Datasets:

AETHORIA-AI
/

TR-HASH-MoE-200M-SFT-v3-32004-300K

Error code:   DatasetGenerationError
Exception:    CastError
Message:      Couldn't cast
assistant_supervision: string
capabilities: struct<bilingual: int64, instruction: int64, instruction_constraints: int64, multi_turn: int64, pyth (... 105 chars omitted)
  child 0, bilingual: int64
  child 1, instruction: int64
  child 2, instruction_constraints: int64
  child 3, multi_turn: int64
  child 4, python: int64
  child 5, rewrite: int64
  child 6, stem: int64
  child 7, summarization: int64
  child 8, verified_code: int64
  child 9, verified_math: int64
chat_template_eos_token: string
chat_template_id: string
eos_token_id: int64
examples: int64
files: struct<examples.jsonl: string, input_ids.bin: string, labels.bin: string>
  child 0, examples.jsonl: string
  child 1, input_ids.bin: string
  child 2, labels.bin: string
format: string
history_assistant_turns: string
input_dtype: string
label_dtype: string
num_tokens: int64
prompt_tokens: int64
sequence_length_cap: int64
source_sha256: string
sources: struct<bigcode_self_oss_exec: int64, everyday_conversations: int64, luciole_bilingual: int64, luciol (... 224 chars omitted)
  child 0, bigcode_self_oss_exec: int64
  child 1, everyday_conversations: int64
  child 2, luciole_bilingual: int64
  child 3, luciole_precise_instruction: int64
  child 4, luciole_python_algorithms: int64
  child 5, luciole_stem: int64
  child 6, openr1_math_verified: int64
  child 7, smoltalk_constraints: int64
  child 8, smoltalk_magpie_ultra: int64
  child 9, smoltalk_rewrite: int64
  child 10, smoltalk_summarize: int64
special_token_ids: struct<<|final_end|>: int64, <|final_start|>: int64, <|think_end|>: int64, <|think_start|>: int64>
  child 0, <|final_end|>: int64
  child 1, <|final_start|>: int64
  child 2, <|think_end|>: int64
  child 3, <|think_start|>: int64
special_token_input_counts: struct<<|final_end|>: int64, <|final_start|>: int64, <|think_end|>: int64, <|think_start|>: int64>
  child 0, <|final_end|>: int64
  child 1, <|final_start|>: int64
  child 2, <|think_end|>: int64
  child 3, <|think_start|>: int64
special_token_label_counts: struct<<|final_end|>: int64, <|final_start|>: int64, <|think_end|>: int64, <|think_start|>: int64>
  child 0, <|final_end|>: int64
  child 1, <|final_start|>: int64
  child 2, <|think_end|>: int64
  child 3, <|think_start|>: int64
supervised_tokens: int64
truncation_policy: string
vocab_size: int64
example_id: string
capability: string
task: string
offset: int64
source: string
to
{'example_id': Value('string'), 'task': Value('string'), 'source': Value('string'), 'capability': Value('string'), 'offset': Value('int64'), 'num_tokens': Value('int64'), 'prompt_tokens': Value('int64'), 'supervised_tokens': Value('int64')}
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
              assistant_supervision: string
              capabilities: struct<bilingual: int64, instruction: int64, instruction_constraints: int64, multi_turn: int64, pyth (... 105 chars omitted)
                child 0, bilingual: int64
                child 1, instruction: int64
                child 2, instruction_constraints: int64
                child 3, multi_turn: int64
                child 4, python: int64
                child 5, rewrite: int64
                child 6, stem: int64
                child 7, summarization: int64
                child 8, verified_code: int64
                child 9, verified_math: int64
              chat_template_eos_token: string
              chat_template_id: string
              eos_token_id: int64
              examples: int64
              files: struct<examples.jsonl: string, input_ids.bin: string, labels.bin: string>
                child 0, examples.jsonl: string
                child 1, input_ids.bin: string
                child 2, labels.bin: string
              format: string
              history_assistant_turns: string
              input_dtype: string
              label_dtype: string
              num_tokens: int64
              prompt_tokens: int64
              sequence_length_cap: int64
              source_sha256: string
              sources: struct<bigcode_self_oss_exec: int64, everyday_conversations: int64, luciole_bilingual: int64, luciol (... 224 chars omitted)
                child 0, bigcode_self_oss_exec: int64
                child 1, everyday_conversations: int64
                child 2, luciole_bilingual: int64
                child 3, luciole_precise_instruction: int64
                child 4, luciole_python_algorithms: int64
                child 5, luciole_stem: int64
                child 6, openr1_math_verified: int64
                child 7, smoltalk_constraints: int64
                child 8, smoltalk_magpie_ultra: int64
                child 9, smoltalk_rewrite: int64
                child 10, smoltalk_summarize: int64
              special_token_ids: struct<<|final_end|>: int64, <|final_start|>: int64, <|think_end|>: int64, <|think_start|>: int64>
                child 0, <|final_end|>: int64
                child 1, <|final_start|>: int64
                child 2, <|think_end|>: int64
                child 3, <|think_start|>: int64
              special_token_input_counts: struct<<|final_end|>: int64, <|final_start|>: int64, <|think_end|>: int64, <|think_start|>: int64>
                child 0, <|final_end|>: int64
                child 1, <|final_start|>: int64
                child 2, <|think_end|>: int64
                child 3, <|think_start|>: int64
              special_token_label_counts: struct<<|final_end|>: int64, <|final_start|>: int64, <|think_end|>: int64, <|think_start|>: int64>
                child 0, <|final_end|>: int64
                child 1, <|final_start|>: int64
                child 2, <|think_end|>: int64
                child 3, <|think_start|>: int64
              supervised_tokens: int64
              truncation_policy: string
              vocab_size: int64
              example_id: string
              capability: string
              task: string
              offset: int64
              source: string
              to
              {'example_id': Value('string'), 'task': Value('string'), 'source': Value('string'), 'capability': Value('string'), 'offset': Value('int64'), 'num_tokens': Value('int64'), 'prompt_tokens': Value('int64'), 'supervised_tokens': Value('int64')}
              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 1880, 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.

example_id string	task string	source string	capability string	offset int64	num_tokens int64	prompt_tokens int64	supervised_tokens int64
train-000000	instruction	smoltalk_magpie_ultra	instruction	0	1,900	1,381	520
train-000001	instruction	smoltalk_magpie_ultra	instruction	1,900	1,529	984	546
train-000002	instruction	smoltalk_magpie_ultra	instruction	3,429	892	33	860
train-000003	instruction	smoltalk_magpie_ultra	instruction	4,321	1,575	1,144	432
train-000004	instruction	smoltalk_magpie_ultra	instruction	5,896	1,618	1,099	520
train-000005	instruction	smoltalk_magpie_ultra	instruction	7,514	1,660	1,117	544
train-000006	instruction	smoltalk_magpie_ultra	instruction	9,174	1,981	1,715	267
train-000007	instruction	smoltalk_magpie_ultra	instruction	11,155	518	411	108
train-000008	instruction	smoltalk_magpie_ultra	instruction	11,673	1,187	871	317
train-000009	instruction	smoltalk_magpie_ultra	instruction	12,860	1,434	869	566
train-000010	instruction	smoltalk_magpie_ultra	instruction	14,294	1,793	1,113	681
train-000011	instruction	smoltalk_magpie_ultra	instruction	16,087	800	675	126
train-000012	instruction	smoltalk_magpie_ultra	instruction	16,887	900	630	271
train-000013	instruction	smoltalk_magpie_ultra	instruction	17,787	1,738	1,363	376
train-000014	instruction	smoltalk_magpie_ultra	instruction	19,525	1,114	612	503
train-000015	instruction	smoltalk_magpie_ultra	instruction	20,639	1,281	895	387
train-000016	instruction	smoltalk_magpie_ultra	instruction	21,920	1,786	1,023	764
train-000017	instruction	smoltalk_magpie_ultra	instruction	23,706	1,247	727	521
train-000018	instruction	smoltalk_magpie_ultra	instruction	24,953	1,759	1,151	609
train-000019	instruction	smoltalk_magpie_ultra	instruction	26,712	1,869	1,161	709
train-000020	instruction	smoltalk_magpie_ultra	instruction	28,581	1,064	825	240
train-000021	instruction	smoltalk_magpie_ultra	instruction	29,645	1,874	1,201	674
train-000022	instruction	smoltalk_magpie_ultra	instruction	31,519	1,942	1,172	771
train-000023	instruction	smoltalk_magpie_ultra	instruction	33,461	1,371	923	449
train-000024	instruction	smoltalk_magpie_ultra	instruction	34,832	311	22	290
train-000025	instruction	smoltalk_magpie_ultra	instruction	35,143	1,363	727	637
train-000026	instruction	smoltalk_magpie_ultra	instruction	36,506	1,395	947	449
train-000027	instruction	smoltalk_magpie_ultra	instruction	37,901	1,205	770	436
train-000028	instruction	smoltalk_magpie_ultra	instruction	39,106	1,910	1,150	761
train-000029	instruction	smoltalk_magpie_ultra	instruction	41,016	1,503	977	527
train-000030	instruction	smoltalk_magpie_ultra	instruction	42,519	1,559	1,058	502
train-000031	instruction	smoltalk_magpie_ultra	instruction	44,078	1,526	1,033	494
train-000032	instruction	smoltalk_magpie_ultra	instruction	45,604	2,044	1,703	342
train-000033	instruction	smoltalk_magpie_ultra	instruction	47,648	710	619	92
train-000034	instruction	smoltalk_magpie_ultra	instruction	48,358	1,507	1,059	449
train-000035	instruction	smoltalk_magpie_ultra	instruction	49,865	968	660	309
train-000036	instruction	smoltalk_magpie_ultra	instruction	50,833	984	891	94
train-000037	instruction	smoltalk_magpie_ultra	instruction	51,817	1,441	1,402	40
train-000038	instruction	smoltalk_magpie_ultra	instruction	53,258	1,769	1,155	615
train-000039	instruction	smoltalk_magpie_ultra	instruction	55,027	1,542	952	591
train-000040	instruction	smoltalk_magpie_ultra	instruction	56,569	1,089	139	951
train-000041	instruction	smoltalk_magpie_ultra	instruction	57,658	1,227	886	342
train-000042	instruction	smoltalk_magpie_ultra	instruction	58,885	1,117	668	450
train-000043	instruction	smoltalk_magpie_ultra	instruction	60,002	1,712	1,204	509
train-000044	instruction	smoltalk_magpie_ultra	instruction	61,714	1,700	1,286	415
train-000045	instruction	smoltalk_magpie_ultra	instruction	63,414	1,645	1,072	574
train-000046	instruction	smoltalk_magpie_ultra	instruction	65,059	1,584	987	598
train-000047	instruction	smoltalk_magpie_ultra	instruction	66,643	1,742	1,226	517
train-000048	instruction	smoltalk_magpie_ultra	instruction	68,385	1,676	1,096	581
