12 Commits
Author SHA1 Message Date
kantan-kanto bb582fd373 Ask local d1 GGUF models in their own prompt format 2026-10-09 00:51:24 +09:00
kantan-kanto 889b0e0228 Run local Clef GGUF models with their decision head 2026-10-05 00:22:24 +09:00
kantan-kanto 1f3bd2f7d4 Batch Jev questions per state and limit max_jev_calls by requests 2026-10-03 14:02:41 +09:00
kantan-kanto 5284e69f90 Print Jev request and response counts per run and quiet httpx/typesafe_sdk INFO logs 2026-10-03 12:54:52 +09:00
kantan-kanto ce81103c06 Decode local Jev prompts with llama_decode directly to pass the Registry code scan 2026-09-29 20:20:27 +09:00
kantan-kanto 7a0dd36c1f Always unload the local Jev model after each run and remove keep_model_loaded 2026-09-28 22:12:05 +09:00
kantan-kanto e874207f33 Fix uniform Jev probabilities with upstream llama-cpp-python by reading logits from the context 2026-09-28 21:34:24 +09:00
kantan-kanto f84abc9e8a Add TypeSafe API backend to MultiOutputScript (Jev) with the key in api_key.txt 2026-09-28 18:52:05 +09:00
kantan-kanto 40b2a5fa92 Add jev.probabilities and an eight-way aspect ratio default Jev script 2026-09-26 18:17:30 +09:00
kantan-kanto d1975cc511 Fix Jev model listing for models on another drive 2026-09-26 18:02:28 +09:00
kantan-kanto a977b97d29 Replace TypeSafe API with local GGUF decisions via llama.cpp in MultiOutputScript (Jev) 2026-09-26 17:54:57 +09:00
kantan-kanto af8f8cc25d Add MultiOutputScript (Jev) node with jev namespace for scripts 2026-09-26 17:06:14 +09:00