Recipe

qwen3-6-27b-ud-q4-k-xl-rtx-3090-ti-24gb-llama-cpp-tp1

qwen3-6-27b-ud-q4-k-xl-rtx-3090-ti-24gb-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
buun-llama-cpp b8959-7343d30c6
Accelerators
1
Tensor parallel
1
Context tokens
4,096
Max concurrency
1
chat
unknown
reasoning
unknown
tools
unknown
vision
unknown

Hugging Face model card

Identity

https://huggingface.co/unsloth/Qwen3.6-27B
Repository
unsloth/Qwen3.6-27B
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/cmohkdafg000nl704l43fxf7j

Observed source tokens

Mechanical split of the source command. Unverified against the engine CLI. The registry does not offer Run for this recipe.

  1. /home/steven/dev/buun-llama-cpp/build/bin/llama-server
  2. -m
  3. Qwen3.6-27B-UD-Q4_K_XL.gguf
  4. -md
  5. dflash-draft-3.6-q4_k_m.gguf
  6. --spec-type
  7. dflash
  8. -dev
  9. CUDA0
  10. -devd
  11. CUDA0
  12. -sm
  13. none
  14. -ngl
  15. 99
  16. -ngld
  17. 99
  18. -np
  19. 1
  20. -c
  21. 4096
  22. -cd
  23. 256
  24. -fa
  25. on
  26. -b
  27. 512
  28. -ub
  29. 128
  30. --draft-max
  31. 16
  32. --draft-min
  33. 1
  34. --temp
  35. 0
  36. --reasoning
  37. off
  38. --host
  39. 127.0.0.1
  40. --port
  41. 8081
  42. --no-webui
FlagValue
-mQwen3.6-27B-UD-Q4_K_XL.gguf
-mddflash-draft-3.6-q4_k_m.gguf
--spec-typedflash
-devCUDA0
-devdCUDA0
-smnone
-ngl99
-ngld99
-np1
-c4096
-cd256
-faon
-b512
-ub128
--draft-max16
--draft-min1
--temp0
--reasoningoff
--host127.0.0.1
--port8081

Measured speed

ConcurrencyContextPrefillDecodeTTFT msStatusSweep
14,096155.967.6observedqwen3-6-27b-ud-q4-k-xl-rtx-3090-ti-24gb-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
rtx-3090-ti-24gb
id
qwen3-6-27b-ud-q4-k-xl-rtx-3090-ti-24gb-llama-cpp-tp1
model instance id
unsloth-qwen3-6-27b--ud-q4-k-xl
recipe source
localmaxxing
schema version
local-ai-registry/v1
speed sweep ids
qwen3-6-27b-ud-q4-k-xl-rtx-3090-ti-24gb-llama-cpp-tp1-sweep
status
candidate

capabilities

engine

name
llama.cpp
version
buun-llama-cpp b8959-7343d30c6

serving

max concurrency
1
max context tokens
4,096
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
cuda
localmaxxing · hardware label
RTX 3090 Ti
localmaxxing · notes
DFlash speculative decoding bench reproducing joaosump's methodology: 2 warmup requests + recorded request against llama-server /v1/chat/completions. Recorded request used temperature=0, max_tokens=500, prompt: Write a Python binary search tree implementation with insert, search, delete, and traversal methods. DFlash acceptance on recorded run: 451 accepted draft tokens out of 697 draft tokens (64.7%) -- workload-equivalent to joaosump's 452/693. Wall-clock 155.94 t/s; server-side eval rate 159.62 t/s. Same target + draft combo as joaosump's #1 entry (UD-Q4_K_XL fp_target + q4_k_m DFlash draft); same buun-llama-cpp build (one commit newer at 7343d30c6 vs his 30759dfa0). Server launched with -c 4096; DFlash recurrent backup makes usable slot n_ctx 2048.
localmaxxing · run id
cmohkdafg000nl704l43fxf7j
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/cmohkdafg000nl704l43fxf7j