episode_id stringlengths 26 26 | parsed_episode dict | turn_rewards listlengths 1 25 | original_rewards listlengths 2 2 |
|---|---|---|---|
01HFDGGH3TK18KFM2EBPRXSMV7 | {
"pk": "01HFDGGH3TK18KFM2EBPRXSMV7",
"scenario": "Two friends participating in a charity event for children in Syria",
"agent_1_name": "Hendrick Heinz",
"agent_2_name": "Ethan Smith",
"agent_1_background": "Hendrick Heinz is a 54-year-old male chef. He/him pronouns. Hendrick Heinz hosts monthly wine tasting ... | [
{
"turn": 0,
"agent": "Hendrick Heinz",
"agent_1_rewards": {
"believability": {
"reasoning": "<naturalness> Hendrick's opening line is natural and aligns with someone who values social gatherings and causes, despite his hedonistic nature. <consistency> His impulsivity may lead him to quick... | [
[
1.8571428571,
{
"believability": 8,
"relationship": 0,
"knowledge": 0,
"secret": 0,
"social_rules": 0,
"financial_and_material_benefits": 0,
"goal": 5,
"overall_score": 1.8571428571
}
],
[
2.7142857143,
{
"believability": 9,
"rela... |
01HNJ3RC6Y2HBYWDMRHZ145752 | {
"pk": "01HNJ3RC6Y2HBYWDMRHZ145752",
"scenario": "Conversation between two friends who have known each other for a long time. One of them used to make fun of the other because English was not their first language.",
"agent_1_name": "Sophia James",
"agent_2_name": "Miles Hawkins",
"agent_1_background": "Sophi... | [
{
"turn": 0,
"agent": "Sophia James",
"agent_1_rewards": {
"believability": {
"reasoning": "<naturalness> Sophia's apology is direct and sincere, aligning with her spontaneous and pleasure-seeking personality, suggesting she genuinely wants to mend the relationship. <consistency> Her actio... | [
[
3.7142857143,
{
"believability": 9,
"relationship": 4,
"knowledge": 3,
"secret": 0,
"social_rules": 0,
"financial_and_material_benefits": 0,
"goal": 10,
"overall_score": 3.7142857143
}
],
[
3.7142857143,
{
"believability": 9,
"rel... |
01HP2CBJ5JBT16ER5AD1X71NPV | {
"pk": "01HP2CBJ5JBT16ER5AD1X71NPV",
"scenario": "Agent1 has recently married into Agent2's family. Agent2 is Agent1's in-law. Agent1 has observed that the way Agent2's family celebrates holidays is quite different from what Agent1 is accustomed to, and would like to suggest some changes for the sake of the childr... | [
{
"turn": 0,
"agent": "Giselle Rousseau",
"agent_1_rewards": {
"believability": {
"reasoning": "<naturalness> Giselle's approach is consistent with their outgoing and inclusive personality. They bring up the topic in a friendly manner, making the suggestion sound like an opportunity for en... | [
[
3.5714285714000003,
{
"believability": 9,
"relationship": 2,
"knowledge": 7,
"secret": 0,
"social_rules": 0,
"financial_and_material_benefits": 0,
"goal": 7,
"overall_score": 3.5714285714000003
}
],
[
3.5714285714000003,
{
"believabilit... |
01HMXBAA9ZF70V0BVD8CCM4BSM | {"pk":"01HMXBAA9ZF70V0BVD8CCM4BSM","scenario":"Two people are in the same room. One has brought some(...TRUNCATED) | [{"turn":0,"agent":"Micah Stevens","agent_1_rewards":{"believability":{"reasoning":"<naturalness> Mi(...TRUNCATED) | [[2.8571428571,{"believability":9.0,"relationship":0.0,"knowledge":0.0,"secret":0.0,"social_rules":0(...TRUNCATED) |
01HJZ0TP3EF57T87S6QGFMMJAS | {"pk":"01HJZ0TP3EF57T87S6QGFMMJAS","scenario":"Agent1 gets extremely upset and loud while playing vi(...TRUNCATED) | [{"turn":0,"agent":"Sasha Ramirez","agent_1_rewards":{"believability":{"reasoning":"<naturalness> Sa(...TRUNCATED) | [[5.0,{"believability":10.0,"relationship":5.0,"knowledge":10.0,"secret":0.0,"social_rules":0.0,"fin(...TRUNCATED) |
01HPCVS2VK39FENATM72Q0KXYX | {"pk":"01HPCVS2VK39FENATM72Q0KXYX","scenario":"Two acquaintances meet at a charity event. One is a r(...TRUNCATED) | [{"turn":0,"agent":"Sasha Ramirez","agent_1_rewards":{"believability":{"reasoning":"<naturalness> Sa(...TRUNCATED) | [[3.1428571429,{"believability":9.0,"relationship":0.0,"knowledge":0.0,"secret":0.0,"social_rules":0(...TRUNCATED) |
01HPDJMCNVW3NBJQZ90MWTGNG9 | {"pk":"01HPDJMCNVW3NBJQZ90MWTGNG9","scenario":"Conversation between two friends at a local charity e(...TRUNCATED) | [{"turn":0,"agent":"Sophia James","agent_1_rewards":{"believability":{"reasoning":"<naturalness> Sop(...TRUNCATED) | [[3.7142857143,{"believability":9.0,"relationship":3.0,"knowledge":7.0,"secret":0.0,"social_rules":0(...TRUNCATED) |
