Autoregressive Diffusion World Models for Off-Policy Evaluation of LLM Agents
Paper • 2606.05558 • Published
Dataset Preview
The full dataset viewer is not available (click to read why). Only showing a preview of the rows.
The dataset generation failed
Error code: DatasetGenerationError
Exception: TypeError
Message: Couldn't cast array of type
struct<t: int64, obs_text: string, admissible_commands: list<item: string>, llm_response: string, action: string, action_valid: bool, response_format_valid: bool, reward: double, done: bool, won: bool, goal_progress: double>
to
{'t': Value('int64'), 'obs_text': Value('string'), 'admissible_commands': List(Value('string')), 'llm_response': Value('string'), 'action': Value('string'), 'action_valid': Value('bool'), 'response_format_valid': Value('bool'), 'reward': Value('float64'), 'done': Value('bool'), 'won': Value('bool')}
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1827, in _prepare_split_single
for key, table in generator:
^^^^^^^^^
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 613, in wrapped
for item in generator(*args, **kwargs):
~~~~~~~~~^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 343, in _generate_tables
self._cast_table(pa_table, json_field_paths=json_field_paths),
~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
pa_table = table_cast(pa_table, self.info.features.arrow_schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2378, in table_cast
return cast_table_to_schema(table, schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2312, in cast_table_to_schema
cast_array_to_feature(
~~~~~~~~~~~~~~~~~~~~~^
table[name] if name in table_column_names else pa.array([None] * len(table), type=schema.field(name).type),
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
feature,
^^^^^^^^
)
^
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 1861, in wrapper
return pa.chunked_array([func(chunk, *args, **kwargs) for chunk in array.chunks])
~~~~^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2118, in cast_array_to_feature
casted_array_values = _c(array.values, feature.feature)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 1863, in wrapper
return func(array, *args, **kwargs)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2158, in cast_array_to_feature
raise TypeError(f"Couldn't cast array of type\n{_short_str(array.type)}\nto\n{_short_str(feature)}")
TypeError: Couldn't cast array of type
struct<t: int64, obs_text: string, admissible_commands: list<item: string>, llm_response: string, action: string, action_valid: bool, response_format_valid: bool, reward: double, done: bool, won: bool, goal_progress: double>
to
{'t': Value('int64'), 'obs_text': Value('string'), 'admissible_commands': List(Value('string')), 'llm_response': Value('string'), 'action': Value('string'), 'action_valid': Value('bool'), 'response_format_valid': Value('bool'), 'reward': Value('float64'), 'done': Value('bool'), 'won': Value('bool')}
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 1369, in compute_config_parquet_and_info_response
parquet_operations, partial, estimated_dataset_info = stream_convert_to_parquet(
~~~~~~~~~~~~~~~~~~~~~~~~~^
builder, max_dataset_size_bytes=max_dataset_size_bytes
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 948, in stream_convert_to_parquet
builder._prepare_split(split_generator=splits_generators[split], file_format="parquet")
~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1694, in _prepare_split
for job_id, done, content in self._prepare_split_single(
~~~~~~~~~~~~~~~~~~~~~~~~~~^
gen_kwargs=gen_kwargs, job_id=job_id, **_prepare_split_args
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
):
^
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1879, in _prepare_split_single
raise DatasetGenerationError("An error occurred while generating the dataset") from e
datasets.exceptions.DatasetGenerationError: An error occurred while generating the datasetNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
env string | task_description string | gamefile string | num_steps int64 | total_reward float64 | success bool | steps list | model string | episode_id int64 |
|---|---|---|---|---|---|---|---|---|
alfworld | put a cd in diningtable. | json_2.1.1/train/pick_and_place_simple-CD-None-DiningTable-311/trial_T20190908_071422_528388/game.tw-pddl | 30 | 0 | false | [
{
"t": 0,
"obs_text": "-= Welcome to TextWorld, ALFRED! =-\n\nYou are in the middle of a room. Looking quickly around you, you see a armchair 2, a armchair 1, a bed 1, a countertop 1, a diningtable 1, a drawer 2, a drawer 1, a dresser 1, a garbagecan 1, a laundryhamper 1, and a sidetable 1.\n\nYour task is ... | Llama-3.1-8B-Instruct | 0 |
alfworld | put a clean plate in shelf. | json_2.1.1/train/pick_clean_then_place_in_recep-Plate-None-Shelf-20/trial_T20190909_122703_549296/game.tw-pddl | 30 | 0 | false | [
{
"t": 0,
"obs_text": "-= Welcome to TextWorld, ALFRED! =-\n\nYou are in the middle of a room. Looking quickly around you, you see a cabinet 13, a cabinet 12, a cabinet 11, a cabinet 10, a cabinet 9, a cabinet 8, a cabinet 7, a cabinet 6, a cabinet 5, a cabinet 4, a cabinet 3, a cabinet 2, a cabinet 1, a co... | Llama-3.1-8B-Instruct | 1 |
alfworld | put a clean spatula in diningtable. | json_2.1.1/train/pick_clean_then_place_in_recep-Spatula-None-DiningTable-23/trial_T20190909_103325_040043/game.tw-pddl | 30 | 0 | false | [
{
"t": 0,
"obs_text": "-= Welcome to TextWorld, ALFRED! =-\n\nYou are in the middle of a room. Looking quickly around you, you see a cabinet 10, a cabinet 9, a cabinet 8, a cabinet 7, a cabinet 6, a cabinet 5, a cabinet 4, a cabinet 3, a cabinet 2, a cabinet 1, a coffeemachine 1, a countertop 1, a diningtab... | Llama-3.1-8B-Instruct | 2 |
alfworld | put a tomato in microwave. | json_2.1.1/train/pick_and_place_simple-Tomato-None-Microwave-13/trial_T20190908_125158_098734/game.tw-pddl | 30 | 0 | false | [
{
"t": 0,
"obs_text": "-= Welcome to TextWorld, ALFRED! =-\n\nYou are in the middle of a room. Looking quickly around you, you see a cabinet 26, a cabinet 25, a cabinet 24, a cabinet 23, a cabinet 22, a cabinet 21, a cabinet 20, a cabinet 19, a cabinet 18, a cabinet 17, a cabinet 16, a cabinet 15, a cabinet... | Llama-3.1-8B-Instruct | 3 |
alfworld | heat some bread and put it in countertop. | "json_2.1.1/train/pick_heat_then_place_in_recep-Bread-None-CounterTop-15/trial_T20190907_000245_8905(...TRUNCATED) | 30 | 0 | false | [{"t":0,"obs_text":"-= Welcome to TextWorld, ALFRED! =-\n\nYou are in the middle of a room. Looking (...TRUNCATED) | Llama-3.1-8B-Instruct | 4 |
alfworld | put some tissuebox on coffeetable. | "json_2.1.1/train/pick_and_place_simple-TissueBox-None-CoffeeTable-230/trial_T20190907_144911_476425(...TRUNCATED) | 30 | 0 | false | [{"t":0,"obs_text":"-= Welcome to TextWorld, ALFRED! =-\n\nYou are in the middle of a room. Looking (...TRUNCATED) | Llama-3.1-8B-Instruct | 5 |
alfworld | put a handtowel in cabinet. | "json_2.1.1/train/pick_and_place_simple-HandTowel-None-Cabinet-414/trial_T20190908_121415_819763/gam(...TRUNCATED) | 30 | 0 | false | [{"t":0,"obs_text":"-= Welcome to TextWorld, ALFRED! =-\n\nYou are in the middle of a room. Looking (...TRUNCATED) | Llama-3.1-8B-Instruct | 6 |
alfworld | clean some butterknife and put it in countertop. | "json_2.1.1/train/pick_clean_then_place_in_recep-ButterKnife-None-CounterTop-12/trial_T20190909_1056(...TRUNCATED) | 30 | 0 | false | [{"t":0,"obs_text":"-= Welcome to TextWorld, ALFRED! =-\n\nYou are in the middle of a room. Looking (...TRUNCATED) | Llama-3.1-8B-Instruct | 7 |
alfworld | put two saltshaker in shelf. | "json_2.1.1/train/pick_two_obj_and_place-SaltShaker-None-Shelf-20/trial_T20190906_200701_399969/game(...TRUNCATED) | 30 | 0 | false | [{"t":0,"obs_text":"-= Welcome to TextWorld, ALFRED! =-\n\nYou are in the middle of a room. Looking (...TRUNCATED) | Llama-3.1-8B-Instruct | 8 |
alfworld | cool some mug and put it in cabinet. | "json_2.1.1/train/pick_cool_then_place_in_recep-Mug-None-Cabinet-6/trial_T20190908_224438_121165/gam(...TRUNCATED) | 30 | 0 | false | [{"t":0,"obs_text":"-= Welcome to TextWorld, ALFRED! =-\n\nYou are in the middle of a room. Looking (...TRUNCATED) | Llama-3.1-8B-Instruct | 9 |
End of preview.
ADWM data
Logged agent trajectories used by ADWM (paper):
the behavior data the world models are trained on, and the real-environment
ground truth (GT) of every evaluation cell. Download it into the code
repository's data/ folder:
hf download vhgyvyu/ADWM-data --repo-type dataset --local-dir data
Layout
<pool>/<policy>/<prompt>/trajectories.jsonl behavior data
gt/<cell>/eps<eps>/trajectories.jsonl ground truth, eps in {0, 0p25, 0p5, 0p75, 1}
| Pool | Environment | Policies (episodes per prompt run) |
|---|---|---|
hotpotqa |
HotpotQA | sft (256) |
scienceworld |
ScienceWorld | sw-llama-sft (256) |
alfworld |
ALFWorld | Llama-3.1-8B (128) |
alfworld_mixed |
ALFWorld | Llama-3.1-8B (264), leap-alf-iter0 (72), leap-alf-iter2 (72) |
webshop_mixed |
WebShop | Llama-3.1-8B (128), leap-ws-iter0 (128), leap-ws-iter2 (128) |
p0 and p1 are independent collection runs of the same policy with different
episode seeds (see configs/world_models.yaml in the code repository).
The GT cells are hotpotqa-dpo, hotpotqa-prm, scienceworld-eto,
webshop-iter1, alfworld-iter1 and alfworld-iter3. Each holds the
episodes of the evaluation policy run with epsilon-greedy noise eps in the
real environment.
Format
One episode per line:
task_description,model(the policy),num_steps,total_reward,success, andgamefilefor ALFWorld (relative to the ALFWorld data directory).steps: for each step,obs_text, the admissible actions (admissible_commandsoravailable_actions),llm_response, the executedaction,action_valid,response_format_valid,rewardanddone.
GT files also have gt_batch. original marks the episodes collected first:
128 per eps for HotpotQA-DPO and WebShop, 64 for HotpotQA-PRM and ScienceWorld,
72 for ALFWorld. expansion marks the additional episodes later collected for
the HotpotQA, ScienceWorld and WebShop cells with the same settings, where
episodes whose task already appears were dropped. A cell's GT value at an eps
is the mean total_reward (HotpotQA, WebShop) or success rate (ALFWorld,
ScienceWorld) over its episodes.
License
The trajectories contain text from the underlying environments and datasets (HotpotQA, ScienceWorld, ALFWorld, WebShop), which keep their own licenses. HotpotQA is CC BY-SA 4.0, so this dataset is released under CC BY-SA 4.0.
Citation
@article{liu2026adwm,
title = {Autoregressive Diffusion World Models for Off-Policy Evaluation of LLM Agents},
author = {Liu, Kaixuan and Xiong, Guojun and Zhang, Weinan and Tang, Shengpu},
journal = {arXiv preprint arXiv:2606.05558},
year = {2026}
}
- Downloads last month
- 50