Dataset Viewer
Duplicate
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
status: string
checks: list<item: struct<name: string, returncode: int64, seconds: double, stdout: string, stderr: string>>
  child 0, item: struct<name: string, returncode: int64, seconds: double, stdout: string, stderr: string>
      child 0, name: string
      child 1, returncode: int64
      child 2, seconds: double
      child 3, stdout: string
      child 4, stderr: string
source_files: int64
clean_directory: bool
execution_location: string
secret_pattern_text_files: int64
gpu_jobs_started: int64
paid_api_calls: int64
new_full_benchmark_runs: int64
long_metrics: int64
paper_metrics: int64
sessions: int64
values: struct<Model: string, turn-taking/TOR: double, turn-taking/Response Latency: double, pause/TOR_pre:  (... 347 chars omitted)
  child 0, Model: string
  child 1, turn-taking/TOR: double
  child 2, turn-taking/Response Latency: double
  child 3, pause/TOR_pre: double
  child 4, pause/TOR_post: double
  child 5, pause/Response Latency: double
  child 6, hesitation/FinalAcc: double
  child 7, hesitation/TOR_post: double
  child 8, hesitation/TOR_pre: double
  child 9, interruption/TOR: double
  child 10, interruption/Stop Latency: double
  child 11, interruption/FinalAcc: double
  child 12, backchannel/TOR: double
  child 13, backchannel/Resume: double
  child 14, third-party/TOR: double
  child 15, third-party/Resume: double
events: int64
scope: string
to
{'status': Value('string'), 'sessions': Value('int64'), 'events': Value('int64'), 'long_metrics': Value('int64'), 'paper_metrics': Value('int64'), 'scope': Value('string'), 'values': {'Model': Value('string'), 'turn-taking/TOR': Value('float64'), 'turn-taking/Response Latency': Value('float64'), 'pause/TOR_pre': Value('float64'), 'pause/TOR_post': Value('float64'), 'pause/Response Latency': Value('float64'), 'hesitation/FinalAcc': Value('float64'), 'hesitation/TOR_post': Value('float64'), 'hesitation/TOR_pre': Value('float64'), 'interruption/TOR': Value('float64'), 'interruption/Stop Latency': Value('float64'), 'interruption/FinalAcc': Value('float64'), 'backchannel/TOR': Value('float64'), 'backchannel/Resume': Value('float64'), 'third-party/TOR': Value('float64'), 'third-party/Resume': Value('float64')}}
because column names don't match
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/utils.py", line 149, 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 129, 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 544, 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 3052, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2559, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2584, 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 558, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 440, 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 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 2309, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              status: string
              checks: list<item: struct<name: string, returncode: int64, seconds: double, stdout: string, stderr: string>>
                child 0, item: struct<name: string, returncode: int64, seconds: double, stdout: string, stderr: string>
                    child 0, name: string
                    child 1, returncode: int64
                    child 2, seconds: double
                    child 3, stdout: string
                    child 4, stderr: string
              source_files: int64
              clean_directory: bool
              execution_location: string
              secret_pattern_text_files: int64
              gpu_jobs_started: int64
              paid_api_calls: int64
              new_full_benchmark_runs: int64
              long_metrics: int64
              paper_metrics: int64
              sessions: int64
              values: struct<Model: string, turn-taking/TOR: double, turn-taking/Response Latency: double, pause/TOR_pre:  (... 347 chars omitted)
                child 0, Model: string
                child 1, turn-taking/TOR: double
                child 2, turn-taking/Response Latency: double
                child 3, pause/TOR_pre: double
                child 4, pause/TOR_post: double
                child 5, pause/Response Latency: double
                child 6, hesitation/FinalAcc: double
                child 7, hesitation/TOR_post: double
                child 8, hesitation/TOR_pre: double
                child 9, interruption/TOR: double
                child 10, interruption/Stop Latency: double
                child 11, interruption/FinalAcc: double
                child 12, backchannel/TOR: double
                child 13, backchannel/Resume: double
                child 14, third-party/TOR: double
                child 15, third-party/Resume: double
              events: int64
              scope: string
              to
              {'status': Value('string'), 'sessions': Value('int64'), 'events': Value('int64'), 'long_metrics': Value('int64'), 'paper_metrics': Value('int64'), 'scope': Value('string'), 'values': {'Model': Value('string'), 'turn-taking/TOR': Value('float64'), 'turn-taking/Response Latency': Value('float64'), 'pause/TOR_pre': Value('float64'), 'pause/TOR_post': Value('float64'), 'pause/Response Latency': Value('float64'), 'hesitation/FinalAcc': Value('float64'), 'hesitation/TOR_post': Value('float64'), 'hesitation/TOR_pre': Value('float64'), 'interruption/TOR': Value('float64'), 'interruption/Stop Latency': Value('float64'), 'interruption/FinalAcc': Value('float64'), 'backchannel/TOR': Value('float64'), 'backchannel/Resume': Value('float64'), 'third-party/TOR': Value('float64'), 'third-party/Resume': Value('float64')}}
              because column names don't match

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.

