Local AI Registry

Recipe

minimax-m2-7-ud-iq4-xs-intel-arc-pro-b70-32gb-llama-cpp-tp4-x14ylh33

minimax-m2-7-ud-iq4-xs-intel-arc-pro-b70-32gb-llama-cpp-tp4-x14ylh33
JSON APIFind compatible recipes

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

Record

Status
candidate
Source
localmaxxing
Engine
llama.cpp
Engine version
ik_llama.cpp 9a26522 + B70 SYCL patches
Accelerators
4
Tensor parallel
4
Context tokens
64
Max concurrency
1
chat
unknown
reasoning
unknown
tools
unknown
vision
unknown

Hugging Face model card

Identity

https://huggingface.co/unsloth/MiniMax-M2.7
Repository
unsloth/MiniMax-M2.7
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

llama.cpp

Evidence only · candidate · reference

Candidate evidence — not a Run contract

Source
https://www.localmaxxing.com/en/runs/cmow2dknq00k2p101x14ylh33

Observed source tokens

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

  1. llama-bench
  2. -m
  3. MiniMax-M2.7-UD-IQ4_XS-00001-of-00004.gguf
  4. -rpc
  5. 127.0.0.1:50100|0,127.0.0.1:50101|0,127.0.0.1:50102|0,127.0.0.1:50103|0
  6. -p
  7. 0
  8. -n
  9. 64
  10. -r
  11. 1
  12. -ngl
  13. 99
  14. -sm
  15. layer
  16. -ts
  17. 0/1/1/1/1
  18. -fa
  19. 0
  20. -nkvo
  21. 1
  22. -ub
  23. 32
  24. -ctk
  25. f16
  26. -ctv
  27. f16
  28. -threads
  29. 4
  30. -w
  31. 0
  32. -mmp
  33. 1
  34. -rtr
  35. 1
  36. -fmoe
  37. 1
  38. -no-mmad
  39. 0
  40. -muge
  41. 1
  42. -sas
  43. 1
  44. -o
  45. json
FlagValue
-mMiniMax-M2.7-UD-IQ4_XS-00001-of-00004.gguf
-rpc127.0.0.1:50100|0,127.0.0.1:50101|0,127.0.0.1:50102|0,127.0.0.1:50103|0
-p0
-n64
-r1
-ngl99
-smlayer
-ts0/1/1/1/1
-fa0
-nkvo1
-ub32
-ctkf16
-ctvf16
-threads4
-w0
-mmp1
-rtr1
-fmoe1
-no-mmad0
-muge1
-sas1
-ojson

Environment

VariableValue
GGML_DISABLE_FUSED_MUL_UNARY1
GGML_DISABLE_FUSED_RMS_NORM1

Measured speed

ConcurrencyContextPrefillDecodeTTFT msStatusSweep
164—14.4—observedminimax-m2-7-ud-iq4-xs-intel-arc-pro-b70-32gb-llama-cpp-tp4-x14ylh33-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-ud-iq4-xs-intel-arc-pro-b70-32gb-llama-cpp-tp4-x14ylh33
model instance id
unsloth-minimax-m2-7--ud-iq4-xs
recipe source
localmaxxing
schema version
local-ai-registry/v1
speed sweep ids
minimax-m2-7-ud-iq4-xs-intel-arc-pro-b70-32gb-llama-cpp-tp4-x14ylh33-sweep
status
candidate

capabilities

engine

name
llama.cpp
version
ik_llama.cpp 9a26522 + B70 SYCL patches

serving

max concurrency
1
max context tokens
64
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
2026-08-30T09:10:02Z
engine.graph mode · reason runtime-detail-not-published
engine.graph mode · state unknown
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

metadata

localmaxxing · hardware label
Intel Arc Pro B70
localmaxxing · notes
MiniMax M2.7 230B.A10B UD-IQ4_XS GGUF on 4x Intel Arc Pro B70 using ik_llama.cpp process-per-GPU RPC+SYCL workers. Same quality-preserving path as prior 14.146 tok/s result, now with scheduler async enabled (-sas 1). No speculative decoding, no smart expert reduction/expert dropping, no power-limit changes. OpenCL direct graph allocation was investigated separately but is not used here.
localmaxxing · observed command
GGML_DISABLE_FUSED_RMS_NORM=1 GGML_DISABLE_FUSED_MUL_UNARY=1 llama-bench -m MiniMax-M2.7-UD-IQ4_XS-00001-of-00004.gguf -rpc 127.0.0.1:50100|0,127.0.0.1:50101|0,127.0.0.1:50102|0,127.0.0.1:50103|0 -p 0 -n 64 -r 1 -ngl 99 -sm layer -ts 0/1/1/1/1 -fa 0 -nkvo 1 -ub 32 -ctk f16 -ctv f16 -threads 4 -w 0 -mmp 1 -rtr 1 -fmoe 1 -no-mmad 0 -muge 1 -sas 1 -o json
localmaxxing · run id
cmow2dknq00k2p101x14ylh33
localmaxxing · tokenized · arguments
llama-bench, -m, MiniMax-M2.7-UD-IQ4_XS-00001-of-00004.gguf, -rpc, 127.0.0.1:50100|0,127.0.0.1:50101|0,127.0.0.1:50102|0,127.0.0.1:50103|0, -p, 0, -n, 64, -r, 1, -ngl, 99, -sm, layer, -ts, 0/1/1/1/1, -fa, 0, -nkvo, 1, -ub, 32, -ctk, f16, -ctv, f16, -threads, 4, -w, 0, -mmp, 1, -rtr, 1, -fmoe, 1, -no-mmad, 0, -muge, 1, -sas, 1, -o, json
localmaxxing · tokenized · environment · GGML DISABLE FUSED MUL UNARY
1
localmaxxing · tokenized · environment · GGML DISABLE FUSED RMS NORM
1
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/cmow2dknq00k2p101x14ylh33 ↗