Skip to content

Commit 9d53384

Browse files
authored
Dockerfile.rocm.ubi: force using V1 engine (opendatahub-io#244)
V0 engine on `v0.9.1` has some pretty bad accuracy regressions which are solved when forcing using the V1 engine. https://issues.redhat.com/browse/AIPCC-3159
2 parents fca4c64 + d2cb31e commit 9d53384

File tree

1 file changed

+1
-0
lines changed

1 file changed

+1
-0
lines changed

Dockerfile.rocm.ubi

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -129,6 +129,7 @@ ENV HF_HUB_OFFLINE=1 \
129129
TOKENIZERS_PARALLELISM=false \
130130
RAY_EXPERIMENTAL_NOSET_ROCR_VISIBLE_DEVICES=1 \
131131
VLLM_USE_TRITON_FLASH_ATTN=0 \
132+
VLLM_USE_V1=1 \
132133
HIP_FORCE_DEV_KERNARG=1 \
133134
OUTLINES_CACHE_DIR=/tmp/outlines \
134135
NUMBA_CACHE_DIR=/tmp/numba \

0 commit comments

Comments
 (0)