Upload modeling_trm_text_ism.py
Browse files- modeling_trm_text_ism.py +2 -3
modeling_trm_text_ism.py
CHANGED
|
@@ -51,6 +51,7 @@ class TRMBlock(nn.Module):
|
|
| 51 |
|
| 52 |
class TRMTextISMForCausalLM(PreTrainedModel, GenerationMixin):
|
| 53 |
config_class = TRMTextISMConfig
|
|
|
|
| 54 |
def __init__(self, config):
|
| 55 |
super().__init__(config)
|
| 56 |
self.token_emb = nn.Embedding(config.vocab_size, config.dim)
|
|
@@ -76,9 +77,7 @@ class TRMTextISMForCausalLM(PreTrainedModel, GenerationMixin):
|
|
| 76 |
def set_output_embeddings(self, value):
|
| 77 |
self.lm_head = value
|
| 78 |
|
| 79 |
-
|
| 80 |
-
if hasattr(self, 'lm_head'):
|
| 81 |
-
self.lm_head.weight = self.token_emb.weight
|
| 82 |
|
| 83 |
def prepare_inputs_for_generation(self, input_ids, attention_mask=None, **kwargs):
|
| 84 |
return {"input_ids": input_ids, "attention_mask": attention_mask, "use_cache": False}
|
|
|
|
| 51 |
|
| 52 |
class TRMTextISMForCausalLM(PreTrainedModel, GenerationMixin):
|
| 53 |
config_class = TRMTextISMConfig
|
| 54 |
+
|
| 55 |
def __init__(self, config):
|
| 56 |
super().__init__(config)
|
| 57 |
self.token_emb = nn.Embedding(config.vocab_size, config.dim)
|
|
|
|
| 77 |
def set_output_embeddings(self, value):
|
| 78 |
self.lm_head = value
|
| 79 |
|
| 80 |
+
# tie_weights は削除 → lm_head と token_emb を独立したweightとして保持
|
|
|
|
|
|
|
| 81 |
|
| 82 |
def prepare_inputs_for_generation(self, input_ids, attention_mask=None, **kwargs):
|
| 83 |
return {"input_ids": input_ids, "attention_mask": attention_mask, "use_cache": False}
|