A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
[Misc] Rev DeepEP (#27122)
Signed-off-by: Varun Sundar Rabindranath <vsundarr@redhat.com> Co-authored-by: Varun Sundar Rabindranath <vsundarr@redhat.com>
Varun Sundar Rabindranath committed
Oct 18, 2025 at 02:54 UTC
30a33b92eebe1ec89959ef31800f76abdaabb570
2 files changed
+2
-2
tools/ep_kernels/install_python_libraries.sh
+1
-1
@@ -119,7 +119,7 @@ popd
119
120
# build and install deepep, require pytorch installed
121
pushd $WORKSPACE
122
-clone_repo "https://github.com/deepseek-ai/DeepEP" "DeepEP" "setup.py" "e3908bf"
122
+clone_repo "https://github.com/deepseek-ai/DeepEP" "DeepEP" "setup.py" "73b6ea4"
123
cd DeepEP
124
export NVSHMEM_DIR=$WORKSPACE/nvshmem_install
125
$PIP_CMD install --no-build-isolation -vvv -e .
vllm/model_executor/layers/fused_moe/deepep_ll_prepare_finalize.py
+1
-1
@@ -52,7 +52,7 @@ class DeepEPLLPrepareAndFinalize(mk.FusedMoEPrepareAndFinalize):
52
# specific hidden sizes.
53
# NOTE: Keep this list sorted, maybe_roundup_layer_hidden_size depends
54
# on it.
55
- SUPPORTED_HIDDEN_SIZES = [2048, 2560, 4096, 5120, 6144, 7168, 8192]
55
+ SUPPORTED_HIDDEN_SIZES = [2048, 2560, 3072, 4096, 5120, 6144, 7168, 8192]
56
57
@staticmethod
58
def maybe_roundup_layer_hidden_size(hidden_size: int) -> int: