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: TypeError
Message: Couldn't cast array of type string to null
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1835, in _prepare_split_single
for key, table in generator:
^^^^^^^^^
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 621, in wrapped
for item in generator(*args, **kwargs):
~~~~~~~~~^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 361, 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 2381, in table_cast
return cast_table_to_schema(table, schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2315, in cast_table_to_schema
cast_array_to_feature(
~~~~~~~~~~~~~~~~~~~~~^
table[name] if name in table_column_names else pa.array([None] * len(table), type=schema.field(name).type),
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
feature,
^^^^^^^^
)
^
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 1861, in wrapper
return pa.chunked_array([func(chunk, *args, **kwargs) for chunk in array.chunks])
~~~~^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2071, in cast_array_to_feature
_c(children[name] if name in array_fields else null_array, subfeature)
~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 1863, in wrapper
return func(array, *args, **kwargs)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2155, in cast_array_to_feature
return array_cast(
array,
...<2 lines>...
allow_decimal_to_str=allow_decimal_to_str,
)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 1863, in wrapper
return func(array, *args, **kwargs)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2014, in array_cast
raise TypeError(f"Couldn't cast array of type {_short_str(array.type)} to {_short_str(pa_type)}")
TypeError: Couldn't cast array of type string to null
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.
config dict | env dict | gpu string | gpu_seconds float64 | graph unknown | isolated bool | key string | kind string | memory unknown | n_params int64 | n_trainable int64 | ops list | optimizer_steps_before_measure int64 | power_w null | requested_gpu string | schema string | source_file string | split string | status string | tf32 bool | timing dict | work_units int64 | error string | error_code string |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
{
"family": "encoder",
"key": "002d60f5d1578f10",
"kind": "kernels",
"params": {
"attention": "sdpa",
"batch": 8,
"causal": false,
"d_ff": 1536,
"d_model": 384,
"dtype": "bf16",
"gated_mlp": false,
"grad_checkpointing": "none",
"group": "5653ee18d663",
"n_heads": 6,
"... | {
"cuda": "12.8",
"device": "cuda",
"device_name": "NVIDIA A10",
"hip": null,
"multi_processor_count": 72,
"python": "3.11.12",
"torch": "2.8.0+cu128",
"total_memory": 23684186112
} | A10 | 7.874065 | {
"edges": [
[
0,
1
],
[
1,
2
],
[
2,
3
],
[
3,
4
],
[
3,
5
],
[
3,
6
],
[
2,
7
],
[
7,
8
],
[
4,
8
],
[
5,
8
],
[... | true | 002d60f5d1578f10 | kernels | {
"peak_allocated": 3019628544,
"peak_reserved": 3409969152,
"steady_allocated": 556704256
} | 32,973,312 | 32,973,312 | [
{
"bytes": null,
"class": "other",
"count": 1,
"device_s": 0.015365248,
"flops": null,
"input_shapes": [],
"name": "ProfilerStep*"
},
{
"bytes": null,
"class": "other",
"count": 1,
"device_s": 0.0056323663,
"flops": null,
