The dataset viewer is not available for this subset.
Exception: SplitsNotFoundError
Message: The split names could not be parsed from the dataset config.
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 290, in _generate_tables
pa_table = paj.read_json(
io.BytesIO(batch), read_options=paj.ReadOptions(block_size=block_size)
)
File "pyarrow/_json.pyx", line 342, in pyarrow._json.read_json
File "pyarrow/error.pxi", line 155, in pyarrow.lib.pyarrow_internal_check_status
return check_status(status)
File "pyarrow/error.pxi", line 92, in pyarrow.lib.check_status
raise convert_status(status)
pyarrow.lib.ArrowInvalid: JSON parse error: Column() changed from object to string in row 0
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 286, in get_dataset_config_info
for split_generator in builder._split_generators(
~~~~~~~~~~~~~~~~~~~~~~~~~^
StreamingDownloadManager(base_path=builder.base_path, download_config=download_config)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 101, in _split_generators
pa_table = next(iter(self._generate_tables(**splits[0].gen_kwargs, allow_full_read=False)))[1]
~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 304, in _generate_tables
batch = json_encode_fields_in_json_lines(original_batch, json_field_paths)
File "/usr/local/lib/python3.14/site-packages/datasets/utils/json.py", line 111, in json_encode_fields_in_json_lines
examples = [ujson_loads(line) for line in original_batch.splitlines()]
~~~~~~~~~~~^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/utils/json.py", line 20, in ujson_loads
return pd.io.json.ujson_loads(*args, **kwargs)
~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
ValueError: Trailing data
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/split_names.py", line 68, in compute_split_names_from_streaming_response
for split in get_dataset_split_names(
~~~~~~~~~~~~~~~~~~~~~~~^
path=dataset,
^^^^^^^^^^^^^
config_name=config,
^^^^^^^^^^^^^^^^^^^
token=hf_token,
^^^^^^^^^^^^^^^
)
^
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 340, in get_dataset_split_names
info = get_dataset_config_info(
path,
...<6 lines>...
**config_kwargs,
)
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 291, in get_dataset_config_info
raise SplitsNotFoundError("The split names could not be parsed from the dataset config.") from err
datasets.inspect.SplitsNotFoundError: The split names could not be parsed from the dataset config.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.
FLUX.1-dev cache schedule–estimator separability pilot
This public dataset contains the outputs and analysis of a frozen FLUX.1-dev TPU/JAX pilot. It compares raw reuse of the pre-final-block feature with a simple first-order feature extrapolator (taylor_lite_o1) across 24 fixed-refresh-count schedules at nominal N=5, 7, and 10. The final block/head always uses current-step conditioning. No model weights were trained or modified.
Protocol
- 8 fixed DrawBench200 prompts, one seed per prompt (seed = prompt index).
- 50 denoising steps, 1024×1024, guidance 3.5, corrected RoPE theta 10000.
- Full teacher output generated once per prompt/seed and reused for all comparisons.
- 24 schedules per N: one uniform, 12 local jitter, 8 Dirichlet, and 3 warped; fixed full-refresh counts of 13, 10, and 8 for N=5, 7, 10 respectively.
- The existing TPU protocol requires full refreshes at steps 0–3. This is not a zero-warmup benchmark.
- Raw uniform N=5 matched the existing cache-only baseline bit-for-bit in the smoke test.
- LPIPS (AlexNet), PSNR, and SSIM are measured against the full teacher image. Sampling latency excludes model compilation, text encoding, VAE decoding, and metrics.
Pilot findings
| N | Spearman raw vs Taylor | Raw→Taylor transfer regret | Taylor harm rate |
|---|---|---|---|
| 5 | 0.849 | 0.00% | 50.5% |
| 7 | 0.937 | 0.00% | 58.9% |
| 10 | 0.929 | 13.78% | 62.0% |
At N=10, the raw-optimal schedule is n10_17, while the Taylor-optimal schedule is n10_21. Broad rank correlation remains high, but the top schedule does not transfer. Taylor does not uniformly improve raw reuse. See pilot/analysis/REPORT.md and pilot/analysis/summary.json for the full analysis.
These are exploratory in-sample results: schedule selection and evaluation used the same 8 prompts. The prompt-bootstrap 95% interval for N=10 raw→Taylor regret is [0, 23.06]%, so held-out validation is still needed. taylor_lite_o1 is the simple extrapolator defined in this repository, not a verified reproduction of official TaylorSeer. Do not generalize these findings to TaylorSeer without a separate implementation check.
Files
cache_schedule_separability_results_report.zip: images, CSVs, plots, report, schedules, code, and smoke checks (no pilot latents).cache_schedule_separability_full_with_latents.zip: all of the above plus final pilot latents.pilot/results.csv: 1,152 evaluated candidate rows.pilot/generation.csv: generation metadata and measured sampling times.pilot/analysis/: aggregate CSV, JSON summary, plots, and report.schedule_sets/: the exact 24 schedules for each budget.
The archives contain no FLUX model weights, Hugging Face credentials, or other secrets. The manifest.json records the prompt source, model settings, schedule hashes, and timing scope.
- Downloads last month
- 37