YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
Behavioral model ablations: data and model artifacts
This repository accompanies the preprint Better Behavioral Prediction, More Faithful Model Ablations? Evidence from Sequential Choice by Hanbo Xie. It contains synthetic sequential-choice data, evaluation exports, and frozen trained-model artifacts. No new human participant data are included.
Contents
task_a/andtask_b/: 28 LLaMA SFT adapters (two tasks, full-input and choice-only fits, seven reward weights).MANIFEST.jsonrecords their file hashes. The base model is not redistributed and remains subject to its original access terms.review_artifacts/: frozen synthetic training, validation, and test records; independent-test prediction exports; 168 validation-selected GRU and Transformer checkpoints; cognitive-model fits; and audit records. The directory name is retained from the verified review snapshot.preprint_code.zip: source code, requirements, tests, frozen figure inputs, and plotting scripts. Its contents match the linked GitHub code repository.
The main manuscript figures use independent-test results from validation-selected checkpoints. Some supplementary controls and the Figure 1A illustration use training or validation data, as labeled in the paper. Rebuilding figures from saved summaries is different from retraining or rerunning model inference. For commands and dataset splits, start with preprint_code.zip/behavioral_model_ablations_code/README.md.
This repository does not grant a new license for the LLaMA base model or other upstream dependencies.