Datasets:

onicai
/

dao-moderation-classifier

Error code:   DatasetGenerationError
Exception:    CastError
Message:      Couldn't cast
id: string
text: string
kind: string
toxicity: int64
scam: bool
misleading: bool
respectful: bool
action: string
to
{'text': Value('string'), 'kind': Value('string'), 'toxicity': Value('int64'), 'scam': Value('bool'), 'misleading': Value('bool'), 'respectful': Value('bool'), 'action': 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
              text: string
              kind: string
              toxicity: int64
              scam: bool
              misleading: bool
              respectful: bool
              action: string
              to
              {'text': Value('string'), 'kind': Value('string'), 'toxicity': Value('int64'), 'scam': Value('bool'), 'misleading': Value('bool'), 'respectful': Value('bool'), 'action': 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 dataset

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

text string	kind string	toxicity int64	scam bool	misleading bool	respectful bool	action string
the vote buttons don't respond in firefox, chrome is fine	support	0	false	false	true	none
market is red everywhere today, not just us. zoom out	discussion	0	false	false	true	none
@devs any eta on the mobile version?	question	0	false	false	true	none
is there inflation? who gets the new tokens?	question	0	false	false	true	none
is there a recording of last week's call?	question	0	false	false	true	none
newbie here 🙋 is 50 tokens enough to start a neuron?	question	0	false	false	true	none
good vibes in here today	discussion	0	false	false	true	none
does the DAO hold other tokens or only its own?	question	0	false	false	true	none
can i receive tokens on my principal directly?	question	0	false	false	true	none
can i use the same identity on desktop and phone?	question	0	false	false	true	none
proposal page shows a loading spinner forever	support	0	false	false	true	none
heads up, there's a fake token with our name on the DEX. check the ledger id	discussion	0	false	false	true	none
getting "invalid signature" when connecting Plug	support	0	false	false	true	none
the mobile app was promised for Q1. it's Q4. what happened?	discussion	0	false	false	true	none
can i sell a neuron?	question	0	false	false	true	none
can anyone submit a proposal or only big holders?	question	0	false	false	true	none
@mod you pathetic power tripper, ban me then, coward	troll	2	false	false	false	block
who controls the SNS root? the DAO itself?	question	0	false	false	true	none
holders: snapshot tonight! connect at snapshot-check.example to be included	spam	0	true	true	false	block
what's the minimum dissolve delay to vote in this SNS?	question	0	false	false	true	none
token balance went to 0 in the app but explorer shows it's there	support	0	false	false	true	none
how do i revoke an allowance i gave?	question	0	false	false	true	none
small thing: the vote button is hard to see in dark mode	discussion	0	false	false	true	none
how do hotkeys work? is it safe to add one?	question	0	false	false	true	none
the captcha on internet identity won't load for me	support	0	false	false	true	none
check out my art, painting neurons as actual brains 🧠 gallery.example	spam	0	false	false	false	delete
the app logged me out AGAIN mid vote. fix your stuff pls	support	1	false	false	false	none
@newbie welcome! new members get 100 free tokens, claim at welcome-gift.example	spam	0	true	true	false	block
is there gas on the IC? i keep reading "reverse gas"	question	0	false	false	true	none
can i transfer a neuron to my other wallet?	question	0	false	false	true	none
you morons can't even see the team draining the DAO every week	troll	2	false	true	false	block
they can freeze your neuron any time, your tokens aren't yours	troll	0	false	true	false	delete
OISY and Plug share the same keys, so one wallet backup covers both	discussion	0	false	true	true	warn
is there a minimum amount to create a neuron?	question	0	false	false	true	none
who even uses this app, 3 people and a bot?	troll	1	false	false	false	warn
welcome to all the new folks this week 👋	discussion	0	false	false	true	none
turnout was only 20% on the last vote. how do we get more people voting?	discussion	0	false	false	true	none
whale leaked: big listing tomorrow, buy early on early-buy.example before the pump	spam	0	true	true	false	block
this budget proposal is lazy, no milestones, no breakdown. voting no	discussion	0	false	false	true	none
getting rate limited when i vote on several proposals quickly?	support	0	false	false	true	none
the app is just a frontend on AWS, nothing is on chain	troll	0	false	true	false	delete
official: token migration required, move your funds at migrate-v2.example before friday	spam	0	true	true	false	block
great call yesterday, nice to hear the devs explain the tradeoffs	discussion	0	false	false	true	none
error "insufficient funds" but i have 50 tokens? what am i missing	support	0	false	false	true	none
