[Bugfix] support tie_word_embeddings for all models (#5724)
This commit is contained in:
@@ -36,7 +36,7 @@ from vllm.model_executor.layers.quantization.base_config import (
|
||||
QuantizationConfig)
|
||||
from vllm.model_executor.layers.sampler import Sampler
|
||||
from vllm.model_executor.layers.vocab_parallel_embedding import (
|
||||
VocabParallelEmbedding)
|
||||
ParallelLMHead, VocabParallelEmbedding)
|
||||
from vllm.model_executor.model_loader.weight_utils import default_weight_loader
|
||||
from vllm.model_executor.sampling_metadata import SamplingMetadata
|
||||
from vllm.sequence import IntermediateTensors, SamplerOutput
|
||||
@@ -249,7 +249,11 @@ class GPT2LMHeadModel(nn.Module):
|
||||
cache_config,
|
||||
quant_config,
|
||||
prefix="transformer")
|
||||
self.lm_head = self.transformer.wte
|
||||
if self.config.tie_word_embeddings:
|
||||
self.lm_head = self.transformer.wte
|
||||
else:
|
||||
self.lm_head = ParallelLMHead(self.config.vocab_size,
|
||||
self.config.hidden_size)
|
||||
self.logits_processor = LogitsProcessor(config.vocab_size)
|
||||
self.sampler = Sampler()
|
||||
|
||||
|
||||
Reference in New Issue
Block a user