Skip to content

Update MiniMax-M3 NVFP4 B300 8k1k disaggregated Dynamo-vLLM sweep / 更新 MiniMax-M3 NVFP4 B300 8k1k 分离式 Dynamo-vLLM 扫描配置 #2184

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
jasonlizhengjian wants to merge 7 commits into
base: main
Choose a base branch
from
Open

Update MiniMax-M3 NVFP4 B300 8k1k disaggregated Dynamo-vLLM sweep / 更新 MiniMax-M3 NVFP4 B300 8k1k 分离式 Dynamo-vLLM 扫描配置 #2184

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
4786c59
feat: update MiniMax M3 B300 8k1k disagg recipes
Jul 14, 2026
f628a2e
chore: trigger MiniMax M3 8k1k sweep
Jul 14, 2026
b3b15b0
Merge remote-tracking branch 'origin/main' into agent/update-minimax-…
Jul 14, 2026
8840ef6
fix: skip login-host preflight for MiniMax TP1 sweep
Jul 14, 2026
0c65a6c
Merge main and bump MiniMax Dynamo
Jul 14, 2026
5b0d83b
perf: add MiniMax TP1 TEP4 throughput point
Jul 14, 2026
30be934
perf: preserve MiniMax max-throughput point
Jul 14, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
38 changes: 13 additions & 25 deletions ...m3/b300-fp4/8k1k/1p2d-dep2-tep4-8k1k.yaml → ...x-m3/b300-fp4/8k1k/1p1d-tp1-tp4-8k1k.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,47 +1,37 @@
name: "minimax-m3-vllm-disagg-b300-1p2d-fp4-dep2-tep4-8k1k"

name: "minimax-m3-vllm-disagg-b300-1p1d-tp1-tp4-fp4-8k1k"
model:
path: "nvidia/MiniMax-M3-NVFP4"
container: "vllm/vllm-openai:vllm-minimax-m3-perf-x86_64-13.0.1-8b00f41"
container: "vllm/vllm-openai:nightly-2dfaae752b4db0d43cfc0715c780e33be030d0f1"
precision: "fp4"

resources:
gpu_type: "b300"
gpus_per_node: 8
prefill_nodes: 1
decode_nodes: 1
decode_nodes: 0
prefill_workers: 1
decode_workers: 2
gpus_per_prefill: 2
decode_workers: 1
gpus_per_prefill: 1
gpus_per_decode: 4

dynamo:
install: true
version: 1.3.0.dev20260614

version: 1.3.0.dev20260713
frontend:
type: dynamo
enable_multiple_frontends: false

backend:
type: vllm
connector: null

connector:
prefill_environment:
VLLM_FLOAT32_MATMUL_PRECISION: high
UCX_TLS: "cuda_copy,rc"

decode_environment:
VLLM_FLOAT32_MATMUL_PRECISION: high
UCX_TLS: "cuda_copy,rc"

vllm_config:
prefill:
no-enable-flashinfer-autotune: true
tensor-parallel-size: 1
data-parallel-size: 2
data-parallel-rpc-port: 13345
enable-expert-parallel: true
enable-expert-parallel: false
trust-remote-code: true
no-enable-prefix-caching: true
kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
Expand All @@ -52,13 +42,12 @@ backend:
max-model-len: 9472
language-model-only: true
stream-interval: 32
max-cudagraph-capture-size: 2048
max-num-batched-tokens: 16384

enforce-eager: true
max-num-batched-tokens: 8192
decode:
no-enable-flashinfer-autotune: true
tensor-parallel-size: 4
enable-expert-parallel: true
enable-expert-parallel: false
trust-remote-code: true
no-enable-prefix-caching: true
kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
Expand All @@ -72,14 +61,13 @@ backend:
max-num-seqs: 512
max-num-batched-tokens: 16384
max-cudagraph-capture-size: 4096

allow_prefill_decode_colocation: true
health_check:
max_attempts: 360
interval_seconds: 10

benchmark:
type: "sa-bench"
isl: 8192
osl: 1024
concurrencies: "16"
concurrencies: "8x16x24x32x48x64"
req_rate: "inf"
42 changes: 15 additions & 27 deletions ...m3/b300-fp4/8k1k/1p4d-dep2-tep8-8k1k.yaml → ...x-m3/b300-fp4/8k1k/1p1d-tp1-tp8-8k1k.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,47 +1,38 @@
name: "minimax-m3-vllm-disagg-b300-1p4d-fp4-dep2-tep8-8k1k"

name: "minimax-m3-vllm-disagg-b300-1p1d-tp1-tp8-fp4-8k1k"
model:
path: "nvidia/MiniMax-M3-NVFP4"
container: "vllm/vllm-openai:vllm-minimax-m3-perf-x86_64-13.0.1-8b00f41"
container: "vllm/vllm-openai:nightly-2dfaae752b4db0d43cfc0715c780e33be030d0f1"
precision: "fp4"

resources:
gpu_type: "b300"
gpus_per_node: 8
prefill_nodes: 1
decode_nodes: 4
decode_nodes: 1
prefill_workers: 1
decode_workers: 4
gpus_per_prefill: 2
decode_workers: 1
gpus_per_prefill: 1
gpus_per_decode: 8

dynamo:
install: true
version: 1.3.0.dev20260614

version: 1.3.0.dev20260713
frontend:
type: dynamo
enable_multiple_frontends: false

backend:
type: vllm
connector: null

connector:
prefill_environment:
VLLM_FLOAT32_MATMUL_PRECISION: high
UCX_TLS: "cuda_copy,rc"

decode_environment:
VLLM_FLOAT32_MATMUL_PRECISION: high
UCX_TLS: "cuda_copy,rc"

VLLM_FLASHINFER_ALLREDUCE_BACKEND: trtllm
vllm_config:
prefill:
no-enable-flashinfer-autotune: true
tensor-parallel-size: 1
data-parallel-size: 2
data-parallel-rpc-port: 13345
enable-expert-parallel: true
enable-expert-parallel: false
trust-remote-code: true
no-enable-prefix-caching: true
kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
Expand All @@ -52,13 +43,12 @@ backend:
max-model-len: 9472
language-model-only: true
stream-interval: 32
max-cudagraph-capture-size: 2048
max-num-batched-tokens: 16384

enforce-eager: true
max-num-batched-tokens: 8192
decode:
no-enable-flashinfer-autotune: true
tensor-parallel-size: 8
enable-expert-parallel: true
enable-expert-parallel: false
trust-remote-code: true
no-enable-prefix-caching: true
kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
Expand All @@ -69,17 +59,15 @@ backend:
max-model-len: 9472
language-model-only: true
stream-interval: 32
max-num-seqs: 1024
max-num-seqs: 256
max-num-batched-tokens: 16384
max-cudagraph-capture-size: 4096

max-cudagraph-capture-size: 256
health_check:
max_attempts: 360
interval_seconds: 10

benchmark:
type: "sa-bench"
isl: 8192
osl: 1024
concurrencies: "4"
concurrencies: "1x2x4x8x16"
req_rate: "inf"
40 changes: 14 additions & 26 deletions ...m3/b300-fp4/8k1k/1p4d-dep2-tep4-8k1k.yaml → ...x-m3/b300-fp4/8k1k/2p1d-tp1-tp4-8k1k.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,47 +1,37 @@
name: "minimax-m3-vllm-disagg-b300-1p4d-fp4-dep2-tep4-8k1k"

name: "minimax-m3-vllm-disagg-b300-2p1d-tp1-tp4-fp4-8k1k"
model:
path: "nvidia/MiniMax-M3-NVFP4"
container: "vllm/vllm-openai:vllm-minimax-m3-perf-x86_64-13.0.1-8b00f41"
container: "vllm/vllm-openai:nightly-2dfaae752b4db0d43cfc0715c780e33be030d0f1"
precision: "fp4"

resources:
gpu_type: "b300"
gpus_per_node: 8
prefill_nodes: 1
decode_nodes: 2
prefill_workers: 1
decode_workers: 4
gpus_per_prefill: 2
decode_nodes: 0
prefill_workers: 2
decode_workers: 1
gpus_per_prefill: 1
gpus_per_decode: 4

dynamo:
install: true
version: 1.3.0.dev20260614

version: 1.3.0.dev20260713
frontend:
type: dynamo
enable_multiple_frontends: false

backend:
type: vllm
connector: null

connector:
prefill_environment:
VLLM_FLOAT32_MATMUL_PRECISION: high
UCX_TLS: "cuda_copy,rc"

decode_environment:
VLLM_FLOAT32_MATMUL_PRECISION: high
UCX_TLS: "cuda_copy,rc"

vllm_config:
prefill:
no-enable-flashinfer-autotune: true
tensor-parallel-size: 1
data-parallel-size: 2
data-parallel-rpc-port: 13345
enable-expert-parallel: true
enable-expert-parallel: false
trust-remote-code: true
no-enable-prefix-caching: true
kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
Expand All @@ -52,13 +42,12 @@ backend:
max-model-len: 9472
language-model-only: true
stream-interval: 32
max-cudagraph-capture-size: 2048
max-num-batched-tokens: 16384

enforce-eager: true
max-num-batched-tokens: 8192
decode:
no-enable-flashinfer-autotune: true
tensor-parallel-size: 4
enable-expert-parallel: true
enable-expert-parallel: false
trust-remote-code: true
no-enable-prefix-caching: true
kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
Expand All @@ -72,14 +61,13 @@ backend:
max-num-seqs: 512
max-num-batched-tokens: 16384
max-cudagraph-capture-size: 4096

allow_prefill_decode_colocation: true
health_check:
max_attempts: 360
interval_seconds: 10

benchmark:
type: "sa-bench"
isl: 8192
osl: 1024
concurrencies: "16x32x64x128"
concurrencies: "128"
req_rate: "inf"
43 changes: 15 additions & 28 deletions ...-m3/b300-fp4/8k1k/1p1d-dep2-tp4-8k1k.yaml → ...x-m3/b300-fp4/8k1k/3p1d-tp1-tp4-8k1k.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,48 +1,38 @@
name: "minimax-m3-vllm-disagg-b300-1p1d-fp4-dep2-tp4-8k1k"

name: "minimax-m3-vllm-disagg-b300-3p1d-tp1-tp4-fp4-8k1k"
model:
path: "nvidia/MiniMax-M3-NVFP4"
container: "vllm/vllm-openai:vllm-minimax-m3-perf-x86_64-13.0.1-8b00f41"
container: "vllm/vllm-openai:nightly-2dfaae752b4db0d43cfc0715c780e33be030d0f1"
precision: "fp4"

resources:
gpu_type: "b300"
gpus_per_node: 8
prefill_nodes: 1
decode_nodes: 1
prefill_workers: 1
decode_nodes: 0
prefill_workers: 3
decode_workers: 1
gpus_per_prefill: 2
gpus_per_prefill: 1
gpus_per_decode: 4

dynamo:
install: true
version: 1.3.0.dev20260614

version: 1.3.0.dev20260713
frontend:
type: dynamo
enable_multiple_frontends: false

backend:
type: vllm
connector: null
connector:
allow_prefill_decode_colocation: true

prefill_environment:
VLLM_FLOAT32_MATMUL_PRECISION: high
UCX_TLS: "cuda_ipc,cuda_copy,rc"

UCX_TLS: "cuda_copy,rc"
decode_environment:
VLLM_FLOAT32_MATMUL_PRECISION: high
UCX_TLS: "cuda_ipc,cuda_copy,rc"

UCX_TLS: "cuda_copy,rc"
vllm_config:
prefill:
no-enable-flashinfer-autotune: true
tensor-parallel-size: 1
data-parallel-size: 2
data-parallel-rpc-port: 13345
enable-expert-parallel: true
enable-expert-parallel: false
trust-remote-code: true
no-enable-prefix-caching: true
kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
Expand All @@ -53,9 +43,8 @@ backend:
max-model-len: 9472
language-model-only: true
stream-interval: 32
max-cudagraph-capture-size: 2048
max-num-batched-tokens: 16384

enforce-eager: true
max-num-batched-tokens: 8192
decode:
no-enable-flashinfer-autotune: true
tensor-parallel-size: 4
Expand All @@ -70,17 +59,15 @@ backend:
max-model-len: 9472
language-model-only: true
stream-interval: 32
max-num-seqs: 1024
max-num-seqs: 512
max-num-batched-tokens: 16384
max-cudagraph-capture-size: 2048

max-cudagraph-capture-size: 4096
health_check:
max_attempts: 360
interval_seconds: 10

benchmark:
type: "sa-bench"
isl: 8192
osl: 1024
concurrencies: "1x4x8x16"
concurrencies: "256"
req_rate: "inf"
Loading
Loading