2 days ago
-
GPT-5.5 LoRAs: same-day evaluation with SelectionARM (November 4)
-
Add decoder-4b-lora-gpt55-pre-p16: 4B LoRA selector on the same 18,786 panels with GPT-5.5 pre-execution labels
-
Add decoder-4b-lora-gpt55-k3-p16: 4B LoRA selector on 18,786 GPT-5.5 K=3 lookahead panels
5 days ago
-
README: trim secondary details from the LoRA section
-
README: selector call count wording
-
README: add decoder-4b-lora-recipe2-2.6k (4B LoRA selector, Recipe 2 2.6K) with training and SFT/RL-actor results
-
Rename lora-recipe2-2612/ to decoder-4b-lora-recipe2-2.6k/
-
Add 4B LoRA selector, Recipe 2 (2.6K states): merged weights + rank-32 adapter
9 days ago
-
Model card: plain-language options and method description
-
Model card: quick-start and full-step usage versions
-
Add actor prompt builder (build_prompt) and system_prompt.txt; usage example
-
Model card: add o4-mini OM2W numbers, drop offline numbers
-
Add verified standalone inference module and safetensors weights; finalize model card
-
Model card: remove caveats section
-
Original selection-head checkpoints (piotr-data, lookahead-k3-1380), configs and model card
-
initial commit
