Local AI Registry

Recipe

gemma-4-31b-it-qat-q4-0-unquantized-q4-k-xl-rtx-5090-32gb-llama-cpp-tp1

gemma-4-31b-it-qat-q4-0-unquantized-q4-k-xl-rtx-5090-32gb-llama-cpp-tp1
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
_beellama-preview-v0.3.2-bin-win-cuda-13.1-x64
Accelerators
1
Tensor parallel
1
Context tokens
200,000
Max concurrency
1
chat
unknown
reasoning
unknown
tools
unknown
vision
unknown

Hugging Face model card

Identity

https://huggingface.co/unsloth/gemma-4-31B-it-qat-GGUF
Repository
unsloth/gemma-4-31B-it-qat-GGUF
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/cmqyxx9c901nyoe01tdp4i4hx

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-server.exe
  2. -m
  3. "E:/LLM/gemma-4-31B-it-qat-UD-Q4_K_XL.gguf"
  4. --model-draft
  5. "E:/LLM/draft/mtp-gemma-4-31B-it.gguf"
  6. --mmproj
  7. "E:/LLM/mmproj-F16_gat.gguf"
  8. --spec-type
  9. draft-mtp
  10. --spec-draft-n-max
  11. 8
  12. --spec-draft-n-min
  13. 3
  14. --port
  15. 8080
  16. -ngl
  17. all
  18. --spec-draft-ngl
  19. all
  20. -b
  21. 2048
  22. -ub
  23. 256
  24. --ctx-size
  25. 200000
  26. --cache-type-k
  27. q5_0
  28. --cache-type-v
  29. q5_0
  30. --flash-attn
  31. on
  32. --cache-ram
  33. 0
  34. --jinja
  35. --no-mmap
  36. --mlock
  37. --no-host
  38. --metrics
  39. --log-timestamps
  40. --log-prefix
  41. --log-colors
  42. off
  43. --reasoning
  44. off
FlagValue
-m"E:/LLM/gemma-4-31B-it-qat-UD-Q4_K_XL.gguf"
--model-draft"E:/LLM/draft/mtp-gemma-4-31B-it.gguf"
--mmproj"E:/LLM/mmproj-F16_gat.gguf"
--spec-typedraft-mtp
--spec-draft-n-max8
--spec-draft-n-min3
--port8080
-nglall
--spec-draft-nglall
-b2048
-ub256
--ctx-size200000
--cache-type-kq5_0
--cache-type-vq5_0
--flash-attnon
--cache-ram0
--log-colorsoff
--reasoningoff

Measured speed

ConcurrencyContextPrefillDecodeTTFT msStatusSweep
1200,000712.6207.678observedgemma-4-31b-it-qat-q4-0-unquantized-q4-k-xl-rtx-5090-32gb-llama-cpp-tp1-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
rtx-5090-32gb
id
gemma-4-31b-it-qat-q4-0-unquantized-q4-k-xl-rtx-5090-32gb-llama-cpp-tp1
model instance id
unsloth-gemma-4-31b-it-qat-gguf--q4-k-xl
recipe source
localmaxxing
schema version
local-ai-registry/v1
speed sweep ids
gemma-4-31b-it-qat-q4-0-unquantized-q4-k-xl-rtx-5090-32gb-llama-cpp-tp1-sweep
status
candidate

capabilities

draft launch

accelerator backend
nvidia
arguments
-hf, unsloth/gemma-4-31B-it-qat-GGUF:Q4_K_XL, --n-gpu-layers, 999, --host, 0.0.0.0, --port, 8080, -c, 200000
container port
8,080
environment · LLAMA CACHE
/root/.cache/huggingface
host port
8,080
image
ghcr.io/ggml-org/llama.cpp:server-cuda12-b10481@sha256:b2497f8834f5ecb4e38530f6bf2734b8e0be107f0f0857e259672d1cb85b71c2
ipc
host
kind
docker
shm size
16g
synthesized · generated at
2026-08-31T22:12:17Z
synthesized · image provenance
gemma-4-12b-q4-k-m-rtx-3060-12gb-llama-cpp-tp1
synthesized · template
llama-cpp-server-v1

mounts

read onlytarget
No/root/.cache/huggingface

engine

name
llama.cpp
version
_beellama-preview-v0.3.2-bin-win-cuda-13.1-x64

serving

max concurrency
1
max context tokens
200,000
tensor parallel
1
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 · backend
cuda
localmaxxing · hardware label
RTX 5090
localmaxxing · notes
Did couple runs and hits 190 to 205t/s as total t/s. Windows eats 2Gb Vram on default. prompt was "Write a Python function that calculates the Fibonacci sequence up to the nth number. Include type hints and docstrings. Then explain the time complexity." with output cut when 2048 tokens hit.
localmaxxing · observed command
llama-server.exe -m "E:/LLM/gemma-4-31B-it-qat-UD-Q4_K_XL.gguf" --model-draft "E:/LLM/draft/mtp-gemma-4-31B-it.gguf" --mmproj "E:/LLM/mmproj-F16_gat.gguf" --spec-type draft-mtp --spec-draft-n-max 8 --spec-draft-n-min 3 --port 8080 -ngl all --spec-draft-ngl all -b 2048 -ub 256 --ctx-size 200000 --cache-type-k q5_0 --cache-type-v q5_0 --flash-attn on --cache-ram 0 --jinja --no-mmap --mlock --no-host --metrics --log-timestamps --log-prefix --log-colors off --reasoning off
localmaxxing · run id
cmqyxx9c901nyoe01tdp4i4hx
localmaxxing · tokenized · arguments
llama-server.exe, -m, "E:/LLM/gemma-4-31B-it-qat-UD-Q4_K_XL.gguf", --model-draft, "E:/LLM/draft/mtp-gemma-4-31B-it.gguf", --mmproj, "E:/LLM/mmproj-F16_gat.gguf", --spec-type, draft-mtp, --spec-draft-n-max, 8, --spec-draft-n-min, 3, --port, 8080, -ngl, all, --spec-draft-ngl, all, -b, 2048, -ub, 256, --ctx-size, 200000, --cache-type-k, q5_0, --cache-type-v, q5_0, --flash-attn, on, --cache-ram, 0, --jinja, --no-mmap, --mlock, --no-host, --metrics, --log-timestamps, --log-prefix, --log-colors, off, --reasoning, off
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/cmqyxx9c901nyoe01tdp4i4hx ↗