Datasets:

chendeting
/

safe44-terminus2-qwen36-traces

Tags:

License:

Dataset card Files Files and versions

xet

Community

Dataset Viewer

Duplicate

Split (1)

train

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
task: string
idx: int64
lane: int64
status: string
rc: string
reward: string
errors: string
duration_min: string
requests: string
kv_new_tokens: int64
kv_cached_tokens: int64
kv_hit_rate: string
harbor_input_tokens: string
harbor_cache_tokens: string
harbor_output_tokens: string
batch: string
trace_limit_bytes_per_task: int64
sglang: struct<mem_fraction_static: double, prefix_trace_env: string, reasoning_parser: string, tool_call_pa (... 13 chars omitted)
  child 0, mem_fraction_static: double
  child 1, prefix_trace_env: string
  child 2, reasoning_parser: string
  child 3, tool_call_parser: string
model: string
created_at: string
tasks: list<item: string>
  child 0, item: string
task_count: int64
agent: string
lanes: list<item: struct<lane: int64, gpu: int64, sglang_port: int64, max_concurrent_tasks: int64>>
  child 0, item: struct<lane: int64, gpu: int64, sglang_port: int64, max_concurrent_tasks: int64>
      child 0, lane: int64
      child 1, gpu: int64
      child 2, sglang_port: int64
      child 3, max_concurrent_tasks: int64
harbor_delete: bool
model_path: string
terminus_2_adjustments: struct<interleaved_thinking: bool, store_all_messages: bool, chat_template_kwargs: struct<enable_thi (... 38 chars omitted)
  child 0, interleaved_thinking: bool
  child 1, store_all_messages: bool
  child 2, chat_template_kwargs: struct<enable_thinking: bool, preserve_thinking: bool>
      child 0, enable_thinking: bool
      child 1, preserve_thinking: bool
to
{'batch': Value('string'), 'created_at': Value('string'), 'task_count': Value('int64'), 'tasks': List(Value('string')), 'agent': Value('string'), 'model': Value('string'), 'model_path': Value('string'), 'lanes': List({'lane': Value('int64'), 'gpu': Value('int64'), 'sglang_port': Value('int64'), 'max_concurrent_tasks': Value('int64')}), 'trace_limit_bytes_per_task': Value('int64'), 'harbor_delete': Value('bool'), 'terminus_2_adjustments': {'interleaved_thinking': Value('bool'), 'store_all_messages': Value('bool'), 'chat_template_kwargs': {'enable_thinking': Value('bool'), 'preserve_thinking': Value('bool')}}, 'sglang': {'mem_fraction_static': Value('float64'), 'prefix_trace_env': Value('string'), 'reasoning_parser': Value('string'), 'tool_call_parser': 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 478, 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 2818, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2355, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2380, 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 2369, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2297, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              task: string
              idx: int64
              lane: int64
              status: string
              rc: string
              reward: string
              errors: string
              duration_min: string
              requests: string
              kv_new_tokens: int64
              kv_cached_tokens: int64
              kv_hit_rate: string
              harbor_input_tokens: string
              harbor_cache_tokens: string
              harbor_output_tokens: string
              batch: string
              trace_limit_bytes_per_task: int64
              sglang: struct<mem_fraction_static: double, prefix_trace_env: string, reasoning_parser: string, tool_call_pa (... 13 chars omitted)
                child 0, mem_fraction_static: double
                child 1, prefix_trace_env: string
                child 2, reasoning_parser: string
                child 3, tool_call_parser: string
              model: string
              created_at: string
              tasks: list<item: string>
                child 0, item: string
              task_count: int64
              agent: string
              lanes: list<item: struct<lane: int64, gpu: int64, sglang_port: int64, max_concurrent_tasks: int64>>
                child 0, item: struct<lane: int64, gpu: int64, sglang_port: int64, max_concurrent_tasks: int64>
                    child 0, lane: int64
                    child 1, gpu: int64
                    child 2, sglang_port: int64
                    child 3, max_concurrent_tasks: int64
              harbor_delete: bool
              model_path: string
              terminus_2_adjustments: struct<interleaved_thinking: bool, store_all_messages: bool, chat_template_kwargs: struct<enable_thi (... 38 chars omitted)
                child 0, interleaved_thinking: bool
                child 1, store_all_messages: bool
                child 2, chat_template_kwargs: struct<enable_thinking: bool, preserve_thinking: bool>
                    child 0, enable_thinking: bool
                    child 1, preserve_thinking: bool
              to
              {'batch': Value('string'), 'created_at': Value('string'), 'task_count': Value('int64'), 'tasks': List(Value('string')), 'agent': Value('string'), 'model': Value('string'), 'model_path': Value('string'), 'lanes': List({'lane': Value('int64'), 'gpu': Value('int64'), 'sglang_port': Value('int64'), 'max_concurrent_tasks': Value('int64')}), 'trace_limit_bytes_per_task': Value('int64'), 'harbor_delete': Value('bool'), 'terminus_2_adjustments': {'interleaved_thinking': Value('bool'), 'store_all_messages': Value('bool'), 'chat_template_kwargs': {'enable_thinking': Value('bool'), 'preserve_thinking': Value('bool')}}, 'sglang': {'mem_fraction_static': Value('float64'), 'prefix_trace_env': Value('string'), 'reasoning_parser': Value('string'), 'tool_call_parser': Value('string')}}
              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.

Safe44 Terminus-2 + Qwen3.6-27B Trace Dataset

This dataset contains a curated trace package from a Terminal-Bench / Harbor run using terminus-2 against a local Qwen3.6-27B model served by SGLang.

The run was manually stopped after preserving completed traces. The dataset intentionally keeps the important analysis artifacts and compact request ledgers, rather than full task containers.

Run summary

Planned tasks: 44
Started tasks: 42
Completed tasks: 40
Manually stopped tasks: polyglot-rust-c, video-processing
Not started tasks: vulnerable-secret, winning-avg-corewars
Known reward pass / fail: 21 / 12
Mean reward over known-reward completed tasks: 0.6364
Completed-task KV hit rate from SGLang logs: 97.0669%
Completed-task API requests joined to SGLang events: 1828

Files

data/summary.json                         high-level aggregate metrics
data/per_task_summary.csv                 task-level reward, duration, request, token, KV metrics
data/per_task_summary.json                JSON version of task-level summary
data/traceid_kv_report.manual_stop.json   SGLang trace-id to KV-cache report
data/job_summary.manual_stop.txt          text summary generated after manual stop
ledgers/*.jsonl                           per-request proxy ledgers with trace_id, token usage, hashes, sizes
status/*.status                           task lifecycle and trace-size watchdog status
logs/qwen-local.sglang.manual_stop.log.gz compressed SGLang log used for KV analysis
figures/*.png                             visualization figures

Important caveats

polyglot-rust-c was manually stopped to free GPU1.
video-processing was manually stopped when the whole safe44 batch was intentionally aborted.
vulnerable-secret and winning-avg-corewars were not started.
Harbor/LiteLLM n_cache_tokens fields are not used for KV-cache conclusions; KV hit rates here are computed from SGLang prefix-cache log events joined to proxy trace_ids.
Ledgers store request metadata, usage, hashes, and byte sizes, not full raw prompt bodies.

Quick read

The main table is data/per_task_summary.csv. The most important derived metrics are:

result status: status, reward, errors
runtime: duration_min
request volume: requests
SGLang cache behavior: kv_new_tokens, kv_cached_tokens, kv_hit_rate

Visualizations

Downloads last month: 119

Total file size:

3.38 MB