xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-05-24 13:51:18 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Bugfix] Fix OOM tests in initialization test (#21921)

Browse Source 
Signed-off-by: Isotr0py <2037008807@qq.com>
This commit is contained in:
Isotr0py 2025-07-30 22:35:47 +08:00 committed by GitHub
parent 88edf5994c
commit 6e599eebe8
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 9 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
tests/models/test_initialization.py
View File

@ -33,12 +33,6 @@ def can_initialize(model_arch: str, monkeypatch: pytest.MonkeyPatch,
model_info.check_available_online(on_fail="skip") model_info.check_available_online(on_fail="skip")
model_info.check_transformers_version(on_fail="skip") model_info.check_transformers_version(on_fail="skip")
# FIXME: Possible memory leak in the previous tests?
if model_arch in ("Glm4vForConditionalGeneration",
"GraniteSpeechForConditionalGeneration",
"KimiVLForConditionalGeneration"):
pytest.skip("Avoid OOM")
if model_arch in ("Llama4ForCausalLM", "EagleLlama4ForCausalLM"): if model_arch in ("Llama4ForCausalLM", "EagleLlama4ForCausalLM"):
from vllm.model_executor.models.llama4 import Llama4ForCausalLM from vllm.model_executor.models.llama4 import Llama4ForCausalLM
from vllm.model_executor.models.registry import ModelRegistry from vllm.model_executor.models.registry import ModelRegistry
@ -87,6 +81,14 @@ def can_initialize(model_arch: str, monkeypatch: pytest.MonkeyPatch,
"num_hidden_layers": 1, "num_hidden_layers": 1,
}) })
# e.g.: Qwen/Qwen2-Audio-7B-Instruct
if hasattr(hf_config, "audio_config"):
hf_config.audio_config.update({
"num_layers": 1,
"num_hidden_layers": 1,
"encoder_layers": 1,
})
return hf_config return hf_config
# Avoid calling model.forward() # Avoid calling model.forward()

1
vllm/model_executor/models/glm4_1v.py
View File

@ -1275,6 +1275,7 @@ class Glm4vForConditionalGeneration(nn.Module, SupportsMultiModal,
vllm_config=vllm_config, vllm_config=vllm_config,
prefix=maybe_prefix(prefix, ""), prefix=maybe_prefix(prefix, ""),
architectures=["Glm4ForCausalLM"], architectures=["Glm4ForCausalLM"],
hf_config=self.config.get_text_config(),
) )
self.make_empty_intermediate_tensors = ( self.make_empty_intermediate_tensors = (