Add artifact card and cleaning notes
Browse files
README.md
ADDED
|
@@ -0,0 +1,22 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# World-Model Post-Training Audit Artifacts
|
| 2 |
+
|
| 3 |
+
This repository contains selected fine-tuned checkpoints and cleaned training artifacts for *Does Learning to Predict the World Help Agents Act? Auditing World-Model Post-Training*.
|
| 4 |
+
|
| 5 |
+
## Contents
|
| 6 |
+
|
| 7 |
+
- `weights/`: selected final checkpoints for ALFWorld, ScienceWorld, and VisualWebArena.
|
| 8 |
+
- `data/alfworld/`: 20,000-sample GT and mismatched-target training parquet files.
|
| 9 |
+
- `data/sciworld/`: 20,000-sample GT and mismatched-target training parquet files.
|
| 10 |
+
- `data/vwa/`: VisualWebArena COIN training samples and the task metadata used by the release.
|
| 11 |
+
|
| 12 |
+
Evaluation results are intentionally omitted from this repository.
|
| 13 |
+
|
| 14 |
+
## Cleaning applied
|
| 15 |
+
|
| 16 |
+
- Removed the absolute ALFWorld `gamefile` path from `extra_info`; prompts and training targets are unchanged.
|
| 17 |
+
- Removed run-local VisualWebArena provenance fields from `extra_info`, retaining only the benchmark `site` label; prompts and images are unchanged.
|
| 18 |
+
- Removed internal split-history fields from `data/vwa/task_split_v2.json`.
|
| 19 |
+
- Retained `teacher_prompt_ids` for OPSD-compatible training.
|
| 20 |
+
- Retained `true_next_state` in mismatched-target files for auditability.
|
| 21 |
+
|
| 22 |
+
The fine-tuned weights inherit the terms of their base models. Benchmark data and derived artifacts remain subject to the upstream benchmark licenses and terms; please consult the cited upstream projects before redistribution.
|