01HNYRXN79K45S06AVKB850S5E | {"pk":"01HNYRXN79K45S06AVKB850S5E","scenario":"Agent1 and Agent2 are both regulars at a local park. (...TRUNCATED) | [{"turn":0,"agent":"Giselle Rousseau","agent_1_rewards":{"believability":{"reasoning":"<naturalness>(...TRUNCATED) | [[4.8571428571,{"believability":10.0,"relationship":4.0,"knowledge":8.0,"secret":0.0,"social_rules":(...TRUNCATED) |
01HNHZENB00KV5BZSMNDYAHVRB | {"pk":"01HNHZENB00KV5BZSMNDYAHVRB","scenario":"One person is offering a Tile Mate Item Tracker for a(...TRUNCATED) | [{"turn":0,"agent":"Lily Greenberg","agent_1_rewards":{"believability":{"reasoning":"<naturalness> L(...TRUNCATED) | [[3.5714285714000003,{"believability":8.0,"relationship":0.0,"knowledge":7.0,"secret":0.0,"social_ru(...TRUNCATED) |
01HFDET6YCBF63HVXQD1NXPACB | {"pk":"01HFDET6YCBF63HVXQD1NXPACB","scenario":"One person is offering a Tile Mate Item Tracker for a(...TRUNCATED) | [{"turn":0,"agent":"Lily Greenberg","agent_1_rewards":{"believability":{"reasoning":"<naturalness> L(...TRUNCATED) | [[1.7142857142999999,{"believability":10.0,"relationship":0.0,"knowledge":3.0,"secret":0.0,"social_r(...TRUNCATED) |
End of preview. Expand in Data Studio
Mental Model Annotation Dataset
Dataset for Mental Models for Multi-Agent Systems
This is the official annotation release accompanying the NeurIPS 2026 paper Mental Models for Multi-Agent Systems.
Paper resources: Project page | Code | Paper and arXiv links will be added upon release.
The paper studies explicit, recursive mental representations for multi-agent decision-making. This dataset contains the mental-state, reward, rationale, and preference supervision used for its SOTOPIA, BigToM, and MMRole experiments. The files augment these established benchmarks; they are not a new replacement for the original benchmark datasets. Independent configurations preserve each benchmark's schema and remain compatible with the Hugging Face Dataset Viewer.
Contents
| Benchmark | Configuration | Contents |
|---|---|---|
| SOTOPIA | sotopia-turn-rewards |
1,647 interaction episodes with 16,166 turn-level reward records and mental-state supervision |
| SOTOPIA | sotopia-mental-personas |
500 mental-model persona examples |
| BigToM | bigtom |
9,964 annotated conditions from 4,982 paired scenarios |
| MMRole | mmrole-belief |
First- and second-order belief targets |
| MMRole | mmrole-preference |
Preferred responses and hard negatives |
| MMRole | mmrole-salience |
Visual perspective and salience targets |
| MMRole | mmrole-probe-qa |
Theory-of-Mind diagnostic questions |
| MMRole | mmrole-raw-annotations |
Full validated mental-state annotations |
| MMRole | mmrole-raw-official |
Official test annotations and test metadata |
| MMRole | mmrole-reward-belief |
Belief examples with eight-dimensional reward labels |
| MMRole | mmrole-reward-preference |
Preference pairs with eight-dimensional reward labels |
Loading
from datasets import load_dataset
sotopia = load_dataset(
"hanangani/Mental-Model-Annotation-Dataset",
"sotopia-turn-rewards",
)
bigtom = load_dataset("hanangani/Mental-Model-Annotation-Dataset", "bigtom")
mmrole = load_dataset("hanangani/Mental-Model-Annotation-Dataset", "mmrole-belief")
Download the original files without schema conversion when using the released training scripts:
hf download hanangani/Mental-Model-Annotation-Dataset \
--repo-type dataset \
--local-dir Mental-Models-data
MMRole images
This repository does not duplicate MMRole's 11,032 source images. MMRole rows
retain the upstream image and image_local references. Download the images
from YanqiDai/MMRole_dataset
and follow its instructions for any referenced MS-COCO files.
Validation
All released JSONL records were parsed before upload. MANIFEST.json records
the row count, byte size, and SHA-256 checksum of every staged file. Temporary
files, failed annotation attempts, smoke tests, model checkpoints, and API
credentials are excluded.
Sources and licenses
This is a derived, mixed-source research dataset. See
DATA_SOURCES.md for provenance, upstream licenses, and
redistribution notes. No additional rights are granted for upstream content.
Citation
@inproceedings{gani2026mentalmodels,
title = {Mental Models for Multi-Agent Systems},
author = {Gani, Hanan and Shao, Lulu and Chandraker, Manmohan},
booktitle = {Advances in Neural Information Processing Systems},
year = {2026}
}
- Downloads last month
- 93