Local AI Registry

Recipe

qwen3-6-27b-q4-0-intel-arc-pro-b70-32gb-llama-cpp-tp1-2avo1q5t

qwen3-6-27b-q4-0-intel-arc-pro-b70-32gb-llama-cpp-tp1-2avo1q5t
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
f535774-dirty
Accelerators
1
Tensor parallel
1
Context tokens
2,048
chat
unknown
reasoning
unknown
tools
unknown
vision
unknown

Hugging Face model card

Identity

https://huggingface.co/unsloth/Qwen3.6-27B-GGUF
Repository
unsloth/Qwen3.6-27B-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/cmogi7eik0002jx042avo1q5t

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-bench
  2. -m
  3. Qwen3.6-27B-Q4_0.gguf
  4. -dev
  5. Vulkan0
  6. -ngl
  7. 99
  8. -fa
  9. 0
  10. -p
  11. 0
  12. -n
  13. 512
  14. -d
  15. 0
  16. -sm
  17. layer
  18. -b
  19. 512
  20. -ub
  21. 64
  22. -ctk
  23. f16
  24. -ctv
  25. f16
  26. -t
  27. 8
  28. --poll
  29. 50
  30. -r
  31. 3
  32. -o
  33. jsonl
  34. --no-warmup
  35. --progress
FlagValue
-mQwen3.6-27B-Q4_0.gguf
-devVulkan0
-ngl99
-fa0
-p0
-n512
-d0
-smlayer
-b512
-ub64
-ctkf16
-ctvf16
-t8
--poll50
-r3
-ojsonl

Environment

VariableValue
GGML_VK_ALLOW_GRAPHICS_QUEUE1
LD_LIBRARY_PATH/home/occulink/arc-b70-llm/src/mesa/install-anv-main/lib/x86_64-linux-gnu:/home/occulink/arc-b70-llm/tools/vulkan-sdk/1.4.341.1/x86_64/lib:/home/occulink/arc-b70-llm/src/llama.cpp/build-vulkan-sdk/bin
VK_DRIVER_FILES/home/occulink/arc-b70-llm/src/mesa/install-anv-main/share/vulkan/icd.d/intel_icd.x86_64.json
VK_ICD_FILENAMES/home/occulink/arc-b70-llm/src/mesa/install-anv-main/share/vulkan/icd.d/intel_icd.x86_64.json

Measured speed

ConcurrencyContextPrefillDecodeTTFT msStatusSweep
12,048—25.2—observedqwen3-6-27b-q4-0-intel-arc-pro-b70-32gb-llama-cpp-tp1-2avo1q5t-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
intel-arc-pro-b70-32gb
id
qwen3-6-27b-q4-0-intel-arc-pro-b70-32gb-llama-cpp-tp1-2avo1q5t
model instance id
unsloth-qwen3-6-27b-gguf--q4-0
recipe source
localmaxxing
schema version
local-ai-registry/v1
speed sweep ids
qwen3-6-27b-q4-0-intel-arc-pro-b70-32gb-llama-cpp-tp1-2avo1q5t-sweep
status
candidate

capabilities

engine

name
llama.cpp
version
f535774-dirty

serving

max context tokens
2,048
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
2026-08-31T23:03:15Z
serving.max concurrency · reason server-capacity-not-evidenced
serving.max concurrency · state unknown

metadata

localmaxxing · backend
vulkan
localmaxxing · batch size
1
localmaxxing · hardware label
Intel Arc Pro B70
localmaxxing · notes
Linux best-of-3 generation-only llama-bench result on Intel Arc Pro B70. Local Mesa ANV main 26.2.0-devel git-642bed9eba selected via VK_DRIVER_FILES, llama.cpp f535774 with local B70 0xe223 core-count patch, built against LunarG Vulkan SDK 1.4.341.1. Key setting: GGML_VK_ALLOW_GRAPHICS_QUEUE=1 to use the graphics+compute queue family; -fa 0 was faster than -fa 1 on Linux. Samples: 25.1709, 25.1606, 25.1364 tok/s. Telemetry during similar decode: active card power avg ~211.5 W, GT0 ~2788 MHz, VRAM max ~80 C. This still trails the same system's Windows Vulkan result of 28.8 tok/s.
localmaxxing · observed command
GGML_VK_ALLOW_GRAPHICS_QUEUE=1 VK_DRIVER_FILES=/home/occulink/arc-b70-llm/src/mesa/install-anv-main/share/vulkan/icd.d/intel_icd.x86_64.json VK_ICD_FILENAMES=/home/occulink/arc-b70-llm/src/mesa/install-anv-main/share/vulkan/icd.d/intel_icd.x86_64.json LD_LIBRARY_PATH=/home/occulink/arc-b70-llm/src/mesa/install-anv-main/lib/x86_64-linux-gnu:/home/occulink/arc-b70-llm/tools/vulkan-sdk/1.4.341.1/x86_64/lib:/home/occulink/arc-b70-llm/src/llama.cpp/build-vulkan-sdk/bin llama-bench -m Qwen3.6-27B-Q4_0.gguf -dev Vulkan0 -ngl 99 -fa 0 -p 0 -n 512 -d 0 -sm layer -b 512 -ub 64 -ctk f16 -ctv f16 -t 8 --poll 50 -r 3 -o jsonl --no-warmup --progress
localmaxxing · run id
cmogi7eik0002jx042avo1q5t
localmaxxing · tokenized · arguments
llama-bench, -m, Qwen3.6-27B-Q4_0.gguf, -dev, Vulkan0, -ngl, 99, -fa, 0, -p, 0, -n, 512, -d, 0, -sm, layer, -b, 512, -ub, 64, -ctk, f16, -ctv, f16, -t, 8, --poll, 50, -r, 3, -o, jsonl, --no-warmup, --progress
localmaxxing · tokenized · environment · GGML VK ALLOW GRAPHICS QUEUE
1
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/cmogi7eik0002jx042avo1q5t ↗