[CI/Build] Temporary fix to LM Eval Small Models (vllm-project#28324)

Signed-off-by: zhewenli <zhewenli@meta.com>

Author: zhewenl

.buildkite/test-pipeline.yaml

@@ -1253,7 +1253,7 @@ steps:
  - pytest -v -s tests/compile/test_fusions_e2e.py::test_tp2_attn_quant_allreduce_rmsnorm
  - pytest -v -s tests/distributed/test_context_parallel.py
  - CUDA_VISIBLE_DEVICES=1,2 VLLM_ALL2ALL_BACKEND=deepep_high_throughput VLLM_USE_DEEP_GEMM=1 VLLM_LOGGING_LEVEL=DEBUG python3 examples/offline_inference/data_parallel.py --model Qwen/Qwen1.5-MoE-A2.7B --tp-size=1 --dp-size=2 --max-model-len 2048
- - pytest -v -s tests/v1/distributed/test_dbo.py 
+ - pytest -v -s tests/v1/distributed/test_dbo.py
 
 ##### B200 test #####
 - label: Distributed Tests (B200) # optional

tests/evals/gsm8k/configs/Qwen1.5-MoE-W4A16-CT.yaml

@@ -2,4 +2,7 @@ model_name: "nm-testing/Qwen1.5-MoE-A2.7B-Chat-quantized.w4a16"
 accuracy_threshold: 0.45
 num_questions: 1319
 num_fewshot: 5
-max_model_len: 4096
+max_model_len: 4096
+# Duo stream incompatabilbe with this model: https://github.com/vllm-project/vllm/issues/28220
+env:
+ VLLM_DISABLE_SHARED_EXPERTS_STREAM: "1"

tests/evals/gsm8k/test_gsm8k_correctness.py

@@ -62,9 +62,11 @@ def test_gsm8k_correctness_param(config_filename, tp_size):
  str(tp_size),
  ]
 
+ env_dict = eval_config.get("env", None)
+
  # Launch server and run evaluation
  with RemoteOpenAIServer(
- eval_config["model_name"], server_args, max_wait_seconds=480
+ eval_config["model_name"], server_args, env_dict=env_dict, max_wait_seconds=480
  ) as remote_server:
  server_url = remote_server.url_for("v1")