From dc27fa1d0ba99c3ffeadc1777b6fe18f9542096e Mon Sep 17 00:00:00 2001 From: Kedar Potdar <115327600+kedarpotdar-nv@users.noreply.github.com> Date: Wed, 25 Mar 2026 18:10:12 -0700 Subject: [PATCH 01/16] Update nvidia-master.yaml --- .github/configs/nvidia-master.yaml | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-) diff --git a/.github/configs/nvidia-master.yaml b/.github/configs/nvidia-master.yaml index f4570fd2c..b98d8ad37 100644 --- a/.github/configs/nvidia-master.yaml +++ b/.github/configs/nvidia-master.yaml @@ -3208,18 +3208,18 @@ minimaxm2.5-fp8-b200-vllm: - isl: 1024 osl: 1024 search-space: - - { tp: 2, conc-start: 4, conc-end: 64 } - - { tp: 4, conc-start: 4, conc-end: 64 } + - { tp: 2, conc-start: 4, conc-end: 256 } + - { tp: 4, conc-start: 4, conc-end: 256 } - isl: 1024 osl: 8192 search-space: - - { tp: 2, conc-start: 4, conc-end: 64 } - - { tp: 4, conc-start: 4, conc-end: 64 } + - { tp: 2, conc-start: 4, conc-end: 256 } + - { tp: 4, conc-start: 4, conc-end: 256 } - isl: 8192 osl: 1024 search-space: - - { tp: 2, conc-start: 4, conc-end: 64 } - - { tp: 4, conc-start: 4, conc-end: 64 } + - { tp: 2, conc-start: 4, conc-end: 256 } + - { tp: 4, conc-start: 4, conc-end: 256 } gptoss-fp4-h100-vllm: image: vllm/vllm-openai:v0.15.1 From 9ee612f268f99be712ecdb23c5dedd0354c23885 Mon Sep 17 00:00:00 2001 From: Kedar Potdar Date: Wed, 25 Mar 2026 18:18:54 -0700 Subject: [PATCH 02/16] vllm version bump --- .github/configs/nvidia-master.yaml | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) diff --git a/.github/configs/nvidia-master.yaml b/.github/configs/nvidia-master.yaml index b98d8ad37..d990b3ae1 100644 --- a/.github/configs/nvidia-master.yaml +++ b/.github/configs/nvidia-master.yaml @@ -3197,7 +3197,7 @@ gptoss-fp4-b200-vllm: - { tp: 8, conc-start: 4, conc-end: 4 } minimaxm2.5-fp8-b200-vllm: - image: vllm/vllm-openai:v0.17.0-cu130 + image: vllm/vllm-openai:v0.18.0-cu130 model: MiniMaxAI/MiniMax-M2.5 model-prefix: minimaxm2.5 runner: b200 @@ -3210,16 +3210,19 @@ minimaxm2.5-fp8-b200-vllm: search-space: - { tp: 2, conc-start: 4, conc-end: 256 } - { tp: 4, conc-start: 4, conc-end: 256 } + - { tp: 4, ep: 4, conc-start: 32, conc-end: 256 } - isl: 1024 osl: 8192 search-space: - { tp: 2, conc-start: 4, conc-end: 256 } - { tp: 4, conc-start: 4, conc-end: 256 } + - { tp: 4, ep: 4, conc-start: 32, conc-end: 256 } - isl: 8192 osl: 1024 search-space: - { tp: 2, conc-start: 4, conc-end: 256 } - { tp: 4, conc-start: 4, conc-end: 256 } + - { tp: 4, ep: 4, conc-start: 32, conc-end: 256 } gptoss-fp4-h100-vllm: image: vllm/vllm-openai:v0.15.1 From a27edeecd6821374f0bf80a1a50b15ca2132161c Mon Sep 17 00:00:00 2001 From: Kedar Potdar Date: Wed, 25 Mar 2026 18:20:09 -0700 Subject: [PATCH 03/16] add perf changelog --- perf-changelog.yaml | 7 +++++++ 1 file changed, 7 insertions(+) diff --git a/perf-changelog.yaml b/perf-changelog.yaml