Files

T

L'électron rare 5e63ceb782 Refactor rewrite prompt and update tests for narrative generation

- Updated the rewrite prompt to remove markdown elements such as titles, bullet points, and code fences, ensuring a cleaner narrative output.
- Enhanced the GenerationPipeline tests to verify that code fences and chapter titles are stripped before manuscript promotion.
- Adjusted NextLots tests to include new functionality for command runners and added tests for deduplication of headings and timeout handling.
- Introduced new documentation files detailing project context and execution plans for March 2026, outlining current project status and objectives.
- Added operational memory documents to summarize project state and decisions for resuming work on the ai-novel-engine.

2026-03-16 06:03:24 +01:00

2.3 KiB

Raw Permalink Blame History

Memoire de reprise - 14 mars 2026

Memoire operationnelle pour reprendre ai-novel-engine sans refaire toute l'enquete runtime.

Etat confirme

la suite locale passe avec 48 tests
la reference locale reste apple-coreml:qwen3.5-4b-onnx-q4f16
apple-coreml:qwen2.5-0.5b-instruct-onnx reste quality_blocked sur un blocage de fin / garde-fou
ollama:qwen2.5:7b est maintenant a lire comme un blocage runtime, pas comme un simple blocage qualite
ollama:qwen2.5:1.5b est dans le meme etat runtime que le 7b

Reports du 14 mars 2026 a retenir

automation/reports/20260314T085946Z:
- ollama:qwen2.5:7b -> provider_failed
- status=ollama_runtime_unhealthy
- note utile: HTTP 500 Internal Server Error
automation/reports/20260314T100648Z:
- ollama:qwen2.5:1.5b -> provider_failed
- meme symptome utile: HTTP 500 Internal Server Error

Mesures live utiles

Apple warm-up direct sur :8100: 2.76s
Apple requete prose representative (96 tokens): 39.99s
llama-server temporaire sur le blob qwen2.5:1.5b: reponse en 0.31s

Ce que cela veut dire

Apple n'est pas le sujet critique du jour
Ollama natif 0.17.7 reste le composant qui casse
les blobs GGUF qwen2.5 presents dans ~/.ollama/models/blobs/ sont sains et reutilisables via llama.cpp

Decision de reprise

ne plus perdre du temps a tuner rewrite / repair tant que les reruns qwen2.5 passent par un runtime instable
prioriser un contournement llama-server / llama.cpp avant tout nouveau lot comparatif
garder apple-coreml:qwen3.5-4b-onnx-q4f16 comme reference stable

Fichiers et chemins utiles

blob qwen2.5:1.5b: /Users/electron/.ollama/models/blobs/sha256-183715c435899236895da3869489cc30ac241476b4971a20285b1a462818a5b4
blob qwen2.5:7b: /Users/electron/.ollama/models/blobs/sha256-2bada8a7450677000f678be90653b85d364de7db25eb5ea54136ada5f3933730
report 7b: automation/reports/20260314T085946Z/
report 1.5b: automation/reports/20260314T100648Z/
comparatif courant: docs/MODEL_COMPARISON_2026-03-08.md

Prochaine action utile

Rendre le contournement llama.cpp reusable cote runtime, puis rejouer:

python3 scripts/run_next_lots.py --lot priority_models
python3 scripts/run_next_lots.py --lot baselines

2.3 KiB Raw Permalink Blame History