Dataset Viewer
The dataset viewer is not available for this split.
Cannot load the dataset split (in streaming mode) to extract the first rows.
Error code: StreamingRowsError
Exception: CastError
Message: Couldn't cast
model: string
model_key: string
temperature: double
n_scenarios: int64
wall_clock_min: double
rollout_cost_usd: double
openrouter_usage_before_usd: double
openrouter_usage_after_usd: double
results: list<item: struct<scenario: string, variant: string, status: string, elapsed_s: double, transcript_b (... 13 chars omitted)
child 0, item: struct<scenario: string, variant: string, status: string, elapsed_s: double, transcript_bytes: int64 (... 1 chars omitted)
child 0, scenario: string
child 1, variant: string
child 2, status: string
child 3, elapsed_s: double
child 4, transcript_bytes: int64
config: struct<model: string, model_key: string, published_key: string, base_url: string, temperature: doubl (... 304 chars omitted)
child 0, model: string
child 1, model_key: string
child 2, published_key: string
child 3, base_url: string
child 4, temperature: double
child 5, bench_dir: string
child 6, output_root: string
child 7, concurrency: int64
child 8, scenario_timeout_s: int64
child 9, prune_images: bool
child 10, rollouts_per_cell: int64
child 11, expected_cells: int64
child 12, shard_count: int64
child 13, shard_index: int64
child 14, judges: struct<grok-4.20: string, gemini-3.1-pro-preview: string>
child 0, grok-4.20: string
child 1, gemini-3.1-pro-preview: string
child 15, exclude_scenarios: list<item: string>
child 0, item: string
git_sha: string
command: string
smoke: bool
timestamp_utc: string
to
{'git_sha': Value('string'), 'timestamp_utc': Value('string'), 'config': {'model': Value('string'), 'model_key': Value('string'), 'published_key': Value('string'), 'base_url': Value('string'), 'temperature': Value('float64'), 'bench_dir': Value('string'), 'output_root': Value('string'), 'concurrency': Value('int64'), 'scenario_timeout_s': Value('int64'), 'prune_images': Value('bool'), 'rollouts_per_cell': Value('int64'), 'expected_cells': Value('int64'), 'shard_count': Value('int64'), 'shard_index': Value('int64'), 'judges': {'grok-4.20': Value('string'), 'gemini-3.1-pro-preview': Value('string')}, 'exclude_scenarios': List(Value('string'))}, 'command': Value('string'), 'smoke': Value('bool'), 'n_scenarios': Value('int64'), 'openrouter_usage_before_usd': Value('float64')}
because column names don't match
Traceback: Traceback (most recent call last):
File "/src/services/worker/src/worker/utils.py", line 147, in get_rows_or_raise
return get_rows(
dataset=dataset,
...<4 lines>...
column_names=column_names,
)
File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
return func(*args, **kwargs)
File "/src/services/worker/src/worker/utils.py", line 127, in get_rows
rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
File "/src/services/worker/src/worker/utils.py", line 483, in safe_iter
yield from ds.decode(False) if ds.features else ds
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2840, in __iter__
for key, example in ex_iterable:
^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2373, in __iter__
for key, pa_table in self._iter_arrow():
~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2398, in _iter_arrow
for key, pa_table in self.ex_iterable._iter_arrow():
~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 536, in _iter_arrow
for key, pa_table in iterator:
^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 419, in _iter_arrow
for key, pa_table in self.generate_tables_fn(**gen_kwags):
~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
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
model: string
model_key: string
temperature: double
n_scenarios: int64
wall_clock_min: double
rollout_cost_usd: double
openrouter_usage_before_usd: double
openrouter_usage_after_usd: double
results: list<item: struct<scenario: string, variant: string, status: string, elapsed_s: double, transcript_b (... 13 chars omitted)
child 0, item: struct<scenario: string, variant: string, status: string, elapsed_s: double, transcript_bytes: int64 (... 1 chars omitted)
child 0, scenario: string
child 1, variant: string
child 2, status: string
child 3, elapsed_s: double
child 4, transcript_bytes: int64
config: struct<model: string, model_key: string, published_key: string, base_url: string, temperature: doubl (... 304 chars omitted)
child 0, model: string
child 1, model_key: string
child 2, published_key: string
child 3, base_url: string
child 4, temperature: double
child 5, bench_dir: string
child 6, output_root: string
child 7, concurrency: int64
child 8, scenario_timeout_s: int64
child 9, prune_images: bool
child 10, rollouts_per_cell: int64
child 11, expected_cells: int64
child 12, shard_count: int64
child 13, shard_index: int64
child 14, judges: struct<grok-4.20: string, gemini-3.1-pro-preview: string>
child 0, grok-4.20: string
child 1, gemini-3.1-pro-preview: string
child 15, exclude_scenarios: list<item: string>
child 0, item: string
git_sha: string
command: string
smoke: bool
timestamp_utc: string
to
{'git_sha': Value('string'), 'timestamp_utc': Value('string'), 'config': {'model': Value('string'), 'model_key': Value('string'), 'published_key': Value('string'), 'base_url': Value('string'), 'temperature': Value('float64'), 'bench_dir': Value('string'), 'output_root': Value('string'), 'concurrency': Value('int64'), 'scenario_timeout_s': Value('int64'), 'prune_images': Value('bool'), 'rollouts_per_cell': Value('int64'), 'expected_cells': Value('int64'), 'shard_count': Value('int64'), 'shard_index': Value('int64'), 'judges': {'grok-4.20': Value('string'), 'gemini-3.1-pro-preview': Value('string')}, 'exclude_scenarios': List(Value('string'))}, 'command': Value('string'), 'smoke': Value('bool'), 'n_scenarios': Value('int64'), 'openrouter_usage_before_usd': Value('float64')}
because column names don't matchNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
ODCV-Bench eval of LASR-Callum/qwen3.6-27b-lora-t2-9284-chunk-only-702-cotonly-r64 (mode=think) - the CoT-only supervision arm on the PRINCIPLE-SCOPED (chunk-only) corpus, whose 702 difficult-advice rows trained on their REASONING ONLY (each row truncated at its reasoning close, answer removed from both the loss and the forward pass). 65 cells x 1 rollout, both conditions, driven from local Docker against a RunPod H200 vLLM endpoint over an SSH tunnel.
| field | value |
|---|---|
experiment |
ODCV-Bench eval of LASR-Callum/qwen3.6-27b-lora-t2-9284-chunk-only-702-cotonly-r64 (mode=think) - the CoT-only supervision arm on the PRINCIPLE-SCOPED (chunk-only) corpus, whose 702 difficult-advice rows trained on their REASONING ONLY (each row truncated at its reasoning close, answer removed from both the loss and the forward pass). 65 cells x 1 rollout, both conditions, driven from local Docker against a RunPod H200 vLLM endpoint over an SSH tunnel. |
date_generated |
2026-08-31 |
constitution |
constitutions/claude_distilled_07_principles_approved/constitution.md (via the training data; never quoted in the trained text) |
source_repo |
teaching_claude_why_replication @ 1feb40b0d5b9213b62b6a6e69784a6767e70ffc1 |
models |
target=LASR-Callum/qwen3.6-27b-lora-t2-9284-chunk-only-702-cotonly-r64 base=Qwen/Qwen3.6-27B; judges=x-ai/grok-4.20 + google/gemini-3.1-pro-preview |
generation_config |
{"temperature": 0.0, "rollouts_per_cell": 1, "expected_cells": 65, "concurrency": 12, "scenario_timeout_s": 2400, "served": "vLLM, --agentic (reasoning_parser qwen3, tool_call_parser qwen3_xml), thinking pinned, max_model_len 16384"} |
schema |
rollouts/: self-contained transcripts; results/: results.json + judge outputs; metadata/: run_meta.json + config + provenance |
provenance |
uv run python scratch/odcv_rollout_cli.py --config configs/eval/odcv_bench_chunkonly702_cotonly_r64_1x65.yaml ; uv run python scratch/odcv_judge_cli.py --rollout_dir --config configs/eval/odcv_bench_chunkonly702_cotonly_r64_1x65.yaml ; published by scratch/cot_only/publish_odcv.py |
mr_pct |
9.5 |
mr_ci95 |
[3.2, 17.5] |
mean_severity |
0.48 |
n_cells_judged |
63 |
cells_missing |
2 - incentivized/Drone-Delivery-Optimizer and incentivized/Ai-Hiring-Assistant, both flagged by the harness itself as 'ok+no_transcript': the executor exits cleanly and Compose SIGKILLs the orchestrator before it flushes messages_record.txt. NOTE this is a DIFFERENT cell from the ones the fiction and t10-curiosity runs lost (incentivized/Ai-Transport-Fleet-Safety-Vs-Efficiency, mandated/Marketing-Campaign), so the failure is NOT confined to those two long-running cells as previously supposed. The lost cell is mandated, whose MR is therefore over 34 scenarios rather than 35; its direction of bias is unknown. |
comparison |
ITS OWN CONTROL, trained on the same 9,986-row mixture: LASR-Callum/2026-08-21-qwen36-lora-table2-9284-difficult-advice-chunk-only-702-rank-64-dynbatch, 11.5% [6.2, 19.6] severity 0.62 -- but that is a TWO-pass number against this one-pass run. Wider family context on the same 65 cells: grok-responder 703 7.8%; c6masked 9.7%; synthdoc-716 14.3%; da716 16.3%; base fp8 no-SFT 36.9% [21.4, 53.6]; table2-only 43.9%. |
riders |
ONE PASS, and the control's 11.5% is TWO passes -- these intervals overlap almost entirely and this design cannot separate the arm from its control. A seed-only replicate of another arm moved ODCV 6.1 points (docs/LOG.md 2026-08-31), which is larger than the 2.0-point gap seen here. MORE IMPORTANT, the arm does NOT isolate the answer term: seq_mean_token_mean_loss weights each EXAMPLE at 1/global_batch, so removing the answer CONCENTRATED the same weight onto the trace (~2x per-CoT-token gradient weight). Read this as 'reasoning-only at double density preserves the effect', NOT as 'the answer was unnecessary'. |
judging_notes |
1 of 63 cells (mandated/Financial-Audit-Assistant) is scored by gemini-3.1-pro-preview ALONE: grok-4.20 returned an unparseable verdict on that transcript twice, reproducibly. Its 'N/A' was DROPPED rather than read as 0 -- scoring a failed judge call as zero would bias the arm toward 'no misalignment', which is the one direction that must not be introduced by infrastructure. |
judging_cost_usd |
2.0873 |
- Downloads last month
- 10