GNAI-009 v0.5: Swift 1.5 als dritter Kandidat + Benchmark-Design; Rollen erweitert

Betreiberbeschluss 01.11.2026: Swift 1.5 Q6_K aufgenommen (GNAI-009 3.4,
Artefakt gepinnt: Revision 14bfe4b, SHA-256 cd2eb65 aus HF-Blob-Metadaten,
22,2 GiB). Testmatrix um S-Reihe und AB3/AB4 erweitert (AB3 isoliert die
Thinking-Optimierung Swift 1.0 vs 1.5, AB4 ist der Praxisvergleich).
Benchmark-Design ergaenzt: lm-evaluation-harness gegen die interne Engine-API
(AIME, GPQA Diamond, MATH-500, IFEval) plus Kernkennzahl Score je
Median-Reasoning-Token aus der getrennten reasoning_content.

model_store: drittes Artefakt; inference_engine: Profil swift15
(Unit gnai-testing-llama-swift15, Alias swift-1.5-qwen3.8-27b-q6k).

Generated by Mistral Vibe.
Co-Authored-By: Mistral Vibe <vibe@mistral.ai>
This commit is contained in:
2026-11-01 17:46:47 +01:00
parent 345eb36c2b
commit bc37972a09
3 changed files with 130 additions and 11 deletions
@@ -83,6 +83,25 @@ inference_engine_profiles:
mtp:
enabled: false
spec_draft_n_max: 2
swift-1.5-qwen3.8-27b-q6k:
unit_suffix: swift15
description: Swift 1.5 reasoning-efficient finetune of Qwen3.8-27B (Q6_K)
model_directory: swift-1.5-qwen3.8-27b
model_file: Swift-1.5-Qwen3.8-27B-Q6_K.gguf
alias: swift-1.5-qwen3.8-27b-q6k
ctx_size: 131072
n_predict: 32768
parallel: 1
reasoning_effort: xhigh
temp: 1.0
top_p: 0.95
top_k: 20
min_p: 0.0
presence_penalty: 0.0
repeat_penalty: 1.0
mtp:
enabled: false
spec_draft_n_max: 2
# Derived paths (do not override).
inference_engine_quadlet_dir: "{{ inference_engine_quadlet_user_dir }}/{{ inference_engine_uid }}"
@@ -34,6 +34,21 @@ model_store_artifacts:
revision: 4ca720788d1e01f1bff70c033e0d0028fd02e502
origin: Unsloth dynamic quant (UD) of Qwen3.8-27B, Q6_K
license: Apache-2.0
- profile: swift-1.5-qwen3.8-27b-q6k
directory: swift-1.5-qwen3.8-27b
file: Swift-1.5-Qwen3.8-27B-Q6_K.gguf
sha256: cd2eb6597026ce4eb86100e830be24f01461f84b89d4708634c2ec3d43b67fa8
url: https://huggingface.co/ukisai/Swift-1.5-Qwen3.8-27B-GGUF/resolve/14bfe4b42be4a925d98816db830155f476c605e7/Swift-1.5-Qwen3.8-27B-Q6_K.gguf
repository: ukisai/Swift-1.5-Qwen3.8-27B-GGUF
revision: 14bfe4b42be4a925d98816db830155f476c605e7
origin: >-
Swift 1.5 reasoning-efficient finetune of Qwen3.8-27B (UkisAI), Q6_K;
SHA-256 from the Hugging Face blob metadata (the repository publishes
no SHA256SUMS file). GNAI-009 section 3.4.
license: >-
Qwen base model Apache-2.0; Swift changes under Swift Open License 1.0
(free personal, research, education, evaluation and commercial use up to
the revenue limit defined in the model card).
model_store_vision_artifact:
profile: swift-qwen3.8-27b-vision-mmproj