index 03fb6e082..afce65264 100644 --- a/perf-changelog.yaml +++ b/perf-changelog.yaml @@ -1068,3 +1068,10 @@ - "dsr1-fp8-h200-sglang: v0.5.9-cu129-amd64 → v0.5.9-cu130" pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/943 +- config-keys: + - minimaxm2.5-fp8-b200-vllm + description: + - "Update vLLM image from v0.17.0 to v0.18.0 for MiniMax-M2.5 FP8 B200" + - "Add tp4 ep4 search-space entries (conc 32-256) for all seq-len configs" + pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/947 + From a33dc21986e2f1f9fe0940d42df104ad439b60fd Mon Sep 17 00:00:00 2001 From: Kedar Potdar Date: Fri, 27 Mar 2026 12:19:32 -0700 Subject: [PATCH 04/16] update search space and configs --- .github/configs/nvidia-master.yaml | 5 ++--- benchmarks/single_node/minimaxm2.5_fp8_b200.sh | 4 +++- 2 files changed, 5 insertions(+), 4 deletions(-) diff --git a/.github/configs/nvidia-master.yaml b/.github/configs/nvidia-master.yaml index d990b3ae1..e9f7f7603 100644 --- a/.github/configs/nvidia-master.yaml +++ b/.github/configs/nvidia-master.yaml @@ -3210,19 +3210,18 @@ minimaxm2.5-fp8-b200-vllm: search-space: - { tp: 2, conc-start: 4, conc-end: 256 } - { tp: 4, conc-start: 4, conc-end: 256 } - - { tp: 4, ep: 4, conc-start: 32, conc-end: 256 } + - { tp: 4, ep: 4, conc-start: 64, conc-end: 64 } + - { tp: 8, conc-start: 4, conc-end: 8 } - isl: 1024 osl: 8192 search-space: - { tp: 2, conc-start: 4, conc-end: 256 } - { tp: 4, conc-start: 4, conc-end: 256 } - - { tp: 4, ep: 4, conc-start: 32, conc-end: 256 } - isl: 8192 osl: 1024 search-space: - { tp: 2, conc-start: 4, conc-end: 256 } - { tp: 4, conc-start: 4, conc-end: 256 } - - { tp: 4, ep: 4, conc-start: 32, conc-end: 256 } gptoss-fp4-h100-vllm: image: vllm/vllm-openai:v0.15.1 diff --git a/benchmarks/single_node/minimaxm2.5_fp8_b200.sh b/benchmarks/single_node/minimaxm2.5_fp8_b200.sh index 2e5aa4b24..066a870fd 100755 --- a/benchmarks/single_node/minimaxm2.5_fp8_b200.sh +++ b/benchmarks/single_node/minimaxm2.5_fp8_b200.sh @@ -24,8 +24,8 @@ hf download "$MODEL" SERVER_LOG=/workspace/server.log PORT=${PORT:-8888} -export VLLM_USE_FLASHINFER_MOE_FP8=0 export VLLM_MOE_USE_DEEP_GEMM=0 +export VLLM_FLASHINFER_ALLREDUCE_BACKEND=mnnvl if [ "$EP_SIZE" -ge 1 ]; then EP=" --enable-expert-parallel" @@ -43,6 +43,8 @@ $EP \ --gpu-memory-utilization 0.95 \ --max-model-len $MAX_MODEL_LEN \ --block-size=32 \ +--kv-cache-dtype fp8 \ +--stream-interval 20 --no-enable-prefix-caching \ --trust-remote-code > $SERVER_LOG 2>&1 & SERVER_PID=$! From 89acdf2ff13f6ca69f5794c8998e7b5c87636584 Mon Sep 17 00:00:00 2001 From: Kedar Potdar Date: Fri, 27 Mar 2026 13:01:29 -0700 Subject: [PATCH 05/16] fix typo in VLLM_USE_DEEP_GEMM --- benchmarks/single_node/minimaxm2.5_fp8_b200.sh | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/benchmarks/single_node/minimaxm2.5_fp8_b200.sh b/benchmarks/single_node/minimaxm2.5_fp8_b200.sh