Recipe

muse-glimmer-30b-dynamic-q4-k-xl-dflash2-q4-k-m-radeon-ai-pro-r9700-32gb-llama-cpp-tp1

muse-glimmer-30b-dynamic-q4-k-xl-dflash2-q4-k-m-radeon-ai-pro-r9700-32gb-llama-cpp-tp1

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

Record

Status
candidate
Source
localmaxxing
Engine
llama.cpp
Engine version
0.3.0-dev / build 10680 / commit d7bd3bfca
Accelerators
1
Tensor parallel
1
Context tokens
32,768
Max concurrency
1
chat
unknown
reasoning
unknown
tools
unknown
vision
unknown

Hugging Face model card

Identity

https://huggingface.co/meta-models/Muse-Glimmer-30B-GGUF
Repository
meta-models/Muse-Glimmer-30B-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/cmtdonksq005olm01t0itg4cs

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
  2. -m
  3. Muse-Glimmer-30B-KQuant-Dynamic-Q4_K_XL.gguf
  4. --mmproj
  5. mmproj-Muse-Glimmer-30B-Q4_K_M.gguf
  6. -md
  7. Muse-Glimmer-30B-DFlash2-Q4_K_M.gguf
  8. --spec-type
  9. draft-dflash
  10. --spec-draft-n-max
  11. 15
  12. -ngld
  13. 99
  14. -ngl
  15. 99
  16. -c
  17. 32768
  18. -np
  19. 1
  20. -ctk
  21. q8_0
  22. -ctv
  23. q8_0
  24. -fa
  25. on
  26. --jinja
  27. --chat-template-kwargs
  28. {reasoning_strength: low}
FlagValue
-mMuse-Glimmer-30B-KQuant-Dynamic-Q4_K_XL.gguf
--mmprojmmproj-Muse-Glimmer-30B-Q4_K_M.gguf
-mdMuse-Glimmer-30B-DFlash2-Q4_K_M.gguf
--spec-typedraft-dflash
--spec-draft-n-max15
-ngld99
-ngl99
-c32768
-np1
-ctkq8_0
-ctvq8_0
-faon
--chat-template-kwargs{reasoning_strength: low}

Measured speed

ConcurrencyContextPrefillDecodeTTFT msStatusSweep
132,768380.6134.6290.7observedmuse-glimmer-30b-dynamic-q4-k-xl-dflash2-q4-k-m-radeon-ai-pro-r9700-32gb-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
radeon-ai-pro-r9700-32gb
id
muse-glimmer-30b-dynamic-q4-k-xl-dflash2-q4-k-m-radeon-ai-pro-r9700-32gb-llama-cpp-tp1
model instance id
meta-models-muse-glimmer-30b-gguf--dynamic-q4-k-xl-dflash2-q4-k-m
recipe source
localmaxxing
schema version
local-ai-registry/v1
speed sweep ids
muse-glimmer-30b-dynamic-q4-k-xl-dflash2-q4-k-m-radeon-ai-pro-r9700-32gb-llama-cpp-tp1-sweep
status
candidate

capabilities

engine

name
llama.cpp
version
0.3.0-dev / build 10680 / commit d7bd3bfca

serving

max concurrency
1
max context tokens
32,768
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
rocm
localmaxxing · hardware label
Radeon AI Pro R9700
localmaxxing · notes
Lucebox R9700-only Muse Glimmer 30B deployment. Full target is meta-models Dynamic Q4_K_XL (18.30 GiB), paired with z-lab/Muse-Glimmer-30B-DFlash2-GGUF Q4_K_M draft (1.53 GiB) and the Q4 multimodal projector. One discarded warmup plus three measured streamed requests; medians of per-run rates/timing. Low reasoning, deterministic temperature 0/top-k 1, cache_prompt=false, one slot, 32K context, Q8 K/V, flash attention, DFlash2 block size 16 / draft n-max 15. Decode and prefill are server-authoritative; total is (82 prompt + 225 output tokens) / dispatch-to-DONE wall time; TTFT is dispatch to first non-empty generated reasoning/content token. Peak VRAM is the maximum service-process amdgpu DRM fdinfo drm-memory-vram allocation on R9700 PCI 0000:c5:00.0, sampled every 10 ms, not device capacity. All measured outputs were byte-identical (SHA-256 48e5a6129f994b07b862d12db2490a847ad000e55a0ac9f1772dac527004c128). The Q4 DFlash2 draft accelerates verification while preserving the target model output; it is not a standalone model.
localmaxxing · observed command
llama-server -m Muse-Glimmer-30B-KQuant-Dynamic-Q4_K_XL.gguf --mmproj mmproj-Muse-Glimmer-30B-Q4_K_M.gguf -md Muse-Glimmer-30B-DFlash2-Q4_K_M.gguf --spec-type draft-dflash --spec-draft-n-max 15 -ngld 99 -ngl 99 -c 32768 -np 1 -ctk q8_0 -ctv q8_0 -fa on --jinja --chat-template-kwargs '{reasoning_strength: low}'
localmaxxing · run id
cmtdonksq005olm01t0itg4cs
localmaxxing · tokenized · arguments
llama-server, -m, Muse-Glimmer-30B-KQuant-Dynamic-Q4_K_XL.gguf, --mmproj, mmproj-Muse-Glimmer-30B-Q4_K_M.gguf, -md, Muse-Glimmer-30B-DFlash2-Q4_K_M.gguf, --spec-type, draft-dflash, --spec-draft-n-max, 15, -ngld, 99, -ngl, 99, -c, 32768, -np, 1, -ctk, q8_0, -ctv, q8_0, -fa, on, --jinja, --chat-template-kwargs, {reasoning_strength: low}
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/cmtdonksq005olm01t0itg4cs