Recipe
muse-glimmer-30b-unsloth-dynamic-q6-k-xl-rtx-3090-24gb-llama-cpp-tp2
muse-glimmer-30b-unsloth-dynamic-q6-k-xl-rtx-3090-24gb-llama-cpp-tp2Observed LocalMaxxing leaderboard run. Evidence for compatibility, not an executable launch contract.
Record
- Status
- candidate
- Source
- localmaxxing
- Engine
- llama.cpp
- Accelerators
- 2
- Tensor parallel
- 2
- Context tokens
- 2,048
- chat
- unknown
- reasoning
- unknown
- tools
- unknown
- vision
- unknown
Hugging Face model card
Identity
https://huggingface.co/unsloth/Muse-Glimmer-30B-GGUF- Repository
- unsloth/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
Observed source tokens
Mechanical split of the source command. Unverified against the engine CLI. The registry does not offer Run for this recipe.
/home/lotto/llama.cpp/build/bin/llama-bench-m/home/lotto/models/Muse-Glimmer-30B-GGUF/Muse-Glimmer-30B-UD-Q6_K_XL.gguf-p0-n128-ngl999-smlayer-ts1/1-fa1-ctkq8_0-ctvq8_0-r5-ojson
| Flag | Value |
|---|---|
-m | /home/lotto/models/Muse-Glimmer-30B-GGUF/Muse-Glimmer-30B-UD-Q6_K_XL.gguf |
-p | 0 |
-n | 128 |
-ngl | 999 |
-sm | layer |
-ts | 1/1 |
-fa | 1 |
-ctk | q8_0 |
-ctv | q8_0 |
-r | 5 |
-o | json |
Measured speed
| Concurrency | Context | Prefill | Decode | TTFT ms | Status | Sweep |
|---|---|---|---|---|---|---|
| 1 | 2,048 | 1,604 | 31.7 | 176.1 | observed | muse-glimmer-30b-unsloth-dynamic-q6-k-xl-rtx-3090-24gb-llama-cpp-tp2-sweep |
Remaining fields
Identity, launch, related records, and measured speed are shown above. This is the rest of the normalized record.
- hardware count
- 2
- hardware id
- rtx-3090-24gb
- id
- muse-glimmer-30b-unsloth-dynamic-q6-k-xl-rtx-3090-24gb-llama-cpp-tp2
- model instance id
- unsloth-muse-glimmer-30b-gguf--unsloth-dynamic-q6-k-xl
- recipe source
- localmaxxing
- schema version
- local-ai-registry/v1
- speed sweep ids
- muse-glimmer-30b-unsloth-dynamic-q6-k-xl-rtx-3090-24gb-llama-cpp-tp2-sweep
- status
- candidate
capabilities
engine
- name
- llama.cpp
serving
- max context tokens
- 2,048
- tensor parallel
- 2
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
- serving.max concurrency · provenance · captured at
- 2026-08-31T23:03:15Z
serving.max concurrency · reason server-capacity-not-evidenced
serving.max concurrency · state unknown
metadata
- localmaxxing · backend
- cuda
- localmaxxing · batch size
- 1
- localmaxxing · hardware label
- RTX 3090
- localmaxxing · notes
- Setup: llama.cpp 8ad8aef, CUDA, 2x NVIDIA RTX 3090 24GB each (48GB total VRAM), all layers GPU-offloaded with layer split 1:1, flash attention, 131072-token server context, Q8_0 K/V cache, one parallel slot, BF16 Muse Glimmer vision projector. Decode: 31.671169 tok/s from llama-bench, 0 prompt tokens, 128 output tokens, 5 repetitions and default warmup. Prefill: 1604.010013 tok/s from a separate llama-bench prompt-processing test with 2048 prompt tokens, 0 output tokens, ubatch 2048, 5 repetitions; the shorter 66-token endpoint estimate was 374.8 tok/s because fixed TTFT/request overhead dominated. Endpoint timing: median TTFT 176.09 ms and total throughput 46.8 tok/s over 3 measured iterations after 1 warmup, using 66 actual prompt tokens and 128 generated tokens. Server command: /home/lotto/llama.cpp/build/bin/llama-server --model /home/lotto/models/Muse-Glimmer-30B-GGUF/Muse-Glimmer-30B-UD-Q6_K_XL.gguf --mmproj /home/lotto/models/Muse-Glimmer-30B-GGUF/mmproj-Muse-Glimmer-30B-BF16.gguf --alias Muse-Glimmer-30B-Q6 --host 0.0.0.0 --port 8001 --device CUDA0,CUDA1 --n-gpu-layers all --split-mode layer --tensor-split 1,1 --ctx-size 131072 --cache-type-k q8_0 --cache-type-v q8_0 --flash-attn on --parallel 1 --cont-batching --jinja --temp 1.0 --top-p 0.95 --top-k 64 --metrics --perf
- localmaxxing · observed command
- /home/lotto/llama.cpp/build/bin/llama-bench -m /home/lotto/models/Muse-Glimmer-30B-GGUF/Muse-Glimmer-30B-UD-Q6_K_XL.gguf -p 0 -n 128 -ngl 999 -sm layer -ts 1/1 -fa 1 -ctk q8_0 -ctv q8_0 -r 5 -o json
- localmaxxing · run id
- cmsnf37pe00bqo001jze3l8kj
- localmaxxing · tokenized · arguments
- /home/lotto/llama.cpp/build/bin/llama-bench, -m, /home/lotto/models/Muse-Glimmer-30B-GGUF/Muse-Glimmer-30B-UD-Q6_K_XL.gguf, -p, 0, -n, 128, -ngl, 999, -sm, layer, -ts, 1/1, -fa, 1, -ctk, q8_0, -ctv, q8_0, -r, 5, -o, json
- localmaxxing · tokenized · fidelity
- faithful
provenance
- captured at
- 2026-08-30T09:10:02Z
sources
| captured at | kind | url |
|---|---|---|
| 2026-08-30T09:10:02Z | normalized-recipe | www.localmaxxing.com/en/runs/cmsnf37pe00bqo001jze3l8kj ↗ |