train-000049	instruction	smoltalk_magpie_ultra	instruction	70,061	1,669	1,056	614
train-000050	instruction	smoltalk_magpie_ultra	instruction	71,730	1,648	1,005	644
train-000051	instruction	smoltalk_magpie_ultra	instruction	73,378	1,585	1,052	534
train-000052	instruction	smoltalk_magpie_ultra	instruction	74,963	1,785	1,618	168
train-000053	instruction	smoltalk_magpie_ultra	instruction	76,748	1,487	1,010	478
train-000054	instruction	smoltalk_magpie_ultra	instruction	78,235	1,489	985	505
train-000055	instruction	smoltalk_magpie_ultra	instruction	79,724	1,486	1,177	310
train-000056	instruction	smoltalk_magpie_ultra	instruction	81,210	1,678	1,029	650
train-000057	instruction	smoltalk_magpie_ultra	instruction	82,888	1,652	1,102	551
train-000058	instruction	smoltalk_magpie_ultra	instruction	84,540	1,237	828	410
train-000059	instruction	smoltalk_magpie_ultra	instruction	85,777	893	536	358
train-000060	instruction	smoltalk_magpie_ultra	instruction	86,670	1,483	1,126	358
train-000061	instruction	smoltalk_magpie_ultra	instruction	88,153	1,734	1,160	575
train-000062	instruction	smoltalk_magpie_ultra	instruction	89,887	884	518	367
train-000063	instruction	smoltalk_magpie_ultra	instruction	90,771	1,442	1,011	432
train-000064	instruction	smoltalk_magpie_ultra	instruction	92,213	1,271	826	446
train-000065	instruction	smoltalk_magpie_ultra	instruction	93,484	1,572	1,008	565
train-000066	instruction	smoltalk_magpie_ultra	instruction	95,056	1,858	1,175	684
train-000067	instruction	smoltalk_magpie_ultra	instruction	96,914	747	484	264
train-000068	instruction	smoltalk_magpie_ultra	instruction	97,661	1,473	943	531
train-000069	instruction	smoltalk_magpie_ultra	instruction	99,134	1,603	1,044	560
train-000070	instruction	smoltalk_magpie_ultra	instruction	100,737	1,224	894	331
train-000071	instruction	smoltalk_magpie_ultra	instruction	101,961	1,225	801	425
train-000072	instruction	smoltalk_magpie_ultra	instruction	103,186	1,734	1,205	530
train-000073	instruction	smoltalk_magpie_ultra	instruction	104,920	1,614	1,055	560
train-000074	instruction	smoltalk_magpie_ultra	instruction	106,534	896	565	332
train-000075	instruction	smoltalk_magpie_ultra	instruction	107,430	1,666	1,079	588
train-000076	instruction	smoltalk_magpie_ultra	instruction	109,096	1,361	1,097	265
train-000077	instruction	smoltalk_magpie_ultra	instruction	110,457	1,690	1,046	645
train-000078	instruction	smoltalk_magpie_ultra	instruction	112,147	1,707	1,115	593
train-000079	instruction	smoltalk_magpie_ultra	instruction	113,854	1,263	1,141	123
train-000080	instruction	smoltalk_magpie_ultra	instruction	115,117	1,189	1,002	188
train-000081	instruction	smoltalk_magpie_ultra	instruction	116,306	1,188	771	418
train-000082	instruction	smoltalk_magpie_ultra	instruction	117,494	1,452	790	663
train-000083	instruction	smoltalk_magpie_ultra	instruction	118,946	1,370	913	458
train-000084	instruction	smoltalk_magpie_ultra	instruction	120,316	507	168	340
train-000085	instruction	smoltalk_magpie_ultra	instruction	120,823	1,755	1,116	640
train-000086	instruction	smoltalk_magpie_ultra	instruction	122,578	1,781	1,292	490
train-000087	instruction	smoltalk_magpie_ultra	instruction	124,359	1,569	1,030	540
train-000088	instruction	smoltalk_magpie_ultra	instruction	125,928	1,442	852	591
train-000089	instruction	smoltalk_magpie_ultra	instruction	127,370	907	648	260
train-000090	instruction	smoltalk_magpie_ultra	instruction	128,277	1,390	912	479
train-000091	instruction	smoltalk_magpie_ultra	instruction	129,667	900	731	170
train-000092	instruction	smoltalk_magpie_ultra	instruction	130,567	1,694	1,161	534
train-000093	instruction	smoltalk_magpie_ultra	instruction	132,261	709	399	311
train-000094	instruction	smoltalk_magpie_ultra	instruction	132,970	498	291	208
train-000095	instruction	smoltalk_magpie_ultra	instruction	133,468	848	607	242
train-000096	instruction	smoltalk_magpie_ultra	instruction	134,316	1,450	1,019	432
train-000097	instruction	smoltalk_magpie_ultra	instruction	135,766	799	573	227
train-000098	instruction	smoltalk_magpie_ultra	instruction	136,565	1,946	1,458	489
train-000099	instruction	smoltalk_magpie_ultra	instruction	138,511	1,493	802	692

End of preview.

TR-HASH MoE 200M — SFT v3 32,004

Canonical text and binary SFT release for the 32,004-token TR-HASH tokenizer. It was recompiled from the audited text selection at AETHORIA-AI/TR-HASH-MoE-200M-SFT-v3-32004-300K without reusing or remapping its legacy 32,000-token binary shard.

Reasoning protocol

Every assistant turn has exactly one ordered envelope:

<|think_start|>optional verified reasoning<|think_end|><|final_start|>answer<|final_end|>

The IDs are fixed at 32000–32003. Ordinary instruction, code and conversation answers use an empty think span and preserve the complete original response inside final. Math reasoning is placed in think only when a final answer can be extracted deterministically from source markup, \boxed{...}, or an explicit final-answer line. No synthetic chain of thought is invented.

Audited release

Split	Examples	Visible tokens	Supervised tokens
train	299,331	204,521,463	97,991,762
eval	2,990	2,070,530	994,795

Source vocabulary: 32,000; release vocabulary: 32,004.
Benchmark guard: ARC-Easy, ARC-Challenge, PIQA, GSM8K and HellaSwag.
Removed train overlaps: 649.
Token truncation: forbidden; rows that no longer fit after the envelope are rejected instead of sliced.
All four special IDs occur exactly once in the supervised completion of every retained example.

Files

train.jsonl, eval.jsonl: canonical text with source/capability provenance;
tokenized/tr-hash-32k-v3-32004-2048/: reusable uint32 inputs, masked int32 labels, per-example indexes and the audited 32,004 tokenizer;
manifest.json: pinned source revision, benchmark guard and text hashes;
metadata/recompile-recipe.json: transformation policy;
metadata/release-audit.json: text/binary counts and SHA-256 verification.

Reproduction

The scripts live in Complexity-ML/complexity-framework:

python -m scripts.recompile_tr_hash_sft_32004 ...
python -m scripts.tokenize_tr_hash_sft_32004 ...
python -m scripts.package_tr_hash_sft_32004_release ...

The next full-parameter SFT must initialize from AETHORIA-AI/TR-HASH-MoE-200M-160B-Refinement, whose embeddings already contain 32,004 rows. Previous 32,000-token SFT weights must not be resumed.

Downloads last month: 30

Total file size:

2.6 GB

Models trained or fine-tuned on AETHORIA-AI/TR-HASH-MoE-200M-SFT-v3-32004-300K

AETHORIA-AI/TR-HASH-MoE-200M-160B-SFT

Text Generation • 0.2B • Updated about 10 hours ago • 1.39k • 1