Online weight updates from torch.distributed (#2279)
This commit is contained in:
@@ -307,6 +307,8 @@ class LlamaForCausalLM(nn.Module):
|
||||
self.quant_config = quant_config
|
||||
self.torchao_config = global_server_args_dict["torchao_config"]
|
||||
self.model = LlamaModel(config, quant_config=quant_config)
|
||||
# Llama 3.2 1B Insturct set tie_word_embeddings to True
|
||||
# Llama 3.1 8B Insturct set tie_word_embeddings to False
|
||||
if self.config.tie_word_embeddings:
|
||||
self.lm_head = self.model.embed_tokens
|
||||
else:
|
||||
|
||||
Reference in New Issue
Block a user