Datasets:

JitaiHao
/

SWE-lite-trace100

Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
schema: string
timing_boundary: string
timing_quality: string
timing_error_bound_s: null
timing_caveat: string
selection: string
source_run_config: struct<batch_size: int64, chain_cache: bool, cost_limit: double, cost_tracking: string, dataset: str (... 344 chars omitted)
  child 0, batch_size: int64
  child 1, chain_cache: bool
  child 2, cost_limit: double
  child 3, cost_tracking: string
  child 4, dataset: string
  child 5, dataset_rows_sha256: string
  child 6, enable_thinking: bool
  child 7, instance_ids: list<item: string>
      child 0, item: string
  child 8, max_tokens: int64
  child 9, model: string
  child 10, preserve_thinking: bool
  child 11, seed: null
  child 12, seed_control: string
  child 13, served_model_name: string
  child 14, server_manifest_sha256: string
  child 15, split: string
  child 16, step_limit: int64
  child 17, temperature: double
  child 18, top_p: double
  child 19, wall_time_limit_seconds: int64
agents: list<item: struct<instance_id: string, file: string, sha256: string, turns: int64, exit_status: stri (... 27 chars omitted)
  child 0, item: struct<instance_id: string, file: string, sha256: string, turns: int64, exit_status: string, source_ (... 15 chars omitted)
      child 0, instance_id: string
      child 1, file: string
      child 2, sha256: string
      child 3, turns: int64
      child 4, exit_status: string
      child 5, source_sha256: string
instance_count: int64
request_count: int64
excluded_instance_ids: list<item: string>
  child 0, item: string
source_provenance: struct<original_manifest_sha256: string, source_dataset: string, capture_backend: string, capture_mo (... 93 chars omitted)
  child 0, original_manifest_sha256: string
  child 1, source_dataset: string
  child 2, capture_backend: string
  child 3, capture_model: string
  child 4, capture_date: timestamp[s]
  child 5, adapter_commit: null
  child 6, publication_transform: string
turns: null
exit_status: string
instance_id: string
to
{'instance_id': Value('string'), 'exit_status': Value('string'), 'turns': List({'turn': Value('int64'), 'request': {'messages': List(Json(decode=True)), 'model': Value('string'), 'max_tokens': Value('int64'), 'parallel_tool_calls': Value('bool'), 'temperature': Value('float64'), 'tools': List({'type': Value('string'), 'function': {'name': Value('string'), 'description': Value('string'), 'parameters': {'type': Value('string'), 'properties': {'command': {'type': Value('string'), 'description': Value('string')}}, 'required': List(Value('string'))}}}), 'top_p': Value('float64'), 'enable_thinking': Value('bool'), 'preserve_thinking': Value('bool'), 'chat_template_kwargs': {'clear_thinking': Value('bool')}}, 'response': {'id': Value('string'), 'object': Value('string'), 'created': Value('int64'), 'model': Value('string'), 'choices': List({'index': Value('int64'), 'message': {'role': Value('string'), 'content': Value('string'), 'refusal': Value('null'), 'annotations': Value('null'), 'audio': Value('null'), 'function_call': Value('null'), 'tool_calls': List({'id': Value('string'), 'type': Value('string'), 'function': {'name': Value('string'), 'arguments': Value('string')}}), 'reasoning': Value('string')}, 'logprobs': Value('null'), 'finish_reason': Value('string'), 'stop_reason': Value('int64'), 'token_ids': Value('null'), 'routed_experts': Value('null')}), 'service_tier': Value('null'), 'system_fingerprint': Value('string'), 'usage': {'prompt_tokens': Value('int64'), 'total_tokens': Value('int64'), 'completion_tokens': Value('int64'), 'prompt_tokens_details': {'cached_tokens': Value('int64'), 'created_cache_tokens': Value('int64'), 'multimodal_tokens': Value('null')}}, 'prompt_logprobs': Value('null'), 'prompt_token_ids': Value('null'), 'prompt_text': Value('null'), 'kv_transfer_params': Value('null'), 'ec_transfer_params': Value('null'), 'metrics': Value('null')}, 'think_time_s': Value('float64'), 'completion_tokens': Value('int64'), 'source_request_id': Value('string'), 'source_sha256': 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
              schema: string
              timing_boundary: string
              timing_quality: string
              timing_error_bound_s: null
              timing_caveat: string
              selection: string
              source_run_config: struct<batch_size: int64, chain_cache: bool, cost_limit: double, cost_tracking: string, dataset: str (... 344 chars omitted)
                child 0, batch_size: int64
                child 1, chain_cache: bool
                child 2, cost_limit: double
                child 3, cost_tracking: string
                child 4, dataset: string
                child 5, dataset_rows_sha256: string
                child 6, enable_thinking: bool
                child 7, instance_ids: list<item: string>
                    child 0, item: string
                child 8, max_tokens: int64
                child 9, model: string
                child 10, preserve_thinking: bool
                child 11, seed: null
                child 12, seed_control: string
                child 13, served_model_name: string
                child 14, server_manifest_sha256: string
                child 15, split: string
                child 16, step_limit: int64
                child 17, temperature: double
                child 18, top_p: double
                child 19, wall_time_limit_seconds: int64
              agents: list<item: struct<instance_id: string, file: string, sha256: string, turns: int64, exit_status: stri (... 27 chars omitted)
                child 0, item: struct<instance_id: string, file: string, sha256: string, turns: int64, exit_status: string, source_ (... 15 chars omitted)
                    child 0, instance_id: string
                    child 1, file: string
                    child 2, sha256: string
                    child 3, turns: int64
                    child 4, exit_status: string
                    child 5, source_sha256: string
              instance_count: int64
              request_count: int64
              excluded_instance_ids: list<item: string>
                child 0, item: string
              source_provenance: struct<original_manifest_sha256: string, source_dataset: string, capture_backend: string, capture_mo (... 93 chars omitted)
                child 0, original_manifest_sha256: string
                child 1, source_dataset: string
                child 2, capture_backend: string
                child 