index 066a870fd..0b4151e17 100755 --- a/benchmarks/single_node/minimaxm2.5_fp8_b200.sh +++ b/benchmarks/single_node/minimaxm2.5_fp8_b200.sh @@ -24,7 +24,7 @@ hf download "$MODEL" SERVER_LOG=/workspace/server.log PORT=${PORT:-8888} -export VLLM_MOE_USE_DEEP_GEMM=0 +export VLLM_USE_DEEP_GEMM=0 export VLLM_FLASHINFER_ALLREDUCE_BACKEND=mnnvl if [ "$EP_SIZE" -ge 1 ]; then From 83706aee88e68d72e7ba21d8ce9d96f1c5a13e0e Mon Sep 17 00:00:00 2001 From: "claude[bot]" <41898282+claude[bot]@users.noreply.github.com> Date: Sun, 29 Mar 2026 02:26:13 +0000 Subject: [PATCH 06/16] Remove ISL 1024 / OSL 8192 seq-len config for minimaxm2.5-fp8-b200-vllm Co-authored-by: functionstackx --- .github/configs/nvidia-master.yaml | 5 ----- perf-changelog.yaml | 1 + 2 files changed, 1 insertion(+), 5 deletions(-) diff --git a/.github/configs/nvidia-master.yaml b/.github/configs/nvidia-master.yaml index 2309314c6..292169457 100644 --- a/.github/configs/nvidia-master.yaml +++ b/.github/configs/nvidia-master.yaml @@ -3098,11 +3098,6 @@ minimaxm2.5-fp8-b200-vllm: - { tp: 4, conc-start: 4, conc-end: 256 } - { tp: 4, ep: 4, conc-start: 64, conc-end: 64 } - { tp: 8, conc-start: 4, conc-end: 8 } - - isl: 1024 - osl: 8192 - search-space: - - { tp: 2, conc-start: 4, conc-end: 256 } - - { tp: 4, conc-start: 4, conc-end: 256 } - isl: 8192 osl: 1024 search-space: diff --git a/perf-changelog.yaml b/perf-changelog.yaml index 7e0ccf97b..90d1035c0 100644 --- a/perf-changelog.yaml +++ b/perf-changelog.yaml @@ -1195,4 +1195,5 @@ description: - "Update vLLM image from v0.17.0 to v0.18.0 for MiniMax-M2.5 FP8 B200" - "Add tp4 ep4 search-space entries (conc 32-256) for all seq-len configs" + - "Remove ISL 1024 / OSL 8192 seq-len config" pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/947 From e405b60d816399032fa44593503e2bdc4110dda6 Mon Sep 17 00:00:00 2001 From: Kedar Potdar Date: Sun, 29 Mar 2026 23:03:12 -0700 Subject: [PATCH 07/16] update image --- .github/configs/nvidia-master.yaml | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/.github/configs/nvidia-master.yaml b/.github/configs/nvidia-master.yaml index 292169457..2ea513a26 100644 --- a/.github/configs/nvidia-master.yaml +++ b/.github/configs/nvidia-master.yaml @@ -3329,7 +3329,7 @@ gptoss-fp4-h200-vllm: - { tp: 8, conc-start: 4, conc-end: 32 } minimaxm2.5-fp8-h200-vllm: - image: vllm/vllm-openai:v0.18.0 + image: vllm/vllm-openai:nightly-5b8c30d62b754b575e043ce2fc0dcbf8a64f6306 model: MiniMaxAI/MiniMax-M2.5 model-prefix: minimaxm2.5 runner: h200 From 534927b2f1a5b38ff6779179097a182e73754535 Mon Sep 17 00:00:00 2001 From: Kedar Potdar Date: Wed, 1 Apr 2026 10:38:20 -0700 Subject: [PATCH 08/16] update config and remove DEEPGEMM flag --- .github/configs/nvidia-master.yaml | 2 +- benchmarks/single_node/minimaxm2.5_fp8_b200.sh | 1 - 2 files changed, 1 insertion(+), 2 deletions(-) diff --git