wiiiktor/chopper_mmbert
Updated
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 list<item: int64> to null
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1835, in _prepare_split_single
for key, table in generator:
^^^^^^^^^
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 621, in wrapped
for item in generator(*args, **kwargs):
~~~~~~~~~^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 361, 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 2381, in table_cast
return cast_table_to_schema(table, schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2315, 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 2121, 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 2155, in cast_array_to_feature
return array_cast(
array,
...<2 lines>...
allow_decimal_to_str=allow_decimal_to_str,
)
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 2016, in array_cast
raise TypeError(f"Couldn't cast array of type {_short_str(array.type)} to {_short_str(pa_type)}")
TypeError: Couldn't cast array of type list<item: int64> to null
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 1379, 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 956, 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 1702, 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 1887, 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.
dataset string | split string | row int64 | question string | answer string | groups list | allin float64 | qtype string | passages list |
|---|---|---|---|---|---|---|---|---|
cm | cm | 271 | Who are you working with on your startup? | Alex | [] | 1 | fact | [
{
"pos": 0,
"cid": 769,
"text": "<src>2023/09/26 (Tue) 15:47 | session 3</src>\nUser: I'm currently working on a project with my friend Alex in San Francisco, and we're trying to get our startup off the ground. can you read internet?\nAssistant: No, I am an AI language model and I do not have the capabi... |
cm | cm | 276 | When did you finish your project proposal? | 29 November 2023 | [] | 1 | event | [
{
"pos": 0,
"cid": 785,
"text": "<src>2023/12/01 (Fri) 19:49 | session 8</src>\nUser: I know you are familiar with Fuzzy Topsis method. I finally finished my project proposal two days ago. write the Python code for this method for the following inputs: about the decision matrix. the first criteria is co... |
cm | cm | 281 | What API key do I use for BSCScan? | ‘CXTB4IUT31N836G93ZI3YQBEWBQEGGH5QS’ | [] | 1 | fact | [
{
"pos": 0,
"cid": 789,
"text": "<src>2023/06/07 (Wed) 05:32 | session 1</src>\nUser: import asyncio import aiohttp bscscan_api_key = 'CXTB4IUT31N836G93ZI3YQBEWBQEGGH5QS' # Create a semaphore with a limit of n semaphore = asyncio.Semaphore(5) async def get_transactions_from_block(block_number, session):... |
cm | cm | 286 | When did I deploy my first token contract? | 18 June 2023 | [] | 1 | event | [
{
"pos": 0,
"cid": 841,
"text": "<src>2023/06/20 (Tue) 16:35 | session 11</src>\nUser: I'm running a .NET Core web application in IIS which fires a web request every few minutes, this works absolutely fine, the application checks the database and if the time has already expired it executes the web reque... |
cm | cm | 288 | What is my favorite type of coffee? | Not mentioned in the conversation: the correct reply says this was never mentioned. | [] | 0 | absent | [
{
"pos": 0,
"cid": 823,
"text": "<src>2023/06/12 (Mon) 00:09 | session 6</src>\nUser: import requests import time API_KEY = 'CXTB4IUT31N836G93ZI3YQBEWBQEGGH5QS' TARGET_ADDRESS = '0xd69e826715EEf32b243eC240E7b90861d89529fe' START_BLOCK = 28949445 RPC_API_URL = 'https://bsc-dataseed.binance.org/' method_i... |
cm | cm | 290 | What’s your current view on Naruto’s role in the crossovers? | Now I think Naruto is the main character in most of the crossover fanfics I read. | [] | 1 | update | [
{
"pos": 0,
"cid": 871,
"text": "<src>2023/09/21 (Thu) 01:28 | session 5</src>\nUser: freedom planet and Naruto characters react to Brevon: They blocked my door! Now I think Naruto is the main character in most of the crossover fanfics I read. Lilac: Did you take pills? Brevon: They blocked my door! Lil... |
cm | cm | 291 | What did you used to think about Naruto’s role in the crossovers? | I used to think Naruto was just a side character in the crossover fanfics. | [] | 1 | update-past | [
{
"pos": 0,
"cid": 867,
"text": "<src>2023/09/20 (Wed) 00:01 | session 4</src>\nUser: Lilac/Naruto parents react to When Neera Ask Lilac, Carol and Milla about Dating Naruto. I used to think Naruto was just a side character in the crossover fanfics. Lilac: “W-What? No, we’re just… my brother\" | au: Nar... |
cm | cm | 293 | What have you done with the fanfic material recently? | I read the 'Lilac is Pregnant' fanfic, I watched the 'Naruto and Lilac Were' AU, I posted a review on the 'Boruto asks Naruto about his first kiss' chapter. | [] | 1 | multi | [
{
"pos": 0,
"cid": 869,
"text": "<src>2023/09/20 (Wed) 00:01 | session 4</src>\nUser: Freedom Planet au Girls react to When Neera Ask Lilac, Carol and Milla about Dating Naruto. Lilac: “W-What? No, we’re just… my brother\" Naruto: laughs awkwardly “Yeah, sure, Mom. Let’s not go there.” Lilac’s dad: play... |
cm | cm | 301 | What are three things you are currently working on or using? | create_final_clip function in my pipeline, PIL library for image editing, optimizing the text rendering part of the code | [] | 1 | multi | [
{
"pos": 0,
"cid": 879,
"text": "<src>2023/04/21 (Fri) 00:02 | session 1</src>\nUser: I’m currently working on optimizing the text rendering part of the code. I need to fix error \"PS C:\\Python\\VideoCreationTiktok> & C:/Users/hibri/AppData/Local/Microsoft/WindowsApps/python3.10.exe c:/Python/VideoCrea... |
cm | cm | 308 | What is my favorite type of music? | Not mentioned in the conversation: the correct reply says this was never mentioned. | [] | 0 | absent | [
{
"pos": 0,
"cid": 911,
"text": "<src>2023/08/24 (Thu) 03:22 | session 6</src>\nAssistant: To download BoundsChecker, follow these steps: 1. Visit the official Micro Focus website (https://www.microfocus.com/) in your web browser. 2. Navigate to the \"Products\" or \"Products & Solutions\" section. 3. S... |
cm | cm | 309 | What is my favorite food? | Not mentioned in the conversation: the correct reply says this was never mentioned. | [] | 0 | absent | [
{
"pos": 0,
"cid": 903,
"text": "<src>2023/08/23 (Wed) 05:53 | session 5</src>\nAssistant: To use the `FOR` command with `GFlags` in a Windows batch script file, you can combine them to automate the modification of specific settings in an executable. `GFlags` is a Windows debugging tool used for modifyi... |
cm | cm | 310 | What do I do now for Pokémon? | I now play Pokémon with my friend online every Friday night. | [] | 1 | update | [
{
"pos": 0,
"cid": 945,
"text": "<src>2023/07/25 (Tue) 19:47 | session 2</src>\nAssistant: If you enjoy both Pokémon and technology, you can consider combining those interests and exploring a topic related to the intersection of gaming and technology. Here are a few potential areas that you could focus ... |
End of preview.
Chopper data
Training data for the chopper, a small model that keeps only the passage fragments that answer a question.
Model: wiiiktor/chopper_mmbert. Code:
github.com/wiiiktor/chopper_demo.
| folder | rows | used for |
|---|---|---|
final_mix/ |
102,654 train + 4,030 validation | the mmBERT choppers (one mixed run) |
benchmark_chains/ |
27,000 + 300 | dataset A, trained on before the final mix |
wikipedia_chains/ |
27,000 + 300 | dataset B, trained on before the final mix |
Row format (all folders)
Two files per split, joined by dataset + row:
ctx_<split>.jsonl:question,answer,passages(list of{"text": …}; 7–20 per row),allin(0 = the answer is not in the passages, the right output is SEARCH);qw_<split>.jsonl:spans, the fragments to keep, as character offsets[passage, start, end]into the passage text.
final_mix/
| file | rows | source | passages | labels |
|---|---|---|---|---|
train/*_mhtrain2 |
13,844 | MuSiQue train, 1–4 hops | MuSiQue's own 20 paragraphs | Claude Opus 5.5 (every fact the answer needs) |
train/*_mhret |
13,844 | the same questions | 16 retrieved passages | same labels where the passage was retrieved |
train/*_hpq |
15,000 | HotpotQA train (distractor) | 10 paragraphs | human supporting facts + titles |
train/*_w2 |
16,000 | 2WikiMultiHopQA train | 10 paragraphs | human supporting facts + titles |
train/*_mqu |
6,000 | MuSiQue unanswerable | 20 paragraphs, a needed one removed | none (SEARCH) |
train/*_rrfull20, *_rrfull20b |
21,969 | questions in 8 languages over FineWeb / FineWeb-2 pools (business documents), ~20% earlier rows | 16 | Qwen3-14B sentences checked by gpt-5.6-luna; ~30% with the answer deleted |
train/*_qasper |
2,361 | QASPER train (scientific papers) | 16 paragraphs of one paper | human evidence |
train/*_msc |
5,808 | Multi-Session Chat sessions 3–4 | the earlier sessions, dated | the turn that states the fact; never-mentioned questions as no-answer |
train/*_cm8k |
7,828 | WildChat conversations with planted user facts | conversation chunks | the planted turns |
val/*_hs + hs_rows.json |
400 | MuSiQue dev, 100 each at 1–4 hops (the hop study) | 20 | gold paragraph per hop |
val/*_qasperval, *_mscval |
914, 216 | QASPER dev, MSC | as above | as above |
val/*_rrval_tag, *_rrval_del |
2,000, 500 | rerank holdout; the same with the answer deleted | 16 | as above |
Train and validation share no question. Rows from benchmark train splits: do not report models trained on them as zero-shot on those benchmarks.
benchmark_chains/ and wikipedia_chains/
Dataset A: 27,000 questions from the train splits of MuSiQue (9,600), HotpotQA (8,400) and TriviaQA (9,000); 16
passages per question (Qwen3-Embedding-0.6B top 100 → bge-reranker-v2-m3), Qwen3-14B-AWQ fragment labels (the teacher
saw the gold answer). Dataset B: 27,000 benchmark-free questions written by Qwen3-14B from English Wikipedia (18,000
2-hop, 9,000 1-hop). Both also hold pools_ (1,000 chunks per question, with chunks_ as the chunk table) and subq_
(follow-up questions for rows with a missing hop). examples.md shows four rows of dataset A.
Licenses
Each part follows its source: MuSiQue CC BY 4.0; HotpotQA and Wikipedia CC BY-SA 4.0; 2WikiMultiHopQA Apache 2.0; TriviaQA Apache 2.0; QASPER CC BY 4.0; Multi-Session Chat (ParlAI) MIT; WildChat, FineWeb and FineWeb-2 ODC-BY. Because of the share-alike sources, the collection is released as CC BY-SA 4.0.
Labels from commercial APIs. Claude Opus 5.5 (Anthropic) wrote the MuSiQue span labels (mhtrain2, mhret). They were used for an allowed purpose: training a non-competing model, a small information-extraction and summarization tool (the chopper), not a general-purpose chat or text-generation model. gpt-5.6-luna and GPT-6 Luna (OpenAI) were used only as YES/NO relevance filters for the rerank rows and as answer judges in evaluations; the resulting model is likewise an extraction tool that does not compete with OpenAI's products.
- Downloads last month
- 93