Skip to content

Commit ab0c086

Browse files
author
Wei Zhao (Engrg-Hardware 1)
committed
disable AIPERF_USE_DYNAMO_CONV_AWARE_ROUTING
1 parent caa5603 commit ab0c086

5 files changed

Lines changed: 20 additions & 5 deletions

File tree

benchmarks/multi_node/srt-slurm-recipes/vllm/deepseek-v4/agentic/agg-gb200-tp8-agentic.yaml

Lines changed: 4 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -135,7 +135,10 @@ benchmark:
135135
TP: "8"
136136
EP_SIZE: "1"
137137
DP_ATTENTION: "false"
138-
AIPERF_USE_DYNAMO_CONV_AWARE_ROUTING: "1"
138+
# Disabled: with conv-aware routing, aiperf sends nvext.session_control, which
139+
# dynamo 6972dd2a rejects (400: unknown field `session_control`). Re-enable once
140+
# the aiperf submodule and dynamo agree on the conv-aware-routing field.
141+
AIPERF_USE_DYNAMO_CONV_AWARE_ROUTING: "0"
139142
AIPERF_DYNAMO_SESSION_TIMEOUT_SECONDS: "14400"
140143
AIPERF_DATASET_MMAP_CACHE_DIR: "/aiperf_mmap_cache"
141144
HF_HUB_CACHE: "/hf_hub_cache"

benchmarks/multi_node/srt-slurm-recipes/vllm/deepseek-v4/agentic/disagg-gb200-1p1d-dep8-dep12-agentic.yaml

Lines changed: 4 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -182,7 +182,10 @@ benchmark:
182182
RESULT_DIR: /logs/agentic
183183
PORT: "8000"
184184
IS_MULTINODE: "true"
185-
AIPERF_USE_DYNAMO_CONV_AWARE_ROUTING: "1"
185+
# Disabled: with conv-aware routing, aiperf sends nvext.session_control, which
186+
# dynamo 6972dd2a rejects (400: unknown field `session_control`). Re-enable once
187+
# the aiperf submodule and dynamo agree on the conv-aware-routing field.
188+
AIPERF_USE_DYNAMO_CONV_AWARE_ROUTING: "0"
186189
AIPERF_DYNAMO_SESSION_TIMEOUT_SECONDS: "14400"
187190
AIPERF_DATASET_MMAP_CACHE_DIR: "/aiperf_mmap_cache"
188191
HF_HUB_CACHE: "/hf_hub_cache"

benchmarks/multi_node/srt-slurm-recipes/vllm/deepseek-v4/agentic/disagg-gb200-1p1d-dep8-dep8-agentic.yaml

Lines changed: 4 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -182,7 +182,10 @@ benchmark:
182182
RESULT_DIR: /logs/agentic
183183
PORT: "8000"
184184
IS_MULTINODE: "true"
185-
AIPERF_USE_DYNAMO_CONV_AWARE_ROUTING: "1"
185+
# Disabled: with conv-aware routing, aiperf sends nvext.session_control, which
186+
# dynamo 6972dd2a rejects (400: unknown field `session_control`). Re-enable once
187+
# the aiperf submodule and dynamo agree on the conv-aware-routing field.
188+
AIPERF_USE_DYNAMO_CONV_AWARE_ROUTING: "0"
186189
AIPERF_DYNAMO_SESSION_TIMEOUT_SECONDS: "14400"
187190
AIPERF_DATASET_MMAP_CACHE_DIR: "/aiperf_mmap_cache"
188191
HF_HUB_CACHE: "/hf_hub_cache"

benchmarks/multi_node/srt-slurm-recipes/vllm/deepseek-v4/agentic/disagg-gb200-2p1d-dep8-dep12-agentic.yaml

Lines changed: 4 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -182,7 +182,10 @@ benchmark:
182182
RESULT_DIR: /logs/agentic
183183
PORT: "8000"
184184
IS_MULTINODE: "true"
185-
AIPERF_USE_DYNAMO_CONV_AWARE_ROUTING: "1"
185+
# Disabled: with conv-aware routing, aiperf sends nvext.session_control, which
186+
# dynamo 6972dd2a rejects (400: unknown field `session_control`). Re-enable once
187+
# the aiperf submodule and dynamo agree on the conv-aware-routing field.
188+
AIPERF_USE_DYNAMO_CONV_AWARE_ROUTING: "0"
186189
AIPERF_DYNAMO_SESSION_TIMEOUT_SECONDS: "14400"
187190
AIPERF_DATASET_MMAP_CACHE_DIR: "/aiperf_mmap_cache"
188191
HF_HUB_CACHE: "/hf_hub_cache"

benchmarks/multi_node/srt-slurm-recipes/vllm/deepseek-v4/agentic/disagg-gb200-3p1d-dep8-dep16-agentic.yaml

Lines changed: 4 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -182,7 +182,10 @@ benchmark:
182182
RESULT_DIR: /logs/agentic
183183
PORT: "8000"
184184
IS_MULTINODE: "true"
185-
AIPERF_USE_DYNAMO_CONV_AWARE_ROUTING: "1"
185+
# Disabled: with conv-aware routing, aiperf sends nvext.session_control, which
186+
# dynamo 6972dd2a rejects (400: unknown field `session_control`). Re-enable once
187+
# the aiperf submodule and dynamo agree on the conv-aware-routing field.
188+
AIPERF_USE_DYNAMO_CONV_AWARE_ROUTING: "0"
186189
AIPERF_DYNAMO_SESSION_TIMEOUT_SECONDS: "14400"
187190
AIPERF_DATASET_MMAP_CACHE_DIR: "/aiperf_mmap_cache"
188191
HF_HUB_CACHE: "/hf_hub_cache"

0 commit comments

Comments
 (0)