The dataset viewer is not available for this split.
Error code: StreamingRowsError
Exception: CastError
Message: Couldn't cast
text: string
records: list<item: struct<directory: string, status: string, log: string, reason: string, excluded: list<ite (... 92 chars omitted)
child 0, item: struct<directory: string, status: string, log: string, reason: string, excluded: list<item: struct<f (... 80 chars omitted)
child 0, directory: string
child 1, status: string
child 2, log: string
child 3, reason: string
child 4, excluded: list<item: struct<file: string, bytes: int64, reason: string>>
child 0, item: struct<file: string, bytes: int64, reason: string>
child 0, file: string
child 1, bytes: int64
child 2, reason: string
child 5, seconds: double
child 6, returncode: int64
scope: string
to
{'records': List({'directory': Value('string'), 'status': Value('string'), 'log': Value('string'), 'reason': Value('string'), 'excluded': List({'file': Value('string'), 'bytes': Value('int64'), 'reason': Value('string')}), 'seconds': Value('float64'), 'returncode': Value('int64')}), 'scope': Value('string')}
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
text: string
records: list<item: struct<directory: string, status: string, log: string, reason: string, excluded: list<ite (... 92 chars omitted)
child 0, item: struct<directory: string, status: string, log: string, reason: string, excluded: list<item: struct<f (... 80 chars omitted)
child 0, directory: string
child 1, status: string
child 2, log: string
child 3, reason: string
child 4, excluded: list<item: struct<file: string, bytes: int64, reason: string>>
child 0, item: struct<file: string, bytes: int64, reason: string>
child 0, file: string
child 1, bytes: int64
child 2, reason: string
child 5, seconds: double
child 6, returncode: int64
scope: string
to
{'records': List({'directory': Value('string'), 'status': Value('string'), 'log': Value('string'), 'reason': Value('string'), 'excluded': List({'file': Value('string'), 'bytes': Value('int64'), 'reason': Value('string')}), 'seconds': Value('float64'), 'returncode': Value('int64')}), 'scope': Value('string')}
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.
One-agent CPU trace pilot — converted traces
Private research dataset: validated canonical (converted) DynamoRIO CPU traces, not the original offline raw recordings. The canonical files are binary ZIP traces for DynamoRIO analysis tools, not human-readable CSV.
Upload status
The transfer is complete only when UPLOAD_COMPLETE.json exists and reports status: complete. Until then the repository may contain only part of the 358 expected canonical files. Final canonical size: 97,565,806,045 bytes, across 117 process-image traces/views. The main trace is approximately 48.6 GB compressed.
Experiment
One live mini-SWE-agent 2.4.6 execution on SWE-bench instance sympy__sympy-15599, with Qwen3-Coder-30B-A3B-Instruct AWQ inference on a separate server. Collection completed normally in approximately 15 minutes with 57 model requests. The agent submitted a patch; benchmark correctness was not evaluated. These are user-space CPU accesses in the agent and inherited tools. GPU inference, model-server CPU activity and kernel accesses are excluded.
Contents
canonical/<process-image>/trace/: validated converted traces and scheduling metadata. Original raw recordings remain preserved on Phoebe and are not included here.CANONICAL_MANIFEST.json: relative file paths, byte sizes and source SHA256 checksums, added during transfer.validation-final/: completed retry report and all 22 new consistency-check logs, commands and exclusion records.validation-snapshot/: successful logs for the previous 95 validated process traces/views and their recorded exclusions.metadata/andcode/: run context and validation helpers. Earlier snapshot metadata is historical; the final validation report below supersedes it.UPLOAD_COMPLETE.json: final size/checksum verification at a pinned repository revision.
Validation and coverage
All 117 usable process-image traces/views passed DynamoRIO conversion and invariant checking: 43 original traces, 52 initial derived views, and 22 longer retries. The longer validation job took approximately 8 hours 20 minutes. validation-final/completed.json is the final status report; entries marked previously validated refer to the earlier logs.
Four original shell-process directories contain only seven-byte raw fragments and lack metadata: drmemtrace.dash.00489.3463.dir, drmemtrace.dash.00490.3463.dir, drmemtrace.dash.00581.7841.dir, and drmemtrace.dash.00582.7841.dir. They have no validated canonical output. Additional empty fork/exec fragments excluded from populated-stream views are listed in the relevant exclusion records. All raw originals are preserved on Phoebe.
117/121 is a count of directories, not a percentage of memory accesses captured. Passing checks establishes internal consistency, not exhaustive CPU coverage, task correctness or a dead-state finding. Allocation/free events, address reuse and Python suballocation need further analysis. Virtual addresses must be interpreted within each process/address-space lifetime.
Download and analyse
Authenticate with a Hugging Face account granted access, then download with huggingface_hub.snapshot_download using repository ID harry1332/swe-agent-cpu-dynamorio-validated-private and repo_type="dataset". Plan for approximately 98 GB of downloaded data.
Use DynamoRIO cronbuild-11.91.20708, matching collection and validation. An example for one downloaded process trace is:
/path/to/DynamoRIO/bin64/drrun -t drmemtrace -indir /path/to/dataset/canonical/PROCESS_DIRECTORY -tool basic_counts
The canonical format is already converted. Do not run raw-to-canonical conversion again. The raw collection environment referenced private container paths; some additional analysis tools may still need binaries or symbols. Contact the dataset owner for the preserved module archive if required.
Only the single-agent pilot is included. The 10-agent and 100-agent experiments have not been collected. No blanket license is asserted over third-party instruction encodings or referenced software.
- Downloads last month
- 50