biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Model] Composite weight loading for multimodal Qwen2 (#10944)

Browse Source 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2024-12-07 22:22:52 +08:00
committed by GitHub
parent b26b4cd03c
commit bf0e382e16
7 changed files with 148 additions and 206 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
vllm/model_executor/models/qwen2.py
View File

@@ -444,14 +444,17 @@ class Qwen2ForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
self.model = Qwen2Model(vllm_config=vllm_config,
prefix=maybe_prefix(prefix, "model"))
if config.tie_word_embeddings:
self.lm_head = self.model.embed_tokens
if get_pp_group().is_last_rank:
if config.tie_word_embeddings:
self.lm_head = self.model.embed_tokens
else:
self.lm_head = ParallelLMHead(config.vocab_size,
config.hidden_size,
quant_config=quant_config,
prefix=maybe_prefix(
prefix, "lm_head"))
else:
self.lm_head = ParallelLMHead(config.vocab_size,
config.hidden_size,
quant_config=quant_config,
prefix=maybe_prefix(
prefix, "lm_head"))
self.lm_head = PPMissingLayer()
self.logits_processor = LogitsProcessor(config.vocab_size)
self.sampler = get_sampler()