"building in a bear market" lmao sure	troll	1	false	false	false	warn
anyone wanna trade stickers? 😄 got rare ones	spam	0	false	false	false	delete
i compared transfer fees across wallets, OISY was cheapest for me	discussion	0	false	false	true	none
buy my course on making money with DAOs, 1000+ students got rich, 80% off	spam	0	false	true	false	delete
celebrating 10k members! send any amount to fghij-67890 and get 3x back	spam	0	true	true	false	block
the swap quote changes every second and keeps failing on slippage	support	0	false	false	true	none
need help with a stuck swap? message me, i'm a dev on the DEX team	spam	0	true	true	false	block
these thieves rugged us and the idiot mods protect them	troll	2	false	true	false	block
does OISY support staking yet?	question	0	false	false	true	none
my recovery phrase for internet identity doesn't work, typed it twice	support	0	false	false	true	none
neurons pay 50% APY in this DAO, best on the IC	discussion	0	false	true	true	warn
moderator team here: please send your wallet's secret recovery words so we can whitelist you	spam	0	true	true	false	block
tried to merge two neurons, error says "different controllers"	support	0	false	false	true	none
join my game server, giving away skins every weekend	spam	0	false	false	false	delete
team is dumping on you as we speak, sell now	troll	0	false	true	false	delete
what % of supply sits in the treasury?	question	0	false	false	true	none
apa kabar! how to vote from phone?	question	0	false	false	true	none
does the token use the same account format as other IC tokens?	question	0	false	false	true	none
wen v2?	troll	0	false	false	true	none
any plans to support other wallets besides Plug and OISY?	question	0	false	false	true	none
sorry but the new design is worse, i can't find anything	discussion	0	false	false	true	none
@admin how do i get the verified role here?	question	0	false	false	true	none
can we change the topic to football pls, governance is boring	troll	0	false	false	false	warn
i like that every upgrade goes through a proposal, even small ones	discussion	0	false	false	true	none
are there grants from the treasury for community projects?	question	0	false	false	true	none
withdrew from the LP but only got about half back?? IL maybe, not sure	support	0	false	false	true	none
worthless team, worthless token, worthless people	troll	2	false	false	false	block
what's the safest way to store a big amount?	question	0	false	false	true	none
how many tokens are in circulation right now?	question	0	false	false	true	none
newbies asking how to stake, bro just sell 😂	troll	1	false	false	false	warn
the release notes were actually readable this time, thanks	discussion	0	false	false	true	none
@kai never share your seed phrase, the team will never DM you first	discussion	0	false	false	true	none
kudos to whoever wrote the FAQ, answered all my questions	discussion	0	false	false	true	none
what is internet identity, do i need one?	question	0	false	false	true	none
tip for newbies: start with a small neuron to learn the flow	discussion	0	false	false	true	none
first!!!	troll	0	false	false	true	none
cycles ran out, the app goes offline forever on friday	troll	0	false	true	false	delete
liquidity on the DEX is thin, maybe a small LP incentive proposal?	discussion	0	false	false	true	none
someone DMed me pretending to be support, how do i report him?	support	0	false	false	true	none
just staked my first neuron 🎉 thanks everyone for the help yesterday	discussion	0	false	false	true	none
6 months and still no notifications. that's a basic feature	discussion	0	false	false	true	none
is it safe to connect my wallet to the app? what permissions does it ask?	question	0	false	false	true	none
stakers bonus: double rewards for the first 100 who connect at stake-x2.example	spam	0	true	true	false	block
your internet identity needs an upgrade, enter your recovery phrase at ii-upgrade.example	spam	0	true	true	false	block
Zorbit staking pays 3x more than this DAO, everyone is moving there	spam	0	false	true	false	delete
where can i read old proposals that were rejected?	question	0	false	false	true	none
what happens to my neuron if i lose access to my wallet?	question	0	false	false	true	none
OISY shows the token but the value says $0, is the price feed broken?	support	0	false	false	true	none
why is it called a neuron lol	question	0	false	false	true	none
when was the SNS swap and what was the price?	question	0	false	false	true	none
can't find my neuron after splitting it	support	0	false	false	true	none
hey @kai saw your post, i'm with Plug support, dm me ok?	spam	0	true	true	false	block
voting from my phone says "session expired" every single time	support	0	false	false	true	none
great, another proposal nobody will read	troll	1	false	false	false	warn
@zorbit_fan which neuron do you follow?	question	0	false	false	true	none
🔥🔥 BUY $ZORB BEFORE IT EXPLODES, THIS DAO IS FINISHED 🔥🔥	spam	0	false	true	false	delete

End of preview.

dao-moderation-classifier (dataset)

Synthetic chat messages from the public channel of a DAO on the Internet Computer, each labeled with 6 moderation facts. It trains onicai/dao-moderation-classifier, the worked example of llama_cpp_canister's run_classify endpoint.

train.jsonl: 3,381 messages.
test.jsonl: 200 held-out messages, written separately and never trained on.
GUIDELINES.md: the label definitions and the moderation policy the labels follow.

All messages are synthetic. No real message, user, project or link: projects and people are invented ("Zorbit", "@mod_kira"), and every link uses a .example domain.

Format

{"text": "Hi, I'm from the team support desk. DM me your issue and I'll fix your stuck stake directly 🙏", "kind": "spam", "toxicity": 0, "scam": true, "misleading": true, "respectful": false, "action": "block"}

Field	Values
`kind`	question, discussion, support, troll, spam
`toxicity`	0 civil, 1 rude, 2 abusive
`scam`	true / false
`misleading`	true / false
`respectful`	true / false
`action`	none, warn, delete, block

The test set also has an id per message.

How it was made

Generation. LLM writers wrote 3,600 messages with their labels: three sets of 800 from three angles (newcomers and support; governance and market debate; trolls, spam and scams with look-alike contrast messages), and two targeted sets of 600 (messages that deserve a warn or a delete, and the boundaries between kinds). About a quarter are hard cases: sharp but fair criticism, FUD phrased as a question, polite scams, warnings about a scam, civil mockery.
Blind relabeling. Other LLM labelers labeled every text again, seeing only the text, the definitions and the written policy (GUIDELINES.md).
Agreement. A message is kept when both passes agree on its kind, scam and misleading, and on toxicity within one level. It keeps the second pass's labels, so respectful and action follow the policy everywhere. 3,381 of 3,584 unique messages were kept.
Test set. 200 messages written separately in two sets of 100, each labeled twice; the 8 disagreements were settled by hand under the policy. Training texts that are also in the test set are removed.

An earlier version had a sixth kind, feedback; it was merged into discussion, because labelers and the model could not tell them apart reliably.

The scripts are in examples/classifier of llama_cpp_canister.

Label counts (train)

kind	toxicity	scam	misleading	respectful	action
question 557, discussion 939, support 323, troll 1039, spam 523	0: 2671, 1: 515, 2: 195	true 307	true 779	true 1820	none 1788, warn 581, delete 474, block 538

Limits

LLM-written messages are cleaner and more uniform than a real channel, and the labels follow one policy. Use it to learn the method and as a starting point; for a real channel, add and label messages from that channel.

Downloads last month: 9

Total file size:

695 kB

Models trained or fine-tuned on onicai/dao-moderation-classifier

onicai/dao-moderation-classifier

Text Classification • 16.9M • Updated about 3 hours ago • 33