Update PyTorch to 2.9.0+cu129 (#24994)
Co-authored-by: Luka Govedič <ProExpertProg@users.noreply.github.com>
This commit is contained in:
@@ -109,7 +109,7 @@ def test_chatglm3_lora_tp4_fully_sharded_loras(chatglm3_lora_files):
|
||||
tensor_parallel_size=4,
|
||||
trust_remote_code=True,
|
||||
fully_sharded_loras=True,
|
||||
gpu_memory_utilization=0.85,
|
||||
gpu_memory_utilization=0.8,
|
||||
compilation_config=vllm.config.CompilationConfig( # Avoid OOM
|
||||
cudagraph_specialize_lora=False,
|
||||
),
|
||||
|
||||
@@ -3,6 +3,7 @@
|
||||
from typing import Any
|
||||
|
||||
import pytest
|
||||
import torch._dynamo.config as dynamo_config
|
||||
|
||||
from vllm import SamplingParams
|
||||
|
||||
@@ -12,6 +13,7 @@ from ...models.utils import check_outputs_equal
|
||||
MODEL = "Qwen/Qwen3-0.6B"
|
||||
|
||||
|
||||
@dynamo_config.patch(cache_size_limit=16)
|
||||
def test_preempt_and_async_scheduling_e2e(monkeypatch: pytest.MonkeyPatch):
|
||||
"""Test consistency of combos of async scheduling, preemption,
|
||||
uni/multiproc executor, and various sampling parameters."""
|
||||
|
||||
Reference in New Issue
Block a user