/v1/models/mlx-examples-main/llms/qwen/convert.py", line 63, in convert
    weights = {replace_key(k): v.numpy() for k, v in state_dict.items()}
  File "/Volumes/v1/models/mlx-examples-main/llms/qwen/convert.py", line 63, in <dictcomp>
    weights = {replace_key(k): v.numpy() for k, v in state_dict.items()}
TypeError: Got unsupported ScalarType BFloat16
```

Fix: almost same as [#10](429ddb30dc)
This commit is contained in:
Yifan
2023-12-25 20:18:26 +08:00
committed by GitHub
parent 647e48870a
commit b9607f9510
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
llms/qwen/convert.py
View File

@@ -60,7 +60,7 @@ def convert(args):
args.model, trust_remote_code=True, torch_dtype=torch.float16
)
state_dict = model.state_dict()
weights = {replace_key(k): v.numpy() for k, v in state_dict.items()}
weights = {replace_key(k): (v.numpy() if v.dtype != torch.bfloat16 else v.to(torch.float32).numpy()) for k, v in state_dict.items()}
config = model.config.to_dict()
if args.quantize: