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

[Hardware][AMD][CI][Bugfix] Fix regressions from deprecated env vars (#32837)

Browse Source 
Signed-off-by: Matthew Wong <Matthew.Wong2@amd.com>
This commit is contained in:
Matt
2026-01-22 10:59:15 -06:00
committed by GitHub
parent fc37187a51
commit c517d8c934
3 changed files with 14 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
.buildkite/test-amd.yaml
View File

@@ -1475,7 +1475,7 @@ steps:
- tests/v1/kv_connector/nixl_integration/
commands:
- uv pip install --system -r /vllm-workspace/requirements/kv_connectors_rocm.txt
- bash v1/kv_connector/nixl_integration/config_sweep_accuracy_test.sh --attention-backend ROCM_ATTN
- ROCM_ATTN=1 bash v1/kv_connector/nixl_integration/config_sweep_accuracy_test.sh
- label: DP EP NixlConnector PD accuracy tests (Distributed) # 15min
mirror_hardwares: [amdexperimental, amdproduction]
@@ -1489,7 +1489,7 @@ steps:
- tests/v1/kv_connector/nixl_integration/
commands:
- uv pip install --system -r /vllm-workspace/requirements/kv_connectors_rocm.txt
- DP_EP=1 bash v1/kv_connector/nixl_integration/config_sweep_accuracy_test.sh --attention-backend ROCM_ATTN
- DP_EP=1 ROCM_ATTN=1 bash v1/kv_connector/nixl_integration/config_sweep_accuracy_test.sh
##### multi gpus test #####
##### A100 test #####

7
tests/v1/kv_connector/nixl_integration/config_sweep_accuracy_test.sh
View File

@@ -43,7 +43,12 @@ run_tests() {
}
# Run tests
run_tests "default backend" ""
if [[ -n "${ROCM_ATTN:-}" ]]; then
echo "ROCM_ATTN is set, running with --attention-backend ROCM_ATTN"
run_tests "ROCM_ATTN backend" "--attention-backend ROCM_ATTN"
else
run_tests "default backend" ""
fi
# Check if FLASHINFER is set (non-empty)
if [[ -n "${FLASHINFER:-}" ]]; then

9
vllm/platforms/rocm.py
View File

@@ -293,10 +293,13 @@ class RocmPlatform(Platform):
return AttentionBackendEnum.ROCM_AITER_FA.get_path()
# Priority 3: Check for ROCM_ATTN (prefill-decode split)
from vllm.config import get_current_vllm_config
from vllm.config import get_current_vllm_config_or_none
vllm_config = get_current_vllm_config()
if vllm_config.attention_config.use_prefill_decode_attention:
vllm_config = get_current_vllm_config_or_none()
if (
vllm_config is not None
and vllm_config.attention_config.use_prefill_decode_attention
):
logger.info("Using Rocm Attention backend.")
return AttentionBackendEnum.ROCM_ATTN.get_path()