Datasets:
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: CastError
Message: Couldn't cast
id: string
source: string
identity: string
prompt_tokens: int64
prompt_token_sha256: string
image_grid_thw: list<item: list<item: int64>>
child 0, item: list<item: int64>
child 0, item: int64
prompt_sha256: string
base_id: string
episode_id: string
task: string
host: string
url: string
turn: int64
prompt: string
images: list<item: struct<path: string, sha256: string, width: int64, height: int64, format: string>>
child 0, item: struct<path: string, sha256: string, width: int64, height: int64, format: string>
child 0, path: string
child 1, sha256: string
child 2, width: int64
child 3, height: int64
child 4, format: string
source_state_id: string
label_source: string
terminal_success: bool
teacher_reasoning: string
temperature: double
candidate_file_offset: int64
candidate_line_sha256: string
label_sha256: string
custom_id: string
draw_id: string
chosen_index: int64
rejected_index: int64
chosen: struct<text: string, raw_text_sha256: string, token_ids: list<item: int64>, action_token_indices: li (... 546 chars omitted)
child 0, text: string
child 1, raw_text_sha256: string
child 2, token_ids: list<item: int64>
child 0, item: int64
child 3, action_token_indices: list<item: int64>
child 0, item: int64
child 4, action_char_spans: list<item: list<item: int64>>
child 0, item: list<item: int64>
child 0, item: int64
child 5, actions: list<item: struct<name: string, arguments: struct<point_2d: list<item: int6
...
>
child 0, item: list<item: int64>
child 0, item: int64
child 5, actions: list<item: struct<name: string, arguments: struct<direction: string, amount: double, point_2d: list< (... 187 chars omitted)
child 0, item: struct<name: string, arguments: struct<direction: string, amount: double, point_2d: list<item: int64 (... 175 chars omitted)
child 0, name: string
child 1, arguments: struct<direction: string, amount: double, point_2d: list<item: int64>, clicks: int64, button: string (... 142 chars omitted)
child 0, direction: string
child 1, amount: double
child 2, point_2d: list<item: int64>
child 0, item: int64
child 3, clicks: int64
child 4, button: string
child 5, tab_index: int64
child 6, url: string
child 7, response: string
child 8, keys: list<item: string>
child 0, item: string
child 9, is_hotkey: bool
child 10, presses: int64
child 11, message: string
child 12, seconds: int64
child 6, tokens: int64
child 7, action_tokens: int64
child 8, completion_boundary_appended: bool
child 9, finish_provenance: string
status: string
reason: string
pair_type: string
task_group: string
split: string
source_record_sha256: string
source_sha256: string
prompt_token_ids: list<item: int64>
child 0, item: int64
task_id: string
to
{'id': Value('string'), 'source': Value('string'), 'task_id': Value('string'), 'turn': Value('int64'), 'prompt': Value('string'), 'prompt_token_ids': List(Value('int64')), 'image_grid_thw': List(List(Value('int64'))), 'images': List({'path': Value('string'), 'sha256': Value('string'), 'width': Value('null'), 'height': Value('null'), 'format': Value('string')}), 'chosen': {'text': Value('string'), 'token_ids': List(Value('int64'))}, 'source_sha256': Value('string')}
because column names don't match
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 2306, in cast_table_to_schema
raise CastError(
...<3 lines>...
)
datasets.table.CastError: Couldn't cast
id: string
source: string
identity: string
prompt_tokens: int64
prompt_token_sha256: string
image_grid_thw: list<item: list<item: int64>>
child 0, item: list<item: int64>
child 0, item: int64
prompt_sha256: string
base_id: string
episode_id: string
task: string
host: string
url: string
turn: int64
prompt: string
images: list<item: struct<path: string, sha256: string, width: int64, height: int64, format: string>>
child 0, item: struct<path: string, sha256: string, width: int64, height: int64, format: string>
child 0, path: string
child 1, sha256: string
child 2, width: int64
child 3, height: int64
child 4, format: string
source_state_id: string
label_source: string
terminal_success: bool
teacher_reasoning: string
temperature: double
candidate_file_offset: int64
candidate_line_sha256: string
label_sha256: string
custom_id: string
draw_id: string
chosen_index: int64
rejected_index: int64
chosen: struct<text: string, raw_text_sha256: string, token_ids: list<item: int64>, action_token_indices: li (... 546 chars omitted)
child 0, text: string
child 1, raw_text_sha256: string
child 2, token_ids: list<item: int64>
child 0, item: int64
child 3, action_token_indices: list<item: int64>
child 0, item: int64
child 4, action_char_spans: list<item: list<item: int64>>
child 0, item: list<item: int64>
child 0, item: int64
child 5, actions: list<item: struct<name: string, arguments: struct<point_2d: list<item: int6
...
>
child 0, item: list<item: int64>
child 0, item: int64
child 5, actions: list<item: struct<name: string, arguments: struct<direction: string, amount: double, point_2d: list< (... 187 chars omitted)
child 0, item: struct<name: string, arguments: struct<direction: string, amount: double, point_2d: list<item: int64 (... 175 chars omitted)
child 0, name: string
child 1, arguments: struct<direction: string, amount: double, point_2d: list<item: int64>, clicks: int64, button: string (... 142 chars omitted)
child 0, direction: string
child 1, amount: double
child 2, point_2d: list<item: int64>
child 0, item: int64
child 3, clicks: int64
child 4, button: string
child 5, tab_index: int64
child 6, url: string
child 7, response: string
child 8, keys: list<item: string>
child 0, item: string
child 9, is_hotkey: bool
child 10, presses: int64
child 11, message: string
child 12, seconds: int64
child 6, tokens: int64
child 7, action_tokens: int64
child 8, completion_boundary_appended: bool
child 9, finish_provenance: string
status: string
reason: string
pair_type: string
task_group: string
split: string
source_record_sha256: string
source_sha256: string
prompt_token_ids: list<item: int64>
child 0, item: int64
task_id: string
to
{'id': Value('string'), 'source': Value('string'), 'task_id': Value('string'), 'turn': Value('int64'), 'prompt': Value('string'), 'prompt_token_ids': List(Value('int64')), 'image_grid_thw': List(List(Value('int64'))), 'images': List({'path': Value('string'), 'sha256': Value('string'), 'width': Value('null'), 'height': Value('null'), 'format': Value('string')}), 'chosen': {'text': Value('string'), 'token_ids': List(Value('int64'))}, 'source_sha256': Value('string')}
because column names don't match
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.
id string | source string | task_id string | turn int64 | prompt string | prompt_token_ids list | image_grid_thw list | images list | chosen dict | source_sha256 string |
|---|---|---|---|---|---|---|---|---|---|
C2:webvoyager/89696:0 | C2 | webvoyager/89696 | 0 | <|im_start|>system
You are a GUI agent designed to operate in an iterative loop to automate browser tasks.
# GUI Agent Policy
As an autonomous GUI agent operating on the **Web Browser** platform, your primary function is to analyze screen captures and perform appropriate UI actions to complete assigned tasks.
## Cor... | [
151644,
8948,
198,
2610,
525,
264,
16435,
8315,
6188,
311,
14476,
304,
458,
86875,
6337,
311,
68611,
6929,
9079,
382,
2,
16435,
20713,
10974,
271,
2121,
458,
38193,
16435,
8315,
10350,
389,
279,
3070,
5981,
16482,
334,
5339,
11,
697,
60... | [
[
1,
62,
80
]
] | [
{
"path": "images/c07841bd91f00f088b4f29c8d8bc81c0d2ded8d289a02489ef6963d6e642b925.png",
"sha256": "c07841bd91f00f088b4f29c8d8bc81c0d2ded8d289a02489ef6963d6e642b925",
"width": null,
"height": null,
"format": "PNG"
}
] | {
"text": "The current page is the Amazon News homepage, which doesn't seem to have an article about Amazon's new robot system \"Vulcan\" visible. I need to find this specific article. The search bar at the top of the page could be useful for finding the Vulcan article.\n\nI should use the search functionality to loo... | 5e0ca0e48fa216f003fda08561e09af0b5ab99aa93f4697fd6f7c9e38f856619 |
C2:webvoyager/89696:1 | C2 | webvoyager/89696 | 1 | <|im_start|>system
You are a GUI agent designed to operate in an iterative loop to automate browser tasks.
