Local AI Registry

Recipe

minimax-m2-7-ud-iq4-xs-dgx-spark-gb10-128gb-llama-cpp-tp1-wtsw8ogj

minimax-m2-7-ud-iq4-xs-dgx-spark-gb10-128gb-llama-cpp-tp1-wtsw8ogj
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
b8833 (45cac7ca7)
Accelerators
1
Tensor parallel
1
Context tokens
108,000
Max concurrency
1
chat
unknown
reasoning
unknown
tools
unknown
vision
unknown

Hugging Face model card

Identity

https://huggingface.co/MiniMaxAI/MiniMax-M2.7
Repository
MiniMaxAI/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/cmon7v1f70007l204wtsw8ogj

Observed source tokens

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

  1. /opt/llama.cpp/build-cuda/bin/llama-server
  2. --jinja
  3. -fa
  4. on
  5. --no-warmup
  6. -t
  7. 20
  8. -c
  9. 108000
  10. -ctk
  11. q8_0
  12. -ctv
  13. q8_0
  14. -np
  15. 1
  16. -ngl
  17. 99
  18. -a
  19. MiniMax-M2.7-UD-IQ4_XS
  20. -m
  21. /models/MiniMax-M2.7-GGUF/UD-IQ4_XS/MiniMax-M2.7-UD-IQ4_XS-00001-of-00004.gguf
  22. --host
  23. 0.0.0.0
  24. --port
  25. 18080
  26. --cache-reuse
  27. 256
  28. --ctx-checkpoints
  29. 0
  30. -cram
  31. 100
  32. --spec-type
  33. ngram-simple
  34. --draft-max
  35. 16
  36. --draft-min
  37. 1
  38. --draft-p-min
  39. 0.5
  40. --spec-ngram-size-n
  41. 4
FlagValue
-faon
-t20
-c108000
-ctkq8_0
-ctvq8_0
-np1
-ngl99
-aMiniMax-M2.7-UD-IQ4_XS
-m/models/MiniMax-M2.7-GGUF/UD-IQ4_XS/MiniMax-M2.7-UD-IQ4_XS-00001-of-00004.gguf
--host0.0.0.0
--port18080
--cache-reuse256
--ctx-checkpoints0
-cram100
--spec-typengram-simple
--draft-max16
--draft-min1
--draft-p-min0.5
--spec-ngram-size-n4

Measured speed

ConcurrencyContextPrefillDecodeTTFT msStatusSweep
1108,000—30.9807observedminimax-m2-7-ud-iq4-xs-dgx-spark-gb10-128gb-llama-cpp-tp1-wtsw8ogj-sweep

Remaining fields

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

hardware count
1
hardware id
dgx-spark-gb10-128gb
id
minimax-m2-7-ud-iq4-xs-dgx-spark-gb10-128gb-llama-cpp-tp1-wtsw8ogj
model instance id
minimaxai-minimax-m2-7--ud-iq4-xs
recipe source
localmaxxing
schema version
local-ai-registry/v1
speed sweep ids
minimax-m2-7-ud-iq4-xs-dgx-spark-gb10-128gb-llama-cpp-tp1-wtsw8ogj-sweep
status
candidate

capabilities

engine

name
llama.cpp
version
b8833 (45cac7ca7)

serving

max concurrency
1
max context tokens
108,000
tensor parallel
1
Provenance & metadata (3)

facts

capabilities.chat · provenance · captured at
2026-08-30T09:26:09Z
capabilities.chat · reason capability-not-verified
capabilities.chat · state unknown
capabilities.reasoning · provenance · captured at
2026-08-30T09:26:09Z
capabilities.reasoning · reason capability-not-verified
capabilities.reasoning · state unknown
capabilities.tools · provenance · captured at
2026-08-30T09:26:09Z
capabilities.tools · reason capability-not-verified
capabilities.tools · state unknown
capabilities.vision · provenance · captured at
2026-08-30T09:26:09Z
capabilities.vision · reason capability-not-verified
capabilities.vision · state unknown
engine.graph mode · provenance · captured at
2026-08-30T09:26:09Z
engine.graph mode · reason runtime-detail-not-published
engine.graph mode · state unknown
serving.kv cache tokens · provenance · captured at
2026-08-30T09:26:09Z
serving.kv cache tokens · reason kv-cache-capacity-not-published
serving.kv cache tokens · state unknown

metadata

localmaxxing · backend
cuda
localmaxxing · hardware label
GB10 Grace Blackwell
localmaxxing · notes
Companion submission to cmomfmks00003k204617egw2l (unsloth/MiniMax-M2.7 GGUF). Same measurement, posted against root model MiniMaxAI/MiniMax-M2.7 with quantization=UD-IQ4_XS explicit so it surfaces on the canonical model page. MiniMax-M2.7-UD-IQ4_XS (101 GB GGUF) on a single DGX Spark with llama.cpp ngram-simple speculative decode at 108K context. Sherlock prose, pp=128, tg=128, c=1, depth=0. tokSOut = 30.88 t/s MEDIAN over n=38 warm trials across 2 independent rounds (mean 30.88, std 4.11, range 24.57-41.85). ttftMs = 807 median. Same headline tg/s at 3.3× context window vs my prior 32K submission, with provably stable long-context behavior (5-phase battery + 10× back-to-back 100K torture, R10/R1=1.006). Image base PINNED to nvcr.io/nvidia/cuda:13.0.0-devel-ubuntu24.04 — CUDA 13.2/13.3 bases produced silent libcublas matmul corruption on GB10 sm_121. Pre-flight: `readlink libcublas.so.13` must show 13.0.0.19. Stability flags overriding defaults: --cache-reuse 256, --ctx-checkpoints 0, -cram 100 (NOT 4096; 4096 collapses decode 5.6→1.0 t/s by run 4). Spec acceptance ~36-82% across runs. Recipe + raw JSONs + Dockerfile: github.com/my-other-github-account/spark-bench-reproducers/tree/main/minimax-m27-longctx-108k-spark By @banana_baeee.
localmaxxing · run id
cmon7v1f70007l204wtsw8ogj
localmaxxing · tokenized · fidelity
faithful

provenance

captured at
2026-08-30T09:26:09Z

sources

captured atkindurl
2026-08-30T09:26:09Znormalized-recipewww.localmaxxing.com/en/runs/cmon7v1f70007l204wtsw8ogj ↗