Local AI Registry

Recipe

qwen3-8-27b-q4-0-rocmfp4-strix-ryzen-ai-max-plus-395-128gb-llama-cpp-tp1

qwen3-8-27b-q4-0-rocmfp4-strix-ryzen-ai-max-plus-395-128gb-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
ROCmFPX b9219 / 1faa48eef / ROCm 7.2.4
Accelerators
1
Tensor parallel
1
Context tokens
65,536
Max concurrency
1
chat
unknown
reasoning
unknown
tools
unknown
vision
unknown

Hugging Face model card

Identity

https://huggingface.co/kingjones777/Qwen3.8-27B-ROCmFP4-STRIX-MTP-GGUF
Repository
kingjones777/Qwen3.8-27B-ROCmFP4-STRIX-MTP-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/cmstq9z3803q3ms01f0r9qsko

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. Qwen3.8-27B-Q4_0_ROCMFP4_STRIX.gguf
  4. --spec-type
  5. draft-mtp
  6. --spec-draft-model
  7. mtp-Qwen3.8-27B-Q4_0.gguf
  8. --spec-draft-ngl
  9. 99
  10. --spec-draft-device
  11. ROCm0
  12. --spec-draft-n-max
  13. 3
  14. --spec-draft-n-min
  15. 0
  16. --spec-draft-p-min
  17. 0.0
  18. -ngl
  19. 999
  20. -fa
  21. 1
  22. --no-mmap
  23. --jinja
  24. -fit
  25. off
  26. --parallel
  27. 1
  28. -dev
  29. ROCm0
  30. -c
  31. 65536
  32. --metrics
FlagValue
-mQwen3.8-27B-Q4_0_ROCMFP4_STRIX.gguf
--spec-typedraft-mtp
--spec-draft-modelmtp-Qwen3.8-27B-Q4_0.gguf
--spec-draft-ngl99
--spec-draft-deviceROCm0
--spec-draft-n-max3
--spec-draft-n-min0
--spec-draft-p-min0.0
-ngl999
-fa1
-fitoff
--parallel1
-devROCm0
-c65536

Measured speed

ConcurrencyContextPrefillDecodeTTFT msStatusSweep
165,53630023.927,469.3observedqwen3-8-27b-q4-0-rocmfp4-strix-ryzen-ai-max-plus-395-128gb-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
ryzen-ai-max-plus-395-128gb
id
qwen3-8-27b-q4-0-rocmfp4-strix-ryzen-ai-max-plus-395-128gb-llama-cpp-tp1
model instance id
kingjones777-qwen3-8-27b-rocmfp4-strix-mtp-gguf--q4-0-rocmfp4-strix
recipe source
localmaxxing
schema version
local-ai-registry/v1
speed sweep ids
qwen3-8-27b-q4-0-rocmfp4-strix-ryzen-ai-max-plus-395-128gb-llama-cpp-tp1-sweep
status
candidate

capabilities

engine

name
llama.cpp
version
ROCmFPX b9219 / 1faa48eef / ROCm 7.2.4

serving

max concurrency
1
max context tokens
65,536
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
serving.max concurrency · provenance · captured at
2026-09-01T10:20:39Z
serving.max concurrency · reason explicit-source-server-capacity
serving.max concurrency · state known
serving.max context tokens · provenance · captured at
2026-09-01T10:20:39Z
serving.max context tokens · reason explicit-source-context-limit
serving.max context tokens · state known

metadata

localmaxxing · backend
rocm
localmaxxing · batch size
1
localmaxxing · hardware label
Ryzen AI Max 395
localmaxxing · notes
Corrected independent Horizon Strix Halo reproduction from internal NVMe-backed Btrfs storage. Exact STRIX base plus Q4_0 MTP draft head; Kyuz0 ROCmFP4 toolbox runtime b9219 / 1faa48eef / ROCm 7.2.4. Median of 3 measured runs after 1 warmup, using fixed cache-busting operator-style ~8K prompts shared across configurations, cache_prompt=false, greedy sampling (temperature 0/top_k 1), ignore_eos=true, and exactly 256 generated tokens. Native decode runs: 23.88, 23.22, 24.08 tok/s; median 23.88. Measured MTP acceptance: 0.567, 0.536, 0.567; median 0.567. No-MTP control was 13.34 tok/s, so n-max 3 improved decode by 78.95%; n-max 2 was 22.66 and n-max 4 was 21.86 tok/s. A pinned newer b9311 build regressed to 23.33 tok/s (23.50 with GPU_MAX_HW_QUEUES=2), so b9219 remains the best verified runtime. Internal cold readiness was 4.222 s. Publisher SHA256SUMS passed after relocation. A bounded Hermes operator workflow completed 9 tool calls, computed and matched SHA-256 and byte count, wrote a report, and read it back; final content was correct. The earlier external-drive measurements are superseded by this corrected internal-storage run.
localmaxxing · observed command
llama-server -m Qwen3.8-27B-Q4_0_ROCMFP4_STRIX.gguf --spec-type draft-mtp --spec-draft-model mtp-Qwen3.8-27B-Q4_0.gguf --spec-draft-ngl 99 --spec-draft-device ROCm0 --spec-draft-n-max 3 --spec-draft-n-min 0 --spec-draft-p-min 0.0 -ngl 999 -fa 1 --no-mmap --jinja -fit off --parallel 1 -dev ROCm0 -c 65536 --metrics
localmaxxing · run id
cmstq9z3803q3ms01f0r9qsko
localmaxxing · tokenized · arguments
llama-server, -m, Qwen3.8-27B-Q4_0_ROCMFP4_STRIX.gguf, --spec-type, draft-mtp, --spec-draft-model, mtp-Qwen3.8-27B-Q4_0.gguf, --spec-draft-ngl, 99, --spec-draft-device, ROCm0, --spec-draft-n-max, 3, --spec-draft-n-min, 0, --spec-draft-p-min, 0.0, -ngl, 999, -fa, 1, --no-mmap, --jinja, -fit, off, --parallel, 1, -dev, ROCm0, -c, 65536, --metrics
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/cmstq9z3803q3ms01f0r9qsko ↗