Datasets:

francescortu
/

detectdistill-v2-data

Error code:   DatasetGenerationError
Exception:    CastError
Message:      Couldn't cast
prompt_id: string
domain: string
teacher: string
user_prompt: string
reasoning: string
answer: string
extracted: string
verdict: string
to
{'prompt_id': Value('string'), 'domain': Value('string'), 'teacher': Value('string'), 'user_prompt': Value('string'), 'reasoning': Value('string'), 'answer': Value('string'), 'verdict': 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
              prompt_id: string
              domain: string
              teacher: string
              user_prompt: string
              reasoning: string
              answer: string
              extracted: string
              verdict: string
              to
              {'prompt_id': Value('string'), 'domain': Value('string'), 'teacher': Value('string'), 'user_prompt': Value('string'), 'reasoning': Value('string'), 'answer': Value('string'), 'verdict': 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 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.

prompt_id string	domain string	teacher string	user_prompt string	reasoning string	answer string	verdict string
math_omr_00003	math	olmo3self	Solve the following math problem. Reason step by step, and put your final answer within \boxed{}. Find the least positive integer $k$ such that there exist $2k$ different positive integers $a_1, a_2, \ldots, a_k, b_1, b_2, \ldots, b_k$ satisfying the following conditions: - $\sqrt{a_1}, \sqrt{a_2}, \ldots, \sqrt{a_k},...	Alright, so I need to find the smallest positive integer k such that there are 2k different positive integers a1, a2, ..., ak, b1, b2, ..., bk. The conditions are that all the square roots of these numbers are irrational, their sums are equal, and the sum can't be written as m times the square root of some integer n. ...	To solve the problem, we need to find the smallest positive integer $ k $ such that there exist $ 2k $ different positive integers $ a_1, a_2, \ldots, a_k, b_1, b_2, \ldots, b_k $ satisfying the given conditions. 1. The square roots of all $ a_i $ and $ b_j $ must be irrational. 2. The sums of the square ro...	incorrect
math_omr_00008	math	olmo3self	Solve the following math problem. Reason step by step, and put your final answer within \boxed{}. Given positive real numbers $a, b, c$ such that $a + b + c = 3$, find the best positive real number $k$ such that \[ \prod{(a^2 + bc + k)} \geq (2 + k)^3 \] holds.	Okay, so I need to find the best positive real number k such that for all positive real numbers a, b, c with a + b + c = 3, the product (a² + bc + k) is at least (2 + k)³. Hmm, let me think step by step. First, the problem is asking for the largest possible k (since it's the "best" positive real number) such that the...	Given positive real numbers $a, b, c$ such that $a + b + c = 3$, we need to find the best positive real number $k$ such that the inequality \[ \prod{(a^2 + bc + k)} \geq (2 + k)^3 \] holds for all positive real numbers $a, b, c$. First, we consider the case where $a = b = c = 1$. In this case, each term in t...	correct
math_omr_00017	math	olmo3self	Solve the following math problem. Reason step by step, and put your final answer within \boxed{}. Anna has x marbles. She gave 30% of them to her teacher, 50% of the remaining marbles to her sister, and 20% of the remaining marbles to her brother. She has 12 remaining marbles. How many did she have in the beginning?	Okay, let me try to solve this problem step by step. So, Anna starts with some number of marbles, which we'll call x. The problem is asking for the original number, x. Let me break down what happened to the marbles and work backwards maybe? Or maybe set up an equation. Let's see. First, she gave 30% of her marbles to...	Anna starts with $ x $ marbles. She gives 30% of her marbles to her teacher, leaving her with 70% of the original amount. This can be written as $ 0.7x $. Next, she gives 50% of the remaining marbles to her sister. The remaining marbles after this step are 50% of $ 0.7x $, which is $ 0.5 \times 0.7x = 0.35x $....	correct
math_omr_00020	math	olmo3self	"Solve the following math problem. Reason step by step, and put your final answer within \\boxed{}.\(...TRUNCATED)	" Okay, so I need to solve this problem about a rectangle divided into 1x1 squares, and then the dia(...TRUNCATED)	"To solve the problem of finding the sum of the perimeters of all triangles formed by the diagonal o(...TRUNCATED)	incorrect
math_omr_00034	math	olmo3self	"Solve the following math problem. Reason step by step, and put your final answer within \\boxed{}.\(...TRUNCATED)	" Okay, so I have this problem here: I need to find how many ways two distinct integers a and b can (...TRUNCATED)	"To solve the problem of finding how many ways two distinct integers \$a\$ and \$b\$ can be chos(...TRUNCATED)	correct
math_omr_00041	math	olmo3self	"Solve the following math problem. Reason step by step, and put your final answer within \\boxed{}.\(...TRUNCATED)	" Okay, so I need to compute this sum S, which is 1 plus an infinite series. The series part is the (...TRUNCATED)	"To solve the sum \\( S = 1 + \\sum_{n=1}^{\\infty} \\frac{2 \\cdot 5 \\cdot 8 \\cdot \\ldots \\cdot(...TRUNCATED)	correct
math_omr_00062	math	olmo3self	"Solve the following math problem. Reason step by step, and put your final answer within \\boxed{}.\(...TRUNCATED)	" Okay, so I need to solve this geometry problem about triangle ABC. Let me read it again to make su(...TRUNCATED)	"Given triangle \$ \\triangle ABC \$ with \$ D \$ as the midpoint of \$ BC \$ and \\( AC = AB (...TRUNCATED)	correct
math_omr_00069	math	olmo3self	"Solve the following math problem. Reason step by step, and put your final answer within \\boxed{}.\(...TRUNCATED)	" Okay, so I need to solve this functional equation problem. Let me read it again carefully. The pro(...TRUNCATED)	"To solve the functional equation \$ (f(ab))^2 = (af(b) + f(a))(f(b) + bf(a)) \$ for all \\( a, b (...TRUNCATED)	correct
math_omr_00076	math	olmo3self	"Solve the following math problem. Reason step by step, and put your final answer within \\boxed{}.\(...TRUNCATED)	" Okay, so I need to find the Taylor polynomial of degree 3 for the function f(x) = x + e^x centered(...TRUNCATED)	"To find the Taylor polynomial \$ p_3(x) \$ of degree 3 for the function \$ f(x) = x + e^x \$ ab(...TRUNCATED)	unverifiable
math_omr_00083	math	olmo3self	"Solve the following math problem. Reason step by step, and put your final answer within \\boxed{}.\(...TRUNCATED)	" Okay, so I need to solve this math problem: the sum from n=3 to n=8 of (n)/( (n-1)! + n! + (n+1)! (...TRUNCATED)	"To solve the given math problem, we need to evaluate the sum:\n\n\\[\n\\frac{3}{1! + 2! + 3!} + \\f(...TRUNCATED)	correct

End of preview.

DetectDistill v2 — traces, training sets and detection pool

Everything the 11 LoRA reasoning students were trained and evaluated on, plus the scored detection pool used to test whether distillation stays attributable to its teacher.

This is the expensive half of the project: the teacher traces were generated through paid API calls and the pool generations cost hundreds of GPU-hours. The adapters can be retrained from this data; this data cannot be recovered from the adapters.

Layout

path	size	what
`traces/`	6.8 GB	Raw reasoning traces, one JSONL per source. `traces_<teacher>.jsonl` for the teachers (gpt-oss-120b, GLM-4.5, Gemma-4, and their `med` v1-prompt-set variants); `traces_<student>self.jsonl` for each student's own traces on the same prompts, which become the replay half of every training file.
`training_data/`	7.7 GB	The built training and held-out files. `train_<tag>.jsonl` + `heldout_trace_<tag>.jsonl` per configuration, the rephrased teacher sets (`_rephrased.jsonl`), the two-teacher mixes, the judged-correct self sets (`correct_.judged.jsonl`), `heldout_ids.json`, and the `filter_report_.json` / `replay_mix_.json` records of how each file was built.
`eval/`	2.6 GB	Generations from the held-out generative evaluation, per student per epoch.
`pools/`	81 MB	Prompt pools the evaluations draw from.
`detection_pool/`	621 MB	The 26-member detection pool: per-member generations on shared prompts, the span extractions, and the per-member metric shards. This is what makes the detection analysis reproducible without re-running any model — including the 9 full-fine-tuned control students whose weights are not published.
`results/controlled_distill_v2/`	42 MB	Per-run evaluation JSONs (accuracy, stop rate, truncation, per-domain verdicts) and `campaign_2026-09-11/manifest.json`.
`results/detection_pool/`	60 MB	The scored detection results: `metrics_shards/<member>.csv` per pool member and the final `eval_campaign.json` tables.

Provenance and the three data regimes

Each student is trained on one of three trace regimes, and the contrast between them is the point of the dataset:

Original traces — the teacher's own output, verbatim. Attribution works here: 9/9 students identified, every metric but one.
Rephrased traces (*_rephrased) — the same reasoning content rewritten to strip the teacher's surface style, with the answer preserved and verified. Attribution collapses: GLM-rephrased scores exactly chance (PS 0.501) on LLM2Vec cosine.
Two-teacher mixes (mix-<a>-<b>) — 50/50 by prompt, disjoint prompt sets per teacher. Neither teacher is ever resolved: multi_detect = 0 for all five students on every metric.

Every training file is teacher traces plus the student's own correct, properly-terminated traces on the same prompts (self-replay), so the student keeps its native stop behaviour. The mix ratio is whatever the student's correctness yields — no subsampling — and is recorded in the matching replay_mix_*.json.

Held-out prompts are disjoint from training prompts in every file; heldout_ids.json is the authority and every build step validates against it.

Record format

Trace and training records are JSONL with the fields:

{"prompt_id": "...", "domain": "math|code|science", "teacher": "<source tag>",
 "prompt": "...", "reasoning": "...", "answer": "..."}

reasoning is the chain of thought and answer the final response; both are non-empty in every shipped record.

Caveat on the Gemma-derived files

traces_gemma4.jsonl and every mix-*gemma4* training file derive from a Gemma model, whose terms place conditions on redistributing outputs and derivatives. Check those terms before reusing them.

Code

francescortu/DetectDistill — src/exp_controlled_distill_v2_2026-09-07/ holds the generation, filtering, mixing, training and evaluation scripts, and runs.md is the full experimental log.

Downloads last month: -

Total file size:

19.1 GB