File tree Expand file tree Collapse file tree 2 files changed +7
-7
lines changed Expand file tree Collapse file tree 2 files changed +7
-7
lines changed Original file line number Diff line number Diff line change 66from dataclasses import asdict
77
88import pytest
9- import tpu_inference .envs as envs
109from vllm import LLM , EngineArgs , SamplingParams
1110
1211
1312@pytest .fixture (autouse = True )
1413def setup_new_model_design ():
1514 """Automatically set NEW_MODEL_DESIGN=True for all tests."""
16- os .environ ['NEW_MODEL_DESIGN' ] = '1 '
15+ os .environ ['NEW_MODEL_DESIGN' ] = 'True '
1716
1817
1918@pytest .fixture
@@ -174,8 +173,8 @@ def test_data_parallelism_correctness(
174173 This test compares outputs from a single-device run with data parallel runs
175174 to ensure correctness, including log probabilities.
176175 """
177- envs . environment_variables ['SKIP_JAX_PRECOMPILE' ] = lambda : True
178- envs . environment_variables ['VLLM_XLA_CHECK_RECOMPILATION' ] = lambda : False
176+ os . environ ['SKIP_JAX_PRECOMPILE' ] = '1'
177+ os . environ ['VLLM_XLA_CHECK_RECOMPILATION' ] = '0'
179178 model_name = "Qwen/Qwen2.5-1.5B-Instruct"
180179 # Use a smaller subset of prompts for correctness testing
181180 small_prompts = test_prompts [:10 ]
Original file line number Diff line number Diff line change 44# This script is a self-contained test that runs a single prompt and
55# compares the output to a known-good output.
66
7+ import os
78from dataclasses import asdict
89
9- import tpu_inference .envs as envs
1010from vllm import LLM , EngineArgs , SamplingParams
1111from vllm .assets .image import ImageAsset
1212from vllm .multimodal .image import convert_image_mode
@@ -24,8 +24,9 @@ def test_multi_modal_inference(monkeypatch):
2424 """
2525 Runs multi-modal inference and verifies the output.
2626 """
27- envs .environment_variables ['SKIP_JAX_PRECOMPILE' ] = lambda : True # Skip warmup to save time.
28- envs .environment_variables ['VLLM_XLA_CHECK_RECOMPILATION' ] = lambda : False # Allow compilation during execution.
27+ os .environ ['SKIP_JAX_PRECOMPILE' ] = '1' # Skip warmup to save time.
28+ os .environ [
29+ 'VLLM_XLA_CHECK_RECOMPILATION' ] = '0' # Allow compilation during execution.
2930
3031 monkeypatch .setenv ("VLLM_WORKER_MULTIPROC_METHOD" , "spawn" )
3132
You can’t perform that action at this time.
0 commit comments