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 9, 02:30 UTC
move spec decode slow test to test_areas.yaml (#33365)
Signed-off-by: shanjiaz <zsjwpianpian@gmail.com>
shanjiaz committed
Feb 2, 2026 at 09:28 UTC
d95b4be47a3b07d38bccd34b282321f479915de2
2 files changed
+17
-3
.buildkite/test_areas/misc.yaml
+16
-2
@@ -16,7 +16,7 @@ steps:
16
- pytest -v -s v1/sample
17
- pytest -v -s v1/logits_processors
18
- pytest -v -s v1/worker
19
- - pytest -v -s v1/spec_decode
19
+ - pytest -v -s -m 'not slow_test' v1/spec_decode
20
- pytest -v -s -m 'not cpu_test' v1/kv_connector/unit
21
- pytest -v -s -m 'not cpu_test' v1/metrics
22
- pytest -v -s v1/test_oracle.py
@@ -166,4 +166,18 @@ steps:
166
- export VLLM_WORKER_MULTIPROC_METHOD=spawn
167
- pip install pytest-timeout pytest-forked
168
- pytest -v -s v1/determinism/test_batch_invariance.py
169
- - pytest -v -s v1/determinism/test_rms_norm_batch_invariant.py
\ No newline at end of file
169
+ - pytest -v -s v1/determinism/test_rms_norm_batch_invariant.py
170
+
171
+- label: Acceptance Length Test (Large Models) # optional
172
+ timeout_in_minutes: 25
173
+ gpu: h100
174
+ optional: true
175
+ num_gpus: 1
176
+ working_dir: "/vllm-workspace/tests"
177
+ source_file_dependencies:
178
+ - vllm/v1/spec_decode/
179
+ - vllm/model_executor/models/mlp_speculator.py
180
+ - tests/v1/spec_decode/test_acceptance_length.py
181
+ commands:
182
+ - export VLLM_ALLOW_INSECURE_SERIALIZATION=1
183
+ - pytest -v -s v1/spec_decode/test_acceptance_length.py -m slow_test
tests/v1/spec_decode/test_acceptance_length.py
+1
-1
@@ -35,7 +35,7 @@ class Eagle3ModelConfig:
35
id: str = ""
36
# Backends that are incompatible with this model (will be skipped)
37
excluded_backends: set[AttentionBackendEnum] = field(default_factory=set)
38
- # Pytest marks for this configuration (e.g., pytest.mark.optional)
38
+ # Pytest marks for this configuration
39
marks: list = field(default_factory=list)
40
# Custom relative tolerance (defaults to DEFAULT_RTOL if None)
41
rtol: float | None = None