"input_shapes": [],
"name": "Optim... | 2 | null | A10 | preflight-corpus/0.1 | kernels/A10.jsonl.gz | calibration | ok | false | {
"n_steps": 5,
"phases": {
"backward": 0.0252159996,
"forward": 0.0154613762,
"optimizer": 0.0056258559
},
"source": "unprofiled steps",
"step_s": 0.046462378,
"step_s_all": [
0.046462378,
0.04657738,
0.04613196,
0.046651562,
0.046413953
]
} | 4,096 | null | null |
{
"family": "mlp",
"key": "0033b491f228350f",
"kind": "kernels",
"params": {
"batch": 8,
"depth": 2,
"dtype": "bf16",
"group": "bd1b25434383",
"in_features": 784,
"role": "probe",
"width": 2048
},
"versions": {
"corpus": "corpus@0.3.0",
"torch": "2.8.0"
}
} | {
"cuda": "12.8",
"device": "cuda",
"device_name": "NVIDIA A10",
"hip": null,
"multi_processor_count": 72,
"python": "3.11.12",
"torch": "2.8.0+cu128",
"total_memory": 23684186112
} | A10 | 3.621669 | {
"edges": [
[
0,
1
],
[
1,
2
],
[
2,
3
],
[
3,
4
],
[
4,
5
],
[
5,
6
]
],
"nodes": [
{
"op": "placeholder",
"params": 0,
"shape": [
8,
784
],
"... | true | 0033b491f228350f | kernels | {
"peak_allocated": 133558784,
"peak_reserved": 148897792,
"steady_allocated": 110259200
} | 5,824,522 | 5,824,522 | [
{
"bytes": null,
"class": "other",
"count": 1,
"device_s": 0.0009985787,
"flops": null,
"input_shapes": [],
"name": "Optimizer.step#AdamW.step"
},
{
"bytes": null,
"class": "optimizer",
"count": 1.3333333333,
"device_s": 0.0002571537,
"flops": null,
"input_sha... | 2 | null | A10 | preflight-corpus/0.1 | kernels/A10.jsonl.gz | calibration | ok | false | {
"n_steps": 5,
"phases": {
"backward": 0.000527488,
"forward": 0.000312256,
"optimizer": 0.001145728
},
"source": "unprofiled steps",
"step_s": 0.002063043,
"step_s_all": [
0.002233781,
0.002063043,
0.00204283,
0.002081391,
0.002040909
]
} | 8 | null | null |
{
"family": "encoder",
"key": "003e4cc1e914daa2",
"kind": "kernels",
"params": {
"attention": "sdpa",
"batch": 16,
"causal": false,
"d_ff": 4096,
"d_model": 1024,
"dtype": "bf16",
"gated_mlp": false,
"grad_checkpointing": "none",
"group": "172502c193d3",
"n_heads": 8,
... | {
"cuda": "12.8",
"device": "cuda",
"device_name": "NVIDIA A10",
"hip": null,
"multi_processor_count": 72,
"python": "3.11.12",
"torch": "2.8.0+cu128",
"total_memory": 23684186112
} | A10 | 9.219637 | {
"edges": [
[
0,
1
],
[
1,
2
],
[
2,
3
],
[
3,
4
],
[
3,
5
],
[
3,
6
],
[
2,
7
],
[
7,
8
],
[
4,
8
],
[
5,
8
],
[... | true | 003e4cc1e914daa2 | kernels | {
"peak_allocated": 4759655936,
"peak_reserved": 4961861632,
"steady_allocated": 2568028672
} | 159,434,752 | 159,434,752 | [
{
"bytes": null,
"class": "other",
"count": 237,
"device_s": 0.0669740107,
"flops": null,
"input_shapes": [],
"name": "Command Buffer Full"
},
{
"bytes": null,
"class": "other",
"count": 1,
"device_s": 0.030976178,
"flops": null,
"input_shapes": [],
"name"... | 2 | null | A10 | preflight-corpus/0.1 | kernels/A10.jsonl.gz | calibration | ok | false | {
"n_steps": 5,
"phases": {
"backward": 0.0625981445,
"forward": 0.0307507191,
"optimizer": 0.0271329288
},
"source": "unprofiled steps",
"step_s": 0.120714144,
"step_s_all": [
0.119250773,
0.120714144,
0.121156385,
0.12048875,
0.12106502
]
} | 4,096 | null | null |
{
"family": "cnn",
"key": "00a5a20171221a1f",
"kind": "kernels",
"params": {
"batch": 4,
"channels_last": false,
"depth": 10,
"dtype": "bf16",
"group": "07b82187fd1d",
"image": 32,
"n_classes": 1000,
"role": "probe",
"width": 96
},
"versions": {
"corpus": "corpus@0.3.... | {
"cuda": "12.8",
"device": "cuda",
"device_name": "NVIDIA A10",
"hip": null,
"multi_processor_count": 72,
"python": "3.11.12",
"torch": "2.8.0+cu128",
"total_memory": 23684186112
} | A10 | 4.756446 | {
"edges": [
[
0,
1
],
[
1,
2
],
[
2,
3
],
[
3,
4
],
[
4,
5
],
[
5,
6
],
[
6,
7
],
[
7,
8
],
[
8,
9
],
[
3,
9
],
[... | true | 00a5a20171221a1f | kernels | {
"peak_allocated": 259530240,
"peak_reserved": 281018368,
"steady_allocated": 211087872
} | 11,784,136 | 11,784,136 | [
{
"bytes": null,
"class": "other",
"count": 1,
"device_s": 0.0031215107,
"flops": null,
"input_shapes": [],
"name": "ProfilerStep*"
},
{
"bytes": null,
"class": "other",
"count": 1,
"device_s": 0.0020035667,
"flops": null,
"input_shapes": [],
"name": "Opti... | 2 | null | A10 | preflight-corpus/0.1 | kernels/A10.jsonl.gz | calibration | ok | false | {
"n_steps": 5,
"phases": {
"backward": 0.002302912,
"forward": 0.0021207039,
"optimizer": 0.002366432
},
"source": "unprofiled steps",
"step_s": 0.006971532,
"step_s_all": [
0.006838531,
0.007250259,
0.007458202,
0.006971532,
0.006708457
]
} | 4 | null | null |
{"family":"lora","key":"00ab125da3491617","kind":"kernels","params":{"attention":"sdpa","batch":2,"c(...TRUNCATED) | {"cuda":"12.8","device":"cuda","device_name":"NVIDIA A10","hip":null,"multi_processor_count":72,"pyt(...TRUNCATED) | A10 | 9.676071 | {"edges":[[0,1],[1,2],[2,3],[3,4],[1,5],[4,6],[6,7],[5,8],[7,8],[8,9],[1,9],[9,11],[10,11],[11,12],[(...TRUNCATED) | true | 00ab125da3491617 | kernels | {
"peak_allocated": 3435648512,
"peak_reserved": 3546284032,
"steady_allocated": 1957048832
} | 472,418,304 | 2,097,152 | [{"bytes":null,"class":"other","count":1.0,"device_s":0.021467962,"flops":null,"input_shapes":[],"na(...TRUNCATED) | 2 | null | A10 | preflight-corpus/0.1 | kernels/A10.jsonl.gz | not_in_calibration | ok | false | {"n_steps":5,"phases":{"backward":0.0153569279,"forward":0.0177674236,"optimizer":0.000388064},"sour(...TRUNCATED) | 512 | null | null |
{"family":"mlp","key":"00ea11d3b5c03c8a","kind":"kernels","params":{"batch":16,"depth":6,"dtype":"bf(...TRUNCATED) | {"cuda":"12.8","device":"cuda","device_name":"NVIDIA A10","hip":null,"multi_processor_count":72,"pyt(...TRUNCATED) | A10 | 4.593063 | {"edges":[[0,1],[1,2],[2,3],[3,4],[4,5],[5,6],[6,7],[7,8],[8,9],[9,10],[10,11],[11,12],[12,13],[13,1(...TRUNCATED) | true | 00ea11d3b5c03c8a | kernels | {
"peak_allocated": 1760351232,
"peak_reserved": 1939865600,
"steady_allocated": 1411698176
} | 87,162,890 | 87,162,890 | [{"bytes":null,"class":"other","count":1.0,"device_s":0.0147924973,"flops":null,"input_shapes":[],"n(...TRUNCATED) | 2 | null | A10 | preflight-corpus/0.1 | kernels/A10.jsonl.gz | calibration | ok | false | {"n_steps":5,"phases":{"backward":0.0019486721,"forward":0.001741824,"optimizer":0.0147671041},"sour(...TRUNCATED) | 16 | null | null |
{"family":"encoder","key":"012662f6003274c1","kind":"kernels","params":{"attention":"math","batch":2(...TRUNCATED) | {"cuda":"12.8","device":"cuda","device_name":"NVIDIA A10","hip":null,"multi_processor_count":72,"pyt(...TRUNCATED) | A10 | 7.067622 | {"edges":[[0,1],[1,2],[2,3],[3,4],[3,5],[3,6],[2,7],[7,8],[4,8],[5,8],[8,9],[2,10],[10,11],[4,11],[5(...TRUNCATED) | true | 012662f6003274c1 | kernels | {
"peak_allocated": 585050624,
"peak_reserved": 624951296,
"steady_allocated": 415154688
} | 24,398,592 | 24,398,592 | [{"bytes":null,"class":"other","count":1.0,"device_s":0.0154949313,"flops":null,"input_shapes":[],"n(...TRUNCATED) | 2 | null | A10 | preflight-corpus/0.1 | kernels/A10.jsonl.gz | calibration | ok | false | {"n_steps":5,"phases":{"backward":0.0117483521,"forward":0.007543808,"optimizer":0.0049807358},"sour(...TRUNCATED) | 512 | null | null |
{"family":"lora","key":"012bcc29a03e700e","kind":"kernels","params":{"attention":"sdpa","batch":8,"c(...TRUNCATED) | {"cuda":"12.8","device":"cuda","device_name":"NVIDIA A10","hip":null,"multi_processor_count":72,"pyt(...TRUNCATED) | A10 | 7.519896 | {"edges":[[0,1],[1,2],[2,3],[3,4],[1,5],[4,6],[6,7],[5,8],[7,8],[8,9],[1,9],[9,11],[10,11],[11,12],[(...TRUNCATED) | true | 012bcc29a03e700e | kernels | {
"peak_allocated": 1380652544,
"peak_reserved": 1398800384,
"steady_allocated": 281244160
} | 64,476,160 | 196,608 | [{"bytes":null,"class":"other","count":1.0,"device_s":0.0260192963,"flops":null,"input_shapes":[],"n(...TRUNCATED) | 2 | null | A10 | preflight-corpus/0.1 | kernels/A10.jsonl.gz | not_in_calibration | ok | false | {"n_steps":5,"phases":{"backward":0.0103772163,"forward":0.0136017923,"optimizer":0.000287744},"sour(...TRUNCATED) | 1,024 | null | null |
{"family":"encoder","key":"0170fb2372159c9f","kind":"kernels","params":{"attention":"sdpa","batch":1(...TRUNCATED) | {"cuda":"12.8","device":"cuda","device_name":"NVIDIA A10","hip":null,"multi_processor_count":72,"pyt(...TRUNCATED) | A10 | 6.244395 | {"edges":[[0,1],[1,2],[2,3],[3,4],[3,5],[3,6],[2,7],[7,8],[4,8],[5,8],[8,9],[2,10],[10,11],[4,11],[5(...TRUNCATED) | true | 0170fb2372159c9f | kernels | {
"peak_allocated": 4774096384,
"peak_reserved": 5354029056,
"steady_allocated": 2628157952
} | 163,192,832 | 163,192,832 | [{"bytes":null,"class":"other","count":1.0,"device_s":0.0278137343,"flops":null,"input_shapes":[],"n(...TRUNCATED) | 2 | null | A10 | preflight-corpus/0.1 | kernels/A10.jsonl.gz | calibration | ok | false | {"n_steps":5,"phases":{"backward":0.0515563507,"forward":0.0270888958,"optimizer":0.0277585926},"sou(...TRUNCATED) | 4,096 | null | null |
{"family":"cnn","key":"0171e4d1e9877d3a","kind":"kernels","params":{"batch":16,"channels_last":true,(...TRUNCATED) | {"cuda":"12.8","device":"cuda","device_name":"NVIDIA A10","hip":null,"multi_processor_count":72,"pyt(...TRUNCATED) | A10 | 5.918856 | {"edges":[[0,1],[1,2],[2,3],[3,4],[4,5],[5,6],[6,7],[7,8],[8,9],[3,9],[9,10],[10,11],[11,12],[12,13](...TRUNCATED) | true | 0171e4d1e9877d3a | kernels | {
"peak_allocated": 1151178752,
"peak_reserved": 1298137088,
"steady_allocated": 810359296
} | 48,628,168 | 48,628,168 | [{"bytes":null,"class":"other","count":1.0,"device_s":0.0088584073,"flops":null,"input_shapes":[],"n(...TRUNCATED) | 2 | null | A10 | preflight-corpus/0.1 | kernels/A10.jsonl.gz | calibration | ok | false | {"n_steps":5,"phases":{"backward":0.0171415672,"forward":0.00890592,"optimizer":0.0082575359},"sourc(...TRUNCATED) | 16 | null | null |
End of preview.
Preflight corpus 0.4.0
Measured training jobs on real GPUs (Modal, Oct 28 to Nov 2, 2026, and one rented AMD MI300X on Nov 2): memory,
step time and per-op profiles for 3129 small-model configs and 100 Hugging Face LoRA
and QLoRA jobs, and the loss, grad-norm and logit series of 1026
short GPT runs with a hindsight divergence label (720 from Nov 1's sweep and
306 from Nov 2's sweep at usual learning rates). Preflight (github.com/uchuuronin/preflight)
uses it to calibrate and check its prediction intervals. The full datasheet is python/preflight/corpus/DATASHEET.md
in the repository.
- File:
preflight-corpus-0.4.0.jsonl.gz, 4255 rows, 31,723,814 bytes, sha2568719593587b6b6741afa6c2f12f83883065119e0634c6b6ac91c55ceaa3101f4. - Row: one JSON object per line;
kind,splitandsource_fileadded to what the corpus worker wrote. - GPUs: NVIDIA L4, A10, L40S, T4, H100-SXM-80GB, A100-PCIe-80GB and A100-SXM-80GB (torch 2.8.0, CUDA 12.8), and
AMD MI300X (torch 2.5.1, ROCm 6.2). A config's
keyincludes the torch version, so pair rows across vendors onconfig.family,config.kindandconfig.params. - Licence: Apache 2.0, the repository's licence, for the measurements. No model weights and no dataset text are included.
- Sources: the configs are generated by
python/preflight/corpus; the Hugging Face jobs time Qwen/Qwen2.5-0.5B, Qwen/Qwen2.5-1.5B, HuggingFaceTB/SmolLM2-360M, HuggingFaceTB/SmolLM2-1.7B and TinyLlama/TinyLlama-1.1B-Chat-v1.0 (all Apache 2.0); the divergence runs train on Tiny Shakespeare (karpathy/char-rnn, public-domain text) and store only training curves, not text. - Splits:
calibration(calibrated on, leave-one-architecture-out),cross_gpu_holdout(every A100 and MI300X row, scored once, never calibrated on),not_in_calibration(the synthetic LoRA family),divergence(both divergence kinds; they score the divergence classifier only). - Limits: small models (at most 1.7B parameters, single GPU), five timed steps per config, random inputs for the timing kinds, one AMD GPU on another torch version than the NVIDIA rows. Not for ranking GPUs for purchase or for estimating how fast a particular production model trains.
| kind | gpu | status | split | rows |
|---|---|---|---|---|
| divergence | A10 | ok | divergence | 343 |
| divergence | L4 | ok | divergence | 377 |
| divergence_real | A10 | ok | divergence | 300 |
| divergence_real | L4 | ok | divergence | 6 |
| hf | A10 | ok | calibration | 49 |
| hf | A10 | oom | calibration | 1 |
| hf | L4 | ok | calibration | 49 |
| hf | L4 | oom | calibration | 1 |
| kernels | A10 | ok | calibration | 665 |
| kernels | A10 | ok | not_in_calibration | 175 |
| kernels | A10 | oom | calibration | 12 |
| kernels | A10 | oom | not_in_calibration | 5 |
| kernels | A100-PCIe-80GB | ok | cross_gpu_holdout | 42 |
| kernels | A100-SXM-80GB | ok | cross_gpu_holdout | 53 |
| kernels | H100-SXM-80GB | ok | calibration | 392 |
| kernels | H100-SXM-80GB | ok | not_in_calibration | 100 |
| kernels | L4 | ok | calibration | 519 |
| kernels | L4 | ok | not_in_calibration | 136 |
| kernels | L4 | oom | calibration | 7 |
| kernels | L4 | oom | not_in_calibration | 4 |
| kernels | L40S | ok | calibration | 465 |
| kernels | L40S | ok | not_in_calibration | 120 |
| kernels | L40S | oom | calibration | 2 |
| kernels | MI300X | ok | cross_gpu_holdout | 295 |
| kernels | T4 | ok | calibration | 115 |
| kernels | T4 | ok | not_in_calibration | 14 |
| kernels | T4 | oom | calibration | 7 |
| kernels | T4 | oom | not_in_calibration | 1 |
- Downloads last month
- 5