biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[torch.compile] Reorganize vllm/compilation and tests/compile (0/N for vLLM IR) (#33731)

Browse Source 
Signed-off-by: Luka Govedič <lgovedic@redhat.com>
Signed-off-by: ProExpertProg <luka.govedic@gmail.com>
Signed-off-by: Luka Govedič <ProExpertProg@users.noreply.github.com>
This commit is contained in:
Luka Govedič
2026-02-06 07:19:49 -05:00
committed by GitHub
parent f79d9dce16
commit ac32e66cf9
47 changed files with 717 additions and 651 deletions
Download Patch File Download Diff File Expand all files Collapse all files

42
.buildkite/test_areas/compile.yaml
View File

@@ -2,7 +2,7 @@ group: Compile
depends_on:
- image-build
steps:
- label: Sequence Parallel Tests (2 GPUs)
- label: Sequence Parallel Correctness Tests (2 GPUs)
timeout_in_minutes: 50
working_dir: "/vllm-workspace/"
num_devices: 2
@@ -11,12 +11,12 @@ steps:
- vllm/compilation/
- vllm/v1/worker/
- vllm/v1/cudagraph_dispatcher.py
- tests/distributed/test_sequence_parallel.py
- tests/compile/correctness_e2e/test_sequence_parallel.py
commands:
- export VLLM_TEST_CLEAN_GPU_MEMORY=1
- pytest -v -s tests/distributed/test_sequence_parallel.py
- pytest -v -s tests/compile/correctness_e2e/test_sequence_parallel.py
- label: Sequence Parallel Tests (2xH100)
- label: Sequence Parallel Correctness Tests (2xH100)
timeout_in_minutes: 50
working_dir: "/vllm-workspace/"
device: h100
@@ -24,24 +24,30 @@ steps:
num_devices: 2
commands:
- export VLLM_TEST_CLEAN_GPU_MEMORY=1
- pytest -v -s tests/distributed/test_sequence_parallel.py
- pytest -v -s tests/compile/correctness_e2e/test_sequence_parallel.py
- label: AsyncTP Correctness Tests (2xH100)
timeout_in_minutes: 50
working_dir: "/vllm-workspace/"
device: h100
optional: true
num_devices: 2
commands:
- export VLLM_TEST_CLEAN_GPU_MEMORY=1
- pytest -v -s tests/compile/correctness_e2e/test_async_tp.py
- label: Distributed Compile Unit Tests (2xH100)
timeout_in_minutes: 40
timeout_in_minutes: 20
working_dir: "/vllm-workspace/"
device: h100
num_devices: 2
source_file_dependencies:
- vllm/compilation/
- vllm/model_executor/layers
- tests/compile/distributed/test_fusion_all_reduce.py
- tests/compile/distributed/test_sequence_parallelism.py
- tests/compile/distributed/test_async_tp.py
- tests/compile/passes/distributed/
commands:
- export VLLM_TEST_CLEAN_GPU_MEMORY=1
- pytest -v -s tests/compile/distributed/test_fusion_all_reduce.py
- pytest -v -s tests/compile/distributed/test_sequence_parallelism.py
- pytest -v -s tests/compile/distributed/test_async_tp.py
- pytest -s -v tests/compile/passes/distributed
- label: Fusion and Compile Unit Tests (B200)
timeout_in_minutes: 20
@@ -55,17 +61,17 @@ steps:
- vllm/model_executor/layers/attention/attention.py
- vllm/v1/attention/backends/flashinfer.py
- vllm/compilation/ # TODO(luka) limit to vllm/compilation/passes
- tests/compile/test_fusion_attn.py
- tests/compile/test_silu_mul_quant_fusion.py
- tests/compile/distributed/test_fusion_all_reduce.py
- tests/compile/passes/test_fusion_attn.py
- tests/compile/passes/test_silu_mul_quant_fusion.py
- tests/compile/passes/distributed/test_fusion_all_reduce.py
- tests/compile/fullgraph/test_full_graph.py
commands:
# b200 runners are limited, so we limit the tests to the minimum set only supported on Blackwell
- nvidia-smi
- pytest -v -s tests/compile/test_fusion_attn.py -k FLASHINFER
- pytest -v -s tests/compile/test_silu_mul_quant_fusion.py
- pytest -v -s tests/compile/passes/test_fusion_attn.py -k FLASHINFER
- pytest -v -s tests/compile/passes/test_silu_mul_quant_fusion.py
# this runner has 2 GPUs available even though num_devices=2 is not set
- pytest -v -s tests/compile/distributed/test_fusion_all_reduce.py
- pytest -v -s tests/compile/passes/distributed/test_fusion_all_reduce.py
# test_fp8_kv_scale_compile requires FlashAttention (not supported on default L4/L40)
# TODO(luka) move to H100 once pass tests run on H100
- pytest -v -s tests/compile/fullgraph/test_full_graph.py::test_fp8_kv_scale_compile