Gahow Wang
445e491123
third_party/vllm/ now tracked in git for direct patch management.
Based on vLLM v0.18.1 release with one patch applied:
vllm/v1/core/sched/scheduler.py:
Replace fatal assert with graceful skip when KV transfer callback
arrives for an already-aborted request during PD disaggregated serving.
Future vLLM modifications should be made directly in third_party/vllm/
and committed normally. The patches/ directory is kept as documentation
of what changed from upstream.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
39 lines
960 B
YAML
39 lines
960 B
YAML
group: Expert Parallelism
|
|
depends_on:
|
|
- image-build
|
|
steps:
|
|
- label: EPLB Algorithm
|
|
timeout_in_minutes: 15
|
|
working_dir: "/vllm-workspace/tests"
|
|
source_file_dependencies:
|
|
- vllm/distributed/eplb
|
|
- tests/distributed/test_eplb_algo.py
|
|
commands:
|
|
- pytest -v -s distributed/test_eplb_algo.py
|
|
|
|
- label: EPLB Execution
|
|
timeout_in_minutes: 20
|
|
working_dir: "/vllm-workspace/tests"
|
|
num_devices: 4
|
|
source_file_dependencies:
|
|
- vllm/distributed/eplb
|
|
- tests/distributed/test_eplb_execute.py
|
|
commands:
|
|
- pytest -v -s distributed/test_eplb_execute.py
|
|
- pytest -v -s distributed/test_eplb_spec_decode.py
|
|
|
|
- label: Elastic EP Scaling Test
|
|
timeout_in_minutes: 20
|
|
device: b200
|
|
optional: true
|
|
working_dir: "/vllm-workspace/tests"
|
|
num_devices: 4
|
|
source_file_dependencies:
|
|
- vllm/distributed/
|
|
- vllm/engine/
|
|
- vllm/executor/
|
|
- vllm/compilation/
|
|
- tests/distributed/
|
|
commands:
|
|
- pytest -v -s distributed/test_elastic_ep.py
|