a/.github/configs/nvidia-master.yaml b/.github/configs/nvidia-master.yaml index c81fe00b1..7c57d267c 100644 --- a/.github/configs/nvidia-master.yaml +++ b/.github/configs/nvidia-master.yaml @@ -3101,7 +3101,7 @@ gptoss-fp4-b200-vllm: - { tp: 8, conc-start: 4, conc-end: 4 } minimaxm2.5-fp8-b200-vllm: - image: vllm/vllm-openai:v0.18.0-cu130 + image: vllm/vllm-openai:nightly-5b8c30d62b754b575e043ce2fc0dcbf8a64f6306 model: MiniMaxAI/MiniMax-M2.5 model-prefix: minimaxm2.5 runner: b200 diff --git a/benchmarks/single_node/minimaxm2.5_fp8_b200.sh b/benchmarks/single_node/minimaxm2.5_fp8_b200.sh index 11e9ce65e..67269571c 100755 --- a/benchmarks/single_node/minimaxm2.5_fp8_b200.sh +++ b/benchmarks/single_node/minimaxm2.5_fp8_b200.sh @@ -24,7 +24,6 @@ hf download "$MODEL" SERVER_LOG=/workspace/server.log PORT=${PORT:-8888} -export VLLM_USE_DEEP_GEMM=0 export VLLM_FLASHINFER_ALLREDUCE_BACKEND=mnnvl if [ "$EP_SIZE" -ge 1 ]; then From 6073a5ed16c31ba6a0b773caf57c51e947366fe7 Mon Sep 17 00:00:00 2001 From: Kedar Potdar Date: Wed, 1 Apr 2026 16:11:54 -0700 Subject: [PATCH 09/16] test tep --- .github/configs/nvidia-master.yaml | 15 ++++++++------- 1 file changed, 8 insertions(+), 7 deletions(-) diff --git a/.github/configs/nvidia-master.yaml b/.github/configs/nvidia-master.yaml index 7c57d267c..1ddeb4d31 100644 --- a/.github/configs/nvidia-master.yaml +++ b/.github/configs/nvidia-master.yaml @@ -3113,14 +3113,15 @@ minimaxm2.5-fp8-b200-vllm: osl: 1024 search-space: - { tp: 2, conc-start: 4, conc-end: 256 } + - { tp: 2, ep: 2, conc-start: 4, conc-end: 256 } - { tp: 4, conc-start: 4, conc-end: 256 } - - { tp: 4, ep: 4, conc-start: 64, conc-end: 64 } - - { tp: 8, conc-start: 4, conc-end: 8 } - - isl: 8192 - osl: 1024 - search-space: - - { tp: 2, conc-start: 4, conc-end: 256 } - - { tp: 4, conc-start: 4, conc-end: 256 } + - { tp: 4, ep: 4, conc-start: 16, conc-end: 64 } + # - { tp: 8, conc-start: 4, conc-end: 8 } + # - isl: 8192 + # osl: 1024 + # search-space: + # - { tp: 2, conc-start: 4, conc-end: 256 } + # - { tp: 4, conc-start: 4, conc-end: 256 } gptoss-fp4-h100-vllm: image: vllm/vllm-openai:v0.18.0 From f6d81f78dc94a5359d8f94d58fd6effd1a519353 Mon Sep 17 00:00:00 2001 From: Kedar Potdar Date: Wed, 1 Apr 2026 18:13:21 -0700 Subject: [PATCH 10/16] fix typo in ep bash script --- benchmarks/single_node/minimaxm2.5_fp8_b200.sh | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/benchmarks/single_node/minimaxm2.5_fp8_b200.sh b/benchmarks/single_node/minimaxm2.5_fp8_b200.sh index 67269571c..5604b553c 100755 --- a/benchmarks/single_node/minimaxm2.5_fp8_b200.sh +++ b/benchmarks/single_node/minimaxm2.5_fp8_b200.sh @@ -26,7 +26,7 @@ PORT=${PORT:-8888} export VLLM_FLASHINFER_ALLREDUCE_BACKEND=mnnvl -if [ "$EP_SIZE" -ge 1 ]; then +if [ "$EP_SIZE" -gt 1 ]; then EP=" --enable-expert-parallel" else EP=" " From d172646bda46052a6743fb7b0075292ba617dbc4 