HFD Motion Bench · 2026-11-09 release

中文说明 · Exact commands · Metric definitions · Results and caveats

This repository bundles the latest integrated motion evaluation protocols, their immutable upstream code/data, and the completed HFD reference artifacts. It supports both a quick CPU audit of the published numbers and a full rerun from generated motion trajectories.

Component Scope Reference experiment
Motion quality 14 metrics; 1024 complete sessions; 1920 fixed scoring windows HFD checkpoint 2949079
Motion interaction 15 paper metrics / 18 long-form metrics; 1024 sessions; 2229 events HFD checkpoint 3276036
Optional SONIC diagnostic Four walking controls and eight selected HoloMotion failure cases A small tracker comparison; not a replacement for the formal benchmark

These are two different completed checkpoints. This release does not assert that they are the latest training checkpoint or that one checkpoint produced both tables. Newer model experiments, speech evaluation, and training are outside this release.

Start here

Linux with Python 3.10 is the supported full-evaluation environment. The quick replay needs only NumPy and SciPy, no CUDA, model weights, paid service, or HFD checkpoint.

python3.10 -m venv ./env-replay
./env-replay/bin/pip install 'huggingface_hub==0.36.2' 'numpy==1.26.4' 'scipy==1.15.3'
./env-replay/bin/hf download Tinghong-Ye22/motion_bench_1109 \
  --repo-type dataset --local-dir ./motion_bench_1109
cd motion_bench_1109
../env-replay/bin/python bench.py verify
../env-replay/bin/python bench.py replay --output ../replayed-results

verify checks the publication SHA256 manifest and both original upstream package locks. replay executes original metric functions on the saved features/statistics and original interaction aggregation on the saved per-event labels. It verifies every resulting reference value before succeeding. It does not rerun neural feature extraction, physics simulation, or automatic action recognition. See QUICKSTART for those complete reruns and for evaluating another model.

For a small download sufficient for replay, use the --include patterns in QUICKSTART. For permanent experiments, pin the HF commit shown in this repository's history with --revision COMMIT_SHA; all external scoring models are already pinned by revision and SHA256. Public downloads require no account token.

What is included

bench.py                         portable prepare / verify / replay / unpack entry point
quality/                         original quality release, code and data locks unchanged
interaction/                     original interaction release, code and data locks unchanged
shared/inputs.tar                one copy of the shared Test1024 causal inputs
payloads/                        1024+1024 full predictions, packed quality targets,
                                 HoloMotion executed trajectories, final recognition labels
results/quality/HFD_2949079/      original quality scores, features, per-window statistics
results/interaction/HFD_3276036/  original final metrics, boundaries and paper exports
results/baselines/                seven interaction baseline aggregate rows and provenance
diagnostics/sonic/                selected-case tracker comparison and review videos
tools/                           CPU replay and narrow runtime compatibility adapters
docs/                            protocols, definitions, schemas, commands and limitations
RELEASE.json                     immutable source revisions and experiment identities
RELEASE_MANIFEST.json            path / size / SHA256 of publication files
validation/                      validation performed for this publication

The complete input archive is stored once. bench.py prepare creates the local quality/inputs.tar alias required by the unchanged original quality CLI. The archive is not duplicated in the publication.

Scoring weights (~12 GB for quality; ~9 GB for Qwen recognition) remain in their official HF repositories. Included downloaders fetch the exact locked revisions and verify their bytes. HFD model weights are not included: the supplied complete predictions allow reproduction of scoring without them; generating new native HFD predictions requires the corresponding model implementation and checkpoint.

Interpretation that matters for the paper

  • Quality FID is raw FID in the paper-unit CSV; the original delivery JSON scales it by 100. C-Slide is m/s in the paper CSV; the delivery JSON scales it by 10. Do not mix the two exports.
  • The six quality tracking columns measure one fixed HoloMotion v1.2 policy following the generated target in simulation. They are not instruction success, human-reference pose accuracy, or physical-robot execution results.
  • Interaction percentages count eligible unknown events at the adverse bound. Latencies are observed medians, not failure-imputed zeroes or worst-case delays. FinalAcc is final intent selection.
  • Interaction recognition uses the released unknown-recovery extensions and an uncalibrated similarity threshold. Residual unknown duration is 0.13984%; this is coverage, not 99.86% recognition accuracy.
  • HFD used causal audio without oracle transcript input; the supplied quality baselines used released oracle text. Input modality differences must be disclosed.

See PROTOCOL, METRICS, RESULTS and REPRODUCIBILITY before quoting a number. Upstream, model, and data terms remain applicable; see sources and licenses. No blanket license is asserted over third-party assets.

Downloads last month
244