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

[Model] Extend collect_children and no_init_weights contexts (#32757)

Browse Source 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2026-01-22 16:20:27 +08:00
committed by GitHub
parent 1bf1a34b19
commit 2b8a38b6d6
20 changed files with 444 additions and 257 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
vllm/model_executor/models/nano_nemotron_vl.py
View File

@@ -1513,7 +1513,7 @@ class NemotronH_Nano_VL_V2(
self.video_pruning_rate = multimodal_config.video_pruning_rate
with self._mark_language_model(vllm_config):
self.language_model = language_model = init_vllm_registered_model(
self.language_model = init_vllm_registered_model(
vllm_config=vllm_config,
hf_config=config.text_config,
prefix=maybe_prefix(prefix, "language_model"),
@@ -1542,7 +1542,7 @@ class NemotronH_Nano_VL_V2(
ReLUSquaredActivation(),
nn.Linear(vision_projection_hidden_size, llm_hidden_size, bias=False),
)
self.mlp1 = mlp1.to(language_model.config.dtype)
self.mlp1 = mlp1.to(self.language_model.config.dtype)
self.config = config
self.model_config = vllm_config.model_config