Mon Sep 17 00:00:00 2001 From: Kedar Potdar Date: Thu, 2 Apr 2026 19:32:16 -0700 Subject: [PATCH 11/16] add max cudagraph size --- benchmarks/single_node/minimaxm2.5_fp8_b200.sh | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/benchmarks/single_node/minimaxm2.5_fp8_b200.sh b/benchmarks/single_node/minimaxm2.5_fp8_b200.sh index 5604b553c..d892d72f1 100755 --- a/benchmarks/single_node/minimaxm2.5_fp8_b200.sh +++ b/benchmarks/single_node/minimaxm2.5_fp8_b200.sh @@ -43,10 +43,12 @@ set -x vllm serve $MODEL --port $PORT \ --tensor-parallel-size=$TP \ $EP \ ---gpu-memory-utilization 0.95 \ +--gpu-memory-utilization 0.90 \ --max-model-len $MAX_MODEL_LEN \ --block-size=32 \ --kv-cache-dtype fp8 \ +--max-cudagraph-capture-size 2048 \ +--max-num-batched-tokens "$((ISL * 2 ))" \ --stream-interval 20 --no-enable-prefix-caching \ --trust-remote-code > $SERVER_LOG 2>&1 & From 97b25f16d41bc9ff7951845849951f1219561cdf Mon Sep 17 00:00:00 2001 From: Kedar Potdar Date: Thu, 2 Apr 2026 19:34:10 -0700 Subject: [PATCH 12/16] upgrade to vllm 0.19 --- .github/configs/nvidia-master.yaml | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-) diff --git a/.github/configs/nvidia-master.yaml b/.github/configs/nvidia-master.yaml index 1ddeb4d31..fc38bf490 100644 --- a/.github/configs/nvidia-master.yaml +++ b/.github/configs/nvidia-master.yaml @@ -3101,7 +3101,7 @@ gptoss-fp4-b200-vllm: - { tp: 8, conc-start: 4, conc-end: 4 } minimaxm2.5-fp8-b200-vllm: - image: vllm/vllm-openai:nightly-5b8c30d62b754b575e043ce2fc0dcbf8a64f6306 + image: vllm/vllm-openai:0.19.0-aarch64-cu130-ubuntu2404 model: MiniMaxAI/MiniMax-M2.5 model-prefix: minimaxm2.5 runner: b200 @@ -3117,11 +3117,11 @@ minimaxm2.5-fp8-b200-vllm: - { tp: 4, conc-start: 4, conc-end: 256 } - { tp: 4, ep: 4, conc-start: 16, conc-end: 64 } # - { tp: 8, conc-start: 4, conc-end: 8 } - # - isl: 8192 - # osl: 1024 - # search-space: - # - { tp: 2, conc-start: 4, conc-end: 256 } - # - { tp: 4, conc-start: 4, conc-end: 256 } + - isl: 8192 + osl: 1024 + search-space: + - { tp: 2, conc-start: 4, conc-end: 256 } + - { tp: 4, conc-start: 4, conc-end: 256 } gptoss-fp4-h100-vllm: image: vllm/vllm-openai:v0.18.0 From beb6a6b1845cfca445527175db8c11d51c81da00 Mon Sep 17 00:00:00 2001 From: Kedar Potdar Date: Thu, 2 Apr 2026 19:34:46 -0700 Subject: [PATCH 13/16] typo --- .github/configs/nvidia-master.yaml | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/.github/configs/nvidia-master.yaml b/.github/configs/nvidia-master.yaml index fc38bf490..6866c4087 100644 --- a/.github/configs/nvidia-master.yaml +++ b/.github/configs/nvidia-master.yaml @@ -3101,7 +3101,7 @@ gptoss-fp4-b200-vllm: - { tp: 8, conc-start: 4, conc-end: 4 } minimaxm2.5-fp8-b200-vllm: - image: vllm/vllm-openai:0.19.0-aarch64-cu130-ubuntu2404 + image: vllm/vllm-openai:v0.19.0-cu130 model: MiniMaxAI/MiniMax-M2.5 model-prefix: minimaxm2.5 runner: b200 From 9af9992aa5c09160c2e7a560f34911f6a1309261 Mon Sep 17 00:00:00 2001 From: Kedar Potdar Date: Fri, 3 Apr 2026 09:09:14 -0700 Subject: [PATCH 14/16] revert h200 change --- .github/configs/nvidia-master.yaml | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/.github/configs/nvidia-master.yaml b/.github/configs/nvidia-master.yaml index 6866c4087..f514157ef 100644 --- a/.github/configs/nvidia-master.yaml +++ b/.github/configs/nvidia-master.yaml @@ -3112,9 +3112,9 @@ minimaxm2.5-fp8-b200-vllm: - isl: 1024 osl: 1024 search-space: - - { tp: 2, conc-start: 4, conc-end: 256 } + - { tp: 2, conc-start: 4, conc-end: 512 } - { tp: 2, ep: 2, conc-start: 4, conc-end: 256 } - - { tp: 4, conc-start: 4, conc-end: 256 } + - { tp: 4, conc-start: 4, conc-end: 512 } - { tp: 4, ep: 4, conc-start: 16, conc-end: 64 } # - { tp: 8, conc-start: 4, conc-end: 8 } - isl: 8192 @@ -3348,7 +3348,7 @@ gptoss-fp4-h200-vllm: - { tp: 8, conc-start: 4, conc-end: 32 } minimaxm2.5-fp8-h200-vllm: - image: vllm/vllm-openai:nightly-5b8c30d62b754b575e043ce2fc0dcbf8a64f6306 + image: vllm/vllm-openai:v0.18.0 model: MiniMaxAI/MiniMax-M2.5 model-prefix: minimaxm2.5 runner: h200 From 0d734c18efe9c23648fa24742091ca8913f59195 Mon Sep 17 00:00:00 2001 From: "claude[bot]" <41898282+claude[bot]@users.noreply.github.com> Date: Fri, 3 Apr 2026 16:27:22 +0000 Subject: [PATCH 15/16] fix: update perf-changelog version to v0.19.0 Co-authored-by: Cameron Quilici --- perf-changelog.yaml | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/perf-changelog.yaml b/perf-changelog.yaml index d041d897e..c490922a4 100644 --- a/perf-changelog.yaml +++ b/perf-changelog.yaml @@ -1239,7 +1239,7 @@ - config-keys: - minimaxm2.5-fp8-b200-vllm description: - - "Update vLLM image from v0.17.0 to v0.18.0 for MiniMax-M2.5 FP8 B200" + - "Update vLLM image from v0.17.0 to v0.19.0 for MiniMax-M2.5 FP8 B200" - "Add tp4 ep4 search-space entries (conc 32-256) for all seq-len configs" - "Remove ISL 1024 / OSL 8192 seq-len config" pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/947 From 4f150049ca231e3efc2b63c0c7df9b7e2027a2a5 Mon Sep 17 00:00:00 2001 From: "claude[bot]" <41898282+claude[bot]@users.noreply.github.com> Date: Fri, 3 Apr 2026 16:27:38 +0000 Subject: [PATCH 16/16] Remove commented-out tp:8 search-space entry Co-authored-by: Cameron Quilici --- .github/configs/nvidia-master.yaml | 1 - 1 file changed, 1 deletion(-) diff --git a/.github/configs/nvidia-master.yaml b/.github/configs/nvidia-master.yaml index fe32ea6b9..7a8d94326 100644 --- a/.github/configs/nvidia-master.yaml +++ b/.github/configs/nvidia-master.yaml @@ -3116,7 +3116,6 @@ minimaxm2.5-fp8-b200-vllm: - { tp: 2, ep: 2, conc-start: 4, conc-end: 256 } - { tp: 4, conc-start: 4, conc-end: 512 } - { tp: 4, ep: 4, conc-start: 16, conc-end: 64 } - # - { tp: 8, conc-start: 4, conc-end: 8 } - isl: 8192 osl: 1024 search-space: