[CI] Shard Quantization job into 4 parallel shards (≤30 min target) (#52328)
Kevin H. Luu committed
Aug 14, 2026 at 15:11 UTC
d87ef456abe54565cef476637df5e585bfddfb37
1 file changed
+4
-3
.buildkite/test_areas/quantization.yaml
+4
-3
@@ -2,10 +2,11 @@ group: Quantization
2
depends_on:
3
- image-build
4
steps:
5
-- label: Quantization
5
+- label: Quantization %N
6
device: h200_35gb
7
key: quantization
8
- timeout_in_minutes: 75
8
+ timeout_in_minutes: 40
9
+ parallelism: 4
10
env:
11
VLLM_USE_V2_MODEL_RUNNER: "0"
12
source_file_dependencies:
@@ -18,7 +19,7 @@ steps:
19
- uv pip install --system conch-triton-kernels
20
# The SM90-only checkpoint currently contains a removed weight_chan_scale
21
# parameter. It was not exercised by the previous L4 job.
21
- - VLLM_TEST_FORCE_LOAD_FORMAT=auto pytest -v -s quantization/ --ignore quantization/test_blackwell_moe.py -k 'not test_compressed_tensors_w4a8_fp8'
22
+ - VLLM_TEST_FORCE_LOAD_FORMAT=auto pytest -v -s quantization/ --ignore quantization/test_blackwell_moe.py -k 'not test_compressed_tensors_w4a8_fp8' --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT
23
24
- label: Quantized Fusions
25
device: h200_35gb