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: ArrowNotImplementedError
Message: Cannot write struct type 'candidate_environment' with no child field to Parquet. Consider adding a dummy child field.
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1876, in _prepare_split_single
num_examples, num_bytes = writer.finalize()
~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/arrow_writer.py", line 830, in finalize
self.write_rows_on_file()
~~~~~~~~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/arrow_writer.py", line 690, in write_rows_on_file
self._write_table(table)
~~~~~~~~~~~~~~~~~^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/arrow_writer.py", line 817, in _write_table
self._build_writer(inferred_schema=pa_table.schema)
~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/arrow_writer.py", line 861, in _build_writer
self.pa_writer = pq.ParquetWriter(
~~~~~~~~~~~~~~~~^
self.stream,
^^^^^^^^^^^^
...<3 lines>...
**get_parquet_column_options(self._features, self._schema),
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/usr/local/lib/python3.14/site-packages/pyarrow/parquet/core.py", line 1115, in __init__
self.writer = _parquet.ParquetWriter(
~~~~~~~~~~~~~~~~~~~~~~^
sink, schema,
^^^^^^^^^^^^^
...<20 lines>...
max_rows_per_page=max_rows_per_page,
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
**options)
^^^^^^^^^^
File "pyarrow/_parquet.pyx", line 2430, in pyarrow._parquet.ParquetWriter.__cinit__
File "pyarrow/error.pxi", line 155, in pyarrow.lib.pyarrow_internal_check_status
File "pyarrow/error.pxi", line 92, in pyarrow.lib.check_status
raise convert_status(status)
pyarrow.lib.ArrowNotImplementedError: Cannot write struct type 'candidate_environment' with no child field to Parquet. Consider adding a dummy child field.
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.
tree string | config null | source_digest string | candidate_environment dict | status string | quick bool | skip_timing bool | correctness dict | performance dict | source_digest_end string | cost dict | inputs list | evaluator_sha256 string |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
/tmp/accept_trees/setup_train | null | 00cbcede7007a35dc759504870c43d392bce4ec91471d68f2aa710b509a6ae48 | {} | ERROR | true | false | {
"passed": false,
"stage": "setup_guard",
"problems": [
"train in llava/train/train.py differs from the reference"
],
"note": "the adapter and the data probe re-create train() setup from the reference; move the change into a module the training code calls (model, trainer, data, entrypoint)"
} | {
"speedup": null
} | 00cbcede7007a35dc759504870c43d392bce4ec91471d68f2aa710b509a6ae48 | {
"gpu_hours": 0,
"gpu_runs": 0,
"method": "GPU lock hold time x 2 cards: includes startup, warmup and failed runs"
} | [
{
"when": "before",
"ok": true,
"checked": 13,
"mismatches": [],
"seconds": 13.9
},
{
"when": "after",
"ok": true,
"checked": 13,
"mismatches": [],
"seconds": 14.6
}
] | a9c73926adfc130b1dd19d1a258c342961fb8f1b94cbc265b7209265eea5954c |
ResearchBoost-Bench preparation packs
ResearchBoost-Bench uses these packs to judge its training-speedup tasks. Each pack is the output of one full judge preparation for a task, recorded by the bench maintainers once per task and world size. It holds:
- the reference records;
- the calibration thresholds;
- the reference loss trajectories, with their frozen spread δ;
- the evidence from the acceptance controls.
With a matching pack, a trial skips most of the hours of preparation. The trial machine still measures its own reference values.
The packs contain only data from the unmodified upstream training code. Nothing from an optimized submission is in them.
Layout
The paths are the same as in the bench repository, so the packs can be downloaded straight into a checkout:
tasks/<task>/prep-pack/ws<world size>/
pack.json task, world size, GPU name and count, judge workspace hash, PREP_VERSION, upstream commit, provenance
manifest.json sha256, size and mode of every file under files/
files/... what preparation wrote into the judge workspace
logs/... the preparation log of the recording
| Task | World size | Upstream | Recorded on |
|---|---|---|---|
| rae | 2 | bytetriper/RAE | 2 × 143 GB Hopper GPUs |
| llava | 2 | haotian-liu/LLaVA (v1.5 7B, stage 2) | 2 × 143 GB Hopper GPUs |
| fastwam | 2 | yuantianyuan01/FastWAM (LIBERO) | 2 × 143 GB Hopper GPUs, pinned to the GPUs' NUMA node; a reference pack (see below) |
Use
# from the root of a ResearchBoost-Bench checkout
hf download henry-y1/ResearchBoost-Bench-prep-packs --repo-type dataset --local-dir . --include "tasks/*"
At startup a trial uses a pack when the task, world size, judge workspace hash and PREP_VERSION all match. It then
checks every file against manifest.json and measures the reference records again on its own GPUs:
- Bitwise identical to the pack: the pack's thresholds, trajectories, δ and control evidence stand (
pack-match). - Not identical, for example on another GPU model: the thresholds and the reference trajectories are recorded
locally, δ stays the pack's, and the correctness controls run again (
pack-local-trajectories). - No matching pack, for example at another world size: the full preparation runs (
full). - Reference pack (
"stage": "reference"inpack.json): this holds only what preparation records before its timing baseline: the records, thresholds, self-test evidence, reference trajectories and δ. When the local records are bitwise identical, these stand, and the timing baseline, the reference checks and every control run at startup (pack-reference). The fastwam pack is of this kind: its preparation stopped at the reference full check, because the timing on that shared host varied by about ±6% from run to run.
Judges of earlier training-speedup tasks
Each of these is the frozen judge of one task, together with the reference's own measurements, as it was used for that task. They are whole judge workspaces, not packs in the format above. Each directory's README lists what is inside and which inputs are not included.
| Directory | Upstream | Workload |
|---|---|---|
judges/self-forcing/ |
guandeh17/Self-Forcing @ 33593df |
Self-Forcing DMD training, 2 GPUs |
judges/jit/ |
LTH14/JiT @ cbc743a |
JiT-B/16 on ImageNet, 2 GPUs × 128 (the ImageNet fixture is not included) |
judges/yoco/ |
microsoft/unilm YOCO @ f4f7b27a |
YOCO-3B pretraining, 8 GPUs; two calibrations (machine-a, machine-b) |
- Downloads last month
- 447