Update ModelLoader to set default vocab_size if not defined in model config, enhancing compatibility with tokenizer defaults.
This commit is contained in:
@@ -762,6 +762,10 @@ class ModelLoader:
|
|||||||
)
|
)
|
||||||
|
|
||||||
else:
|
else:
|
||||||
|
if not hasattr(self.model_config, 'vocab_size'):
|
||||||
|
LOG.warning("Model config does not have vocab_size attribute, setting to 50257")
|
||||||
|
self.model_config.vocab_size = 50257
|
||||||
|
|
||||||
self.model = getattr(transformers, self.model_type).from_pretrained(
|
self.model = getattr(transformers, self.model_type).from_pretrained(
|
||||||
self.base_model,
|
self.base_model,
|
||||||
config=self.model_config,
|
config=self.model_config,
|
||||||
|
|||||||
Reference in New Issue
Block a user