A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 13, 14:30 UTC
[CI/Build] Temporary fix to LM Eval Small Models (#28324)
Signed-off-by: zhewenli <zhewenli@meta.com>
Zhewen Li committed
Nov 9, 2025 at 13:08 UTC
a65a934ebeeddd8ecfc8034d784f3e284a365971
3 files changed
+8
-3
.buildkite/test-pipeline.yaml
+1
-1
@@ -1253,7 +1253,7 @@ steps:
1253
- pytest -v -s tests/compile/test_fusions_e2e.py::test_tp2_attn_quant_allreduce_rmsnorm
1254
- pytest -v -s tests/distributed/test_context_parallel.py
1255
- 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
1256
- - pytest -v -s tests/v1/distributed/test_dbo.py
1256
+ - pytest -v -s tests/v1/distributed/test_dbo.py
1257
1258
##### B200 test #####
1259
- label: Distributed Tests (B200) # optional
tests/evals/gsm8k/configs/Qwen1.5-MoE-W4A16-CT.yaml
+4
-1
@@ -2,4 +2,7 @@ model_name: "nm-testing/Qwen1.5-MoE-A2.7B-Chat-quantized.w4a16"
2
accuracy_threshold: 0.45
3
num_questions: 1319
4
num_fewshot: 5
5
-max_model_len: 4096
\ No newline at end of file
5
+max_model_len: 4096
6
+# Duo stream incompatabilbe with this model: https://github.com/vllm-project/vllm/issues/28220
7
+env:
8
+ VLLM_DISABLE_SHARED_EXPERTS_STREAM: "1"
tests/evals/gsm8k/test_gsm8k_correctness.py
+3
-1
@@ -62,9 +62,11 @@ def test_gsm8k_correctness_param(config_filename, tp_size):
62
str(tp_size),
63
]
64
65
+ env_dict = eval_config.get("env", None)
66
+
67
# Launch server and run evaluation
68
with RemoteOpenAIServer(
67
- eval_config["model_name"], server_args, max_wait_seconds=480
69
+ eval_config["model_name"], server_args, env_dict=env_dict, max_wait_seconds=480
70
) as remote_server:
71
server_url = remote_server.url_for("v1")
72