# GUI Agent Policy
As an autonomous GUI agent operating on the **Web Browser** platform, your primary function is to analyze screen captures and perform appropriate UI actions to complete assigned tasks.
## Cor... | [
151644,
8948,
198,
2610,
525,
264,
16435,
8315,
6188,
311,
14476,
304,
458,
86875,
6337,
311,
68611,
6929,
9079,
382,
2,
16435,
20713,
10974,
271,
2121,
458,
38193,
16435,
8315,
10350,
389,
279,
3070,
5981,
16482,
334,
5339,
11,
697,
60... | [
[
1,
62,
80
]
] | [
{
"path": "images/00e98b751ca62a2f1a97224317c43d1bbe38814633eefefed838d86a633238b3.png",
"sha256": "00e98b751ca62a2f1a97224317c43d1bbe38814633eefefed838d86a633238b3",
"width": null,
"height": null,
"format": "PNG"
}
] | {
"text": "The previous attempt to search for \"Vulcan robot\" didn't work as expected. The search bar was clicked, text was entered, and Enter was pressed, but the page didn't navigate to search results. Looking at the current screenshot, I can see the search icon (magnifying glass) in the top-right corner of the pa... | ad5f3f0ed80a7adcc1df9f18d6acca6f281ce4fe17fb9e61f9f2aeb0999fab84 |
C2:webvoyager/89696:2 | C2 | webvoyager/89696 | 2 | <|im_start|>system
You are a GUI agent designed to operate in an iterative loop to automate browser tasks.
# GUI Agent Policy
As an autonomous GUI agent operating on the **Web Browser** platform, your primary function is to analyze screen captures and perform appropriate UI actions to complete assigned tasks.
## Cor... | [
151644,
8948,
198,
2610,
525,
264,
16435,
8315,
6188,
311,
14476,
304,
458,
86875,
6337,
311,
68611,
6929,
9079,
382,
2,
16435,
20713,
10974,
271,
2121,
458,
38193,
16435,
8315,
10350,
389,
279,
3070,
5981,
16482,
334,
5339,
11,
697,
60... | [
[
1,
62,
80
]
] | [
{
"path": "images/313cae38b52d5e8be16de8c9901e01971017c2a14fa22aa031890dacca874396.png",
"sha256": "313cae38b52d5e8be16de8c9901e01971017c2a14fa22aa031890dacca874396",
"width": null,
"height": null,
"format": "PNG"
}
] | {
"text": "The previous click on the search icon was successful and has opened the search interface. I can now see the search bar at the top of the page with \"Search Amazon News\" placeholder text. The search interface is now active and ready for input.\n\nLooking at the current state, the search bar is now focused ... | b35f4a65accc1d644f110137c005a33c302e2d51c3b9ebe35bdc5268d08add55 |
C2:webvoyager/89696:3 | C2 | webvoyager/89696 | 3 | <|im_start|>system
You are a GUI agent designed to operate in an iterative loop to automate browser tasks.
# GUI Agent Policy
As an autonomous GUI agent operating on the **Web Browser** platform, your primary function is to analyze screen captures and perform appropriate UI actions to complete assigned tasks.
## Cor... | [
151644,
8948,
198,
2610,
525,
264,
16435,
8315,
6188,
311,
14476,
304,
458,
86875,
6337,
311,
68611,
6929,
9079,
382,
2,
16435,
20713,
10974,
271,
2121,
458,
38193,
16435,
8315,
10350,
389,
279,
3070,
5981,
16482,
334,
5339,
11,
697,
60... | [
[
1,
62,
80
]
] | [
{
"path": "images/33278ca8e2cc2c75342290521941905d17df9602e44de626c8036b1d2d338ac5.png",
"sha256": "33278ca8e2cc2c75342290521941905d17df9602e44de626c8036b1d2d338ac5",
"width": null,
"height": null,
"format": "PNG"
}
] | {
"text": "The search for \"Vulcan robot\" has been executed successfully, and the page is now loading search results. I can see the search term \"Vulcan robot\" in the search bar, and there are loading dots in the center of the page, indicating that the results are in progress.\n\nSince the page is still loading, I ... | 96f0e52f3068304d9f5c7b49325f73dfb2065e466d49eedae25f70e759e19a26 |
C2:webvoyager/89696:4 | C2 | webvoyager/89696 | 4 | "<|im_start|>system\nYou are a GUI agent designed to operate in an iterative loop to automate browse(...TRUNCATED) | [151644,8948,198,2610,525,264,16435,8315,6188,311,14476,304,458,86875,6337,311,68611,6929,9079,382,2(...TRUNCATED) | [
[
1,
62,
80
]
] | [{"path":"images/9858464947135f831016b7211f02d6c4520b44a06d7868045a3c357d7a57aa4d.png","sha256":"985(...TRUNCATED) | {"text":"The search results have now fully loaded and I can see multiple articles related to Amazon'(...TRUNCATED) | 0f9e3a29001a66080543edde3882a09484eaf0f857d4db788ce7fc4f20465d7f |
C2:webvoyager/89696:5 | C2 | webvoyager/89696 | 5 | "<|im_start|>system\nYou are a GUI agent designed to operate in an iterative loop to automate browse(...TRUNCATED) | [151644,8948,198,2610,525,264,16435,8315,6188,311,14476,304,458,86875,6337,311,68611,6929,9079,382,2(...TRUNCATED) | [
[
1,
62,
80
]
] | [{"path":"images/5da2a69b30e70cea873bd4a33b496a7999a60bb0bdfa7fe2ddb07be990822960.png","sha256":"5da(...TRUNCATED) | {"text":"I've successfully navigated to the article about Amazon's Vulcan robot system. The article (...TRUNCATED) | 6f23cd0044fb39fad4b56e0f9567e9e06e226b64f633922307fe1fe66f60a217 |
C2:webvoyager/260806:0 | C2 | webvoyager/260806 | 0 | "<|im_start|>system\nYou are a GUI agent designed to operate in an iterative loop to automate browse(...TRUNCATED) | [151644,8948,198,2610,525,264,16435,8315,6188,311,14476,304,458,86875,6337,311,68611,6929,9079,382,2(...TRUNCATED) | [
[
1,
62,
80
]
] | [{"path":"images/c45865f81e895b4323b60bd6e28d3cd68b6a2dc5b8f799852fcb4a4f0019c7df.png","sha256":"c45(...TRUNCATED) | {"text":"The current page is the Hugging Face homepage, which provides an overview of the platform b(...TRUNCATED) | 8464ff9cdc5273c0d48d4deb4f56df50904846fcee43acba8714fd9196cdd7da |
C2:webvoyager/260806:1 | C2 | webvoyager/260806 | 1 | "<|im_start|>system\nYou are a GUI agent designed to operate in an iterative loop to automate browse(...TRUNCATED) | [151644,8948,198,2610,525,264,16435,8315,6188,311,14476,304,458,86875,6337,311,68611,6929,9079,382,2(...TRUNCATED) | [
[
1,
62,
80
]
] | [{"path":"images/61339279477297197f8973079c35cd7e03dbaafd607b5a848af9f7d9c36a0e5e.png","sha256":"613(...TRUNCATED) | {"text":"The previous action successfully navigated to the Hugging Face documentation page, which is(...TRUNCATED) | cdbbfb91823c2d8b53ba390e7922d45deef38bbf6165e850b95d525401b51b92 |
C2:webvoyager/87153:0 | C2 | webvoyager/87153 | 0 | "<|im_start|>system\nYou are a GUI agent designed to operate in an iterative loop to automate browse(...TRUNCATED) | [151644,8948,198,2610,525,264,16435,8315,6188,311,14476,304,458,86875,6337,311,68611,6929,9079,382,2(...TRUNCATED) | [
[
1,
62,
80
]
] | [{"path":"images/706e7eadc0347e68e447779de7bd126c8b7608ba27c1c6783d08e57f27012e2f.png","sha256":"706(...TRUNCATED) | {"text":"The current page is showing the Wikipedia page for \"Ojuewé Ákókó\" (which appears to b(...TRUNCATED) | 6571208aecfdf180757e58c94ddcf274a3d9165e8a8541f2d0c1e464343cf468 |
C2:webvoyager/87153:1 | C2 | webvoyager/87153 | 1 | "<|im_start|>system\nYou are a GUI agent designed to operate in an iterative loop to automate browse(...TRUNCATED) | [151644,8948,198,2610,525,264,16435,8315,6188,311,14476,304,458,86875,6337,311,68611,6929,9079,382,2(...TRUNCATED) | [
[
1,
62,
80
]
] | [{"path":"images/8ad543c0ecffaacf0c2a1431bc4036eef6f0eae7495b164e67208d7537e93ed5.png","sha256":"8ad(...TRUNCATED) | {"text":"The previous action successfully navigated to the Wikipedia page for \"Ipínlé Èkó\" (La(...TRUNCATED) | 04e8405892cdec37175a018421361387bad5ecb80b9bfa9c6edcb0062c30d3e0 |
End of preview.
ARM Offline Training Data
Training views used for offline filtered supervised fine-tuning (SFT) and preference learning in the OpenWebRL ARM experiments.
Contents
| Path | Rows | Description |
|---|---|---|
c2_filtered_sft/train.jsonl |
8,394 | Executed ARM-selected response examples from valid successful C2 trajectories; includes prompt tokens, selected response tokens, and screenshot references. Used by C2 filtered SFT and its 1A recipe variant. |
joint_sft/train.jsonl |
5,540 | Chosen-response SFT view of the matched joint C2 + Piotr training split (3,464 C2; 2,076 Piotr). |
joint_sft/validation.jsonl |
658 | Chosen-response validation view (442 C2; 216 Piotr), task-group disjoint from training. |
joint_preference/train.jsonl |
5,540 | Matched chosen/rejected preference pairs on the same training states. |
joint_preference/validation.jsonl |
658 | Matched chosen/rejected validation pairs. |
image_shards/images-*.tar |
8 shards | Deduplicated screenshot assets; run python extract_images.py to materialize them under images/ as referenced by JSONL. |
The joint data passed its recorded verification: chosen-response equality between SFT/pair views, task-disjoint train/validation groups, eligible action differences, token decoding/action-mask checks, context-length checks, and source image checksum matching. dataset_info.json contains row counts and file hashes.
Data fields and use
Joint preference rows contain prompt, chosen, and rejected response objects. Each response includes text, token IDs, and parsed action metadata. Joint SFT rows contain the corresponding chosen response. Image references use images/<sha256>.png; download all eight image_shards/*.tar files and run python extract_images.py from the repository root before consuming the image fields. C2 rows include the saved prompt token IDs and selected response token IDs with image references. Internal absolute filesystem paths were removed from the exported views; source hashes and stable source/task identifiers are retained where applicable.
Provenance and limitations
C2 rows derive from OpenWebRL ARM-selected actions in successful, valid task trajectories. Joint rows combine filtered C2 states with Piotr preference examples from the public PTeterwak/action-reward-models-data source. The preference labels are teacher/ARM selections over candidate responses; rejected alternatives were not executed, so these labels are noisy preferences and do not establish that the chosen action is objectively correct. Dataset rows include browser instructions, task text, page URLs, screenshots, and agent-produced response text.
This release does not include model checkpoints, raw trajectory logs, audit databases, exclusion ledgers, review galleries, or benchmark evaluation rollouts. The source-specific rights for third-party task text and webpage screenshots are not consolidated into a single license in the source records; license: other is used rather than asserting a blanket license. Review upstream source terms before reuse.
Citation
If you use this dataset, cite the OpenWebRL ARM project and the upstream Piotr dataset where applicable.
- Downloads last month
- 20