Recipe
muse-glimmer-30b-ud-q8-k-xl-intel-arc-pro-b70-32gb-llama-cpp-tp4
muse-glimmer-30b-ud-q8-k-xl-intel-arc-pro-b70-32gb-llama-cpp-tp4Observed LocalMaxxing leaderboard run. Evidence for compatibility, not an executable launch contract.
Record
- Status
- candidate
- Source
- localmaxxing
- Engine
- llama.cpp
- Engine version
- 10433; 030ebb558 + Muse TP4/WOQ patch 3965702b96fe
- Accelerators
- 4
- Tensor parallel
- 4
- Context tokens
- 32,768
- Max concurrency
- 1
- 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.
llama-server-mMuse-Glimmer-30B-UD-Q8_K_XL.gguf-ngl99-c32768--parallel1-b1024-ub1024-faon-smtensor-bs-lv4--spec-typedraft-dflash--spec-draft-modeldflash-bf16.gguf--spec-draft-n-max15--spec-draft-p-min0--spec-draft-ngl99
| Flag | Value |
|---|---|
-m | Muse-Glimmer-30B-UD-Q8_K_XL.gguf |
-ngl | 99 |
-c | 32768 |
--parallel | 1 |
-b | 1024 |
-ub | 1024 |
-fa | on |
-sm | tensor |
-lv | 4 |
--spec-type | draft-dflash |
--spec-draft-model | dflash-bf16.gguf |
--spec-draft-n-max | 15 |
--spec-draft-p-min | 0 |
--spec-draft-ngl | 99 |
Environment
| Variable | Value |
|---|---|
GGML_META_PARALLEL_SUBMIT | 1 |
GGML_SYCL_BF16_GRAPH_CONVERSION_CACHE | 0 |
GGML_SYCL_COMM_ARGMAX | 1 |
GGML_SYCL_Q8_0_WOQ_BF16 | 1 |
GGML_SYCL_Q8_0_WOQ_FIXED16 | 1 |
LLAMA_DFLASH_TP_GREEDY | 1 |
LLAMA_TP_BACKEND_SAMPLING | 1 |
ONEAPI_DEVICE_SELECTOR | level_zero:0,1,2,3 |
Measured speed
| Concurrency | Context | Prefill | Decode | TTFT ms | Status | Sweep |
|---|---|---|---|---|---|---|
| 1 | 32,768 | — | 161.9 | 2,872.1 | observed | muse-glimmer-30b-ud-q8-k-xl-intel-arc-pro-b70-32gb-llama-cpp-tp4-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
- muse-glimmer-30b-ud-q8-k-xl-intel-arc-pro-b70-32gb-llama-cpp-tp4
- model instance id
- unsloth-muse-glimmer-30b-gguf--ud-q8-k-xl
- recipe source
- localmaxxing
- schema version
- local-ai-registry/v1
- speed sweep ids
- muse-glimmer-30b-ud-q8-k-xl-intel-arc-pro-b70-32gb-llama-cpp-tp4-sweep
- status
- candidate
capabilities
engine
- name
- llama.cpp
- version
- 10433; 030ebb558 + Muse TP4/WOQ patch 3965702b96fe
serving
- max concurrency
- 1
- max context tokens
- 32,768
- 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
- serving.max concurrency · provenance · captured at
- 2026-08-31T23:03:15Z
serving.max concurrency · reason server-capacity-derived-from-source-evidence
serving.max concurrency · state known
metadata
- localmaxxing · backend
- xpu
- localmaxxing · batch size
- 1
- localmaxxing · hardware label
- Intel Arc Pro B70
- localmaxxing · notes
- Muse-Glimmer-30B UD-Q8_K_XL on four Intel Arc Pro B70 GPUs, TP4, one cold request at a time. Primary result is the median across a frozen 15-prompt suite for the 99 inter-token intervals between generated-token events 1 and 100 after TTFT: 161.900 tok/s; p10 108.574, mean 175.813, prompt-bootstrap one-sided 95% lower bound 127.082. Every prompt ran exactly once, all reported cached_tokens=0, and no prompt/KV/context/response/ngram/history reuse was used. Pretrained BF16 DFlash proposals are verified by the declared Q8 target; no drafter training was performed. Runtime uses fixed-N16 BF16-activation/S8 group32 oneDNN WOQ and distributed ARGMAX. This is a target-verified Q8 result, not BF16/lossless or universally token-exact; two deterministic structured tasks were exact while prose can take target-approved near-tie branches. Full natural-completion median after TTFT was 68.586 tok/s. Complete patches, configs, hashes, raw evidence, and runnable repro are in results/muse-glimmer-30b-q8-woq-b70 and repro/muse-glimmer-30b-q8-woq-b70-100tps-20260813.
- localmaxxing · observed command
- ONEAPI_DEVICE_SELECTOR=level_zero:0,1,2,3 GGML_SYCL_Q8_0_WOQ_BF16=1 GGML_SYCL_Q8_0_WOQ_FIXED16=1 GGML_SYCL_BF16_GRAPH_CONVERSION_CACHE=0 GGML_META_PARALLEL_SUBMIT=1 GGML_SYCL_COMM_ARGMAX=1 LLAMA_DFLASH_TP_GREEDY=1 LLAMA_TP_BACKEND_SAMPLING=1 llama-server -m Muse-Glimmer-30B-UD-Q8_K_XL.gguf -ngl 99 -c 32768 --parallel 1 -b 1024 -ub 1024 -fa on -sm tensor -bs -lv 4 --spec-type draft-dflash --spec-draft-model dflash-bf16.gguf --spec-draft-n-max 15 --spec-draft-p-min 0 --spec-draft-ngl 99
- localmaxxing · run id
- cmss8515c00n0ms01n3begqgg
- localmaxxing · tokenized · arguments
- llama-server, -m, Muse-Glimmer-30B-UD-Q8_K_XL.gguf, -ngl, 99, -c, 32768, --parallel, 1, -b, 1024, -ub, 1024, -fa, on, -sm, tensor, -bs, -lv, 4, --spec-type, draft-dflash, --spec-draft-model, dflash-bf16.gguf, --spec-draft-n-max, 15, --spec-draft-p-min, 0, --spec-draft-ngl, 99
- localmaxxing · tokenized · environment · GGML META PARALLEL SUBMIT
- 1
- localmaxxing · tokenized · environment · GGML SYCL BF16 GRAPH CONVERSION CACHE
- 0
- localmaxxing · tokenized · environment · GGML SYCL COMM ARGMAX
- 1
- localmaxxing · tokenized · environment · GGML SYCL Q8 0 WOQ BF16
- 1
- localmaxxing · tokenized · environment · GGML SYCL Q8 0 WOQ FIXED16
- 1
- localmaxxing · tokenized · environment · LLAMA DFLASH TP GREEDY
- 1
- localmaxxing · tokenized · environment · LLAMA TP BACKEND SAMPLING
- 1
- localmaxxing · tokenized · environment · ONEAPI DEVICE SELECTOR
- level_zero:0,1,2,3
- 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/cmss8515c00n0ms01n3begqgg ↗ |