[CI] Disable flash_attn backend for spec decode (vllm-project#5286)

blinkbear · Jun 6, 2024 · 0f13bd2 · 0f13bd2
1 parent d12b1b7
commit 0f13bd2
Showing 1 changed file with 5 additions and 2 deletions.
diff --git a/.buildkite/test-pipeline.yaml b/.buildkite/test-pipeline.yaml
@@ -45,7 +45,7 @@ steps:
   - TEST_DIST_MODEL=meta-llama/Llama-2-7b-hf DISTRIBUTED_EXECUTOR_BACKEND=mp pytest -v -s distributed/test_basic_distributed_correctness.py
   - TEST_DIST_MODEL=facebook/opt-125m DISTRIBUTED_EXECUTOR_BACKEND=mp pytest -v -s distributed/test_chunked_prefill_distributed.py
   - TEST_DIST_MODEL=meta-llama/Llama-2-7b-hf DISTRIBUTED_EXECUTOR_BACKEND=mp pytest -v -s distributed/test_chunked_prefill_distributed.py
-  - pytest -v -s spec_decode/e2e/test_integration_dist.py 
+  - pytest -v -s spec_decode/e2e/test_integration_dist.py
 
 - label: Distributed Tests (Multiple Groups)
   #mirror_hardwares: [amd]
@@ -124,7 +124,10 @@ steps:
 
 - label: Speculative decoding tests
   #mirror_hardwares: [amd]
-  command: pytest -v -s spec_decode
+  commands:
+    # See https://github.com/vllm-project/vllm/issues/5152
+    - export VLLM_ATTENTION_BACKEND=XFORMERS
+    - pytest -v -s spec_decode
 
 - label: LoRA Test %N
   #mirror_hardwares: [amd]