fix profile run for vl model (#5136)

### What this PR does / why we need it?

### Does this PR introduce _any_ user-facing change?

### How was this patch tested?

- vLLM version: v0.12.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/ad32e3e19ccf0526cb6744a5fed09a138a5fb2f9

---------

Signed-off-by: 李少鹏 <lishaopeng21@huawei.com>
This commit is contained in:
shaopeng-666 authored and GitHub committed 2025-12-17 23:51:31 +08:00
1 parent 43d974c6f7
commit 39bdd4cfaa
4 files changed
+1 -3

No files matched your search

@@ -6,7 +6,6 @@ tasks:
metrics:
- name: "acc,none"
value: 0.58
max_model_len: 8192
tensor_parallel_size: 2
gpu_memory_utilization: 0.7
enable_expert_parallel: True
@@ -6,6 +6,5 @@ tasks:
metrics:
- name: "acc,none"
value: 0.55
max_model_len: 8192
batch_size: 32
gpu_memory_utilization: 0.7
-1
View File
@@ -39,7 +39,6 @@ def test_multimodal_vl(prompt_template):
images = [image] * len(img_questions)
prompts = prompt_template(img_questions)
with VllmRunner("Qwen/Qwen3-VL-8B-Instruct",
max_model_len=4096,
mm_processor_kwargs={
"min_pixels": 28 * 28,
"max_pixels": 1280 * 28 * 28,
+1
View File
@@ -232,6 +232,7 @@ class NPUWorker(WorkerBase):
# Init ModelRunner here, so that we have access to self.device.
self.model_runner = NPUModelRunner(self.vllm_config, self.device)
@torch.inference_mode()
def determine_available_memory(self) -> int:
# Profile the memory usage of the model and get the maximum number of
# cache blocks that can be allocated with the remaining free memory.