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

[Core] Enable inputs_embeds_size separate from hidden_size (#29741)

Browse Source 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2025-11-30 17:31:12 +08:00
committed by GitHub
parent 47539cfd3e
commit 64bc09ba27
10 changed files with 123 additions and 18 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
vllm/v1/worker/gpu/model_runner.py
View File

@@ -98,7 +98,7 @@ class GPUModelRunner(LoRAModelRunnerMixin, KVConnectorModelRunnerMixin):
self.max_model_len = self.model_config.max_model_len
self.max_num_tokens = self.scheduler_config.max_num_batched_tokens
self.max_num_reqs = self.scheduler_config.max_num_seqs
self.hidden_size = self.model_config.get_hidden_size()
self.inputs_embeds_size = self.model_config.get_inputs_embeds_size()
self.dp_size = self.parallel_config.data_parallel_size
self.dp_rank = self.parallel_config.data_parallel_rank
@@ -134,7 +134,7 @@ class GPUModelRunner(LoRAModelRunnerMixin, KVConnectorModelRunnerMixin):
self.input_buffers = InputBuffers(
max_num_reqs=self.max_num_reqs,
max_num_tokens=self.max_num_tokens,
hidden_size=self.hidden_size,
inputs_embeds_size=self.inputs_embeds_size,
vocab_size=self.vocab_size,
dtype=self.dtype,
device=self.device,