Local AI Registry

Recipe

minimax-m2-7-autoround-int4-w4a16-intel-arc-pro-b70-32gb-vllm-tp4-yk0zdcjx

minimax-m2-7-autoround-int4-w4a16-intel-arc-pro-b70-32gb-vllm-tp4-yk0zdcjx
JSON APIFind compatible recipes

Observed LocalMaxxing leaderboard run. Evidence for compatibility, not an executable launch contract.

Record

Status
candidate
Source
localmaxxing
Engine
vllm
Engine version
0.20.2rc1.dev2-local
Graph
piecewise
Accelerators
4
Tensor parallel
4
Context tokens
2,048
Max concurrency
1
chat
unknown
reasoning
unknown
tools
unknown
vision
unknown

Hugging Face model card

Identity

https://huggingface.co/MJPansa/MiniMax-M2.7-REAP-172B-A10B-AutoRound-W4A16
Repository
MJPansa/MiniMax-M2.7-REAP-172B-A10B-AutoRound-W4A16
Status
known
Link type
Exact Hub repository

Public Hugging Face repository confirmed by the Hub API.

Candidate: useful compatibility or speed evidence. The registry does not offer Run until promotion requirements are met.

Observed configuration

vllm

Evidence only · candidate · reference

Candidate evidence — not a Run contract

Source
https://www.localmaxxing.com/en/runs/cmpuesbma00r5mq01yk0zdcjx

Observed source tokens

Mechanical split of the source command. Unverified against the engine CLI. The registry does not offer Run for this recipe.

  1. /home/steve/llm-optimizations/experiments/minimax-m27-reap-autoround-vllm/scripts/bench-decode.sh;
  2. vllm
  3. bench
  4. throughput
  5. --backend
  6. vllm
  7. --model
  8. /mnt/fast-ai/llm-models/minimax-m2.7-reap-autoround-w4a16
  9. --tokenizer
  10. /mnt/fast-ai/llm-models/minimax-m2.7-reap-autoround-w4a16
  11. --trust-remote-code
  12. --dtype
  13. float16
  14. --tensor-parallel-size
  15. 4
  16. --distributed-executor-backend
  17. mp
  18. --max-model-len
  19. 2048
  20. --max-num-batched-tokens
  21. 512
  22. --max-num-seqs
  23. 1
  24. --dataset-name
  25. random
  26. --random-input-len
  27. 512
  28. --random-output-len
  29. 1536
  30. --random-range-ratio
  31. 0
  32. --num-prompts
  33. 1
  34. --disable-log-stats
  35. --async-engine
  36. --block-size
  37. 256
  38. --no-enable-prefix-caching
  39. --compilation-config
  40. {use_inductor_graph_partition:true,compile_sizes:[1],cudagraph_mode:PIECEWISE}
FlagValue
--backendvllm
--model/mnt/fast-ai/llm-models/minimax-m2.7-reap-autoround-w4a16
--tokenizer/mnt/fast-ai/llm-models/minimax-m2.7-reap-autoround-w4a16
--dtypefloat16
--tensor-parallel-size4
--distributed-executor-backendmp
--max-model-len2048
--max-num-batched-tokens512
--max-num-seqs1
--dataset-namerandom
--random-input-len512
--random-output-len1536
--random-range-ratio0
--num-prompts1
--block-size256
--compilation-config{use_inductor_graph_partition:true,compile_sizes:[1],cudagraph_mode:PIECEWISE}

Environment

VariableValue
VLLM_BENCH_TEMPERATURE0

Measured speed

ConcurrencyContextPrefillDecodeTTFT msStatusSweep
—2,048—89.5—observedminimax-m2-7-autoround-int4-w4a16-intel-arc-pro-b70-32gb-vllm-tp4-yk0zdcjx-sweep

Remaining fields

Identity, launch, related records, and measured speed are shown above. This is the rest of the normalized record.

hardware count
4
hardware id
intel-arc-pro-b70-32gb
id
minimax-m2-7-autoround-int4-w4a16-intel-arc-pro-b70-32gb-vllm-tp4-yk0zdcjx
model instance id
mjpansa-minimax-m2-7-reap-172b-a10b-autoround-w4a16--autoround-int4-w4a16
recipe source
localmaxxing
schema version
local-ai-registry/v1
speed sweep ids
minimax-m2-7-autoround-int4-w4a16-intel-arc-pro-b70-32gb-vllm-tp4-yk0zdcjx-sweep
status
candidate

capabilities

engine

graph mode
piecewise
name
vllm
version
0.20.2rc1.dev2-local

serving

max concurrency
1
max context tokens
2,048
tensor parallel
4
Provenance & metadata (3)

facts

capabilities.chat · provenance · captured at
2026-08-30T09:10:02Z
capabilities.chat · reason capability-not-verified
capabilities.chat · state unknown
capabilities.reasoning · provenance · captured at
2026-08-30T09:10:02Z
capabilities.reasoning · reason capability-not-verified
capabilities.reasoning · state unknown
capabilities.tools · provenance · captured at
2026-08-30T09:10:02Z
capabilities.tools · reason capability-not-verified
capabilities.tools · state unknown
capabilities.vision · provenance · captured at
2026-08-30T09:10:02Z
capabilities.vision · reason capability-not-verified
capabilities.vision · state unknown
engine.graph mode · provenance · captured at
2026-09-01T01:32:29Z
engine.graph mode · reason explicit-cudagraph-mode-in-observed-command
engine.graph mode · state known
serving.kv cache tokens · provenance · captured at
2026-08-30T09:10:02Z
serving.kv cache tokens · reason kv-cache-capacity-not-published
serving.kv cache tokens · state unknown
serving.max concurrency · provenance · captured at
2026-08-31T23:03:15Z
serving.max concurrency · reason server-capacity-derived-from-source-evidence
serving.max concurrency · state known

metadata

localmaxxing · batch size
1
localmaxxing · hardware label
Intel Arc Pro B70
localmaxxing · notes
MiniMax M2.7 REAP AutoRound W4A16 run on 4x Intel Arc Pro B70 via vLLM/XPU TP4. This supersedes the corrected greedy REAP baseline cmpuc7tkq00qamq01z61pnb3c. New easy win: force oneCCL Level Zero IPC to pidfd with CCL_IPC=pidfd and CCL_ZE_IPC_EXCHANGE=pidfd while keeping CCL_TOPO_FABRIC_VERTEX_CONNECTION_CHECK=0. Quality gate with explicit pidfd passed: /mnt/fast-ai/bench-results/minimax-m27-reap-autoround-vllm/quality/quality-smoke-20260531T231727Z.json reports passed=true, deterministic output, 1473 generated tokens, and no NUL/control/degenerate output. Same-window p512/n1536 screen: pidfd 89.427389, pidfd repeat 89.326898, default IPC control 89.275803, sockets 88.904930, promoted config 89.499223 output tok/s. No speculative decoding, no expert dropping, no router approximation, no power-limit change.
localmaxxing · observed command
VLLM_BENCH_TEMPERATURE=0 /home/steve/llm-optimizations/experiments/minimax-m27-reap-autoround-vllm/scripts/bench-decode.sh; vllm bench throughput --backend vllm --model /mnt/fast-ai/llm-models/minimax-m2.7-reap-autoround-w4a16 --tokenizer /mnt/fast-ai/llm-models/minimax-m2.7-reap-autoround-w4a16 --trust-remote-code --dtype float16 --tensor-parallel-size 4 --distributed-executor-backend mp --max-model-len 2048 --max-num-batched-tokens 512 --max-num-seqs 1 --dataset-name random --random-input-len 512 --random-output-len 1536 --random-range-ratio 0 --num-prompts 1 --disable-log-stats --async-engine --block-size 256 --no-enable-prefix-caching --compilation-config {"use_inductor_graph_partition":true,"compile_sizes":[1],"cudagraph_mode":"PIECEWISE"}
localmaxxing · run id
cmpuesbma00r5mq01yk0zdcjx
localmaxxing · tokenized · environment · VLLM BENCH TEMPERATURE
0
localmaxxing · tokenized · fidelity
faithful

provenance

captured at
2026-08-30T09:10:02Z

sources

captured atkindurl
2026-08-30T09:10:02Znormalized-recipewww.localmaxxing.com/en/runs/cmpuesbma00r5mq01yk0zdcjx ↗