[PT FE]: fixes type checking for freezing conditions (#19199)
This commit is contained in:
parent
71ac5ee301
commit
89956b65e3
@ -77,7 +77,7 @@ class TorchScriptPythonDecoder (Decoder):
|
||||
preserved_attributes = []
|
||||
for name, module in model.named_modules():
|
||||
if hasattr(module, "weight"):
|
||||
if module.weight.dtype in [torch.int8, torch.uint8]:
|
||||
if module.weight is not None and module.weight.dtype in [torch.int8, torch.uint8]:
|
||||
preserved_attributes.append(name)
|
||||
return preserved_attributes
|
||||
|
||||
@ -176,7 +176,7 @@ class TorchScriptPythonDecoder (Decoder):
|
||||
first_input = next(n.inputs())
|
||||
if first_input.node().kind() == "prim::Constant":
|
||||
ivalue = first_input.toIValue()
|
||||
if ivalue is not None and ivalue.dtype in [torch.bfloat16, torch.float16]:
|
||||
if isinstance(ivalue, torch.Tensor) and ivalue.dtype in [torch.bfloat16, torch.float16]:
|
||||
# do not freeze models with compressed constants
|
||||
skip_freeze = True
|
||||
break
|
||||
|
Loading…
Reference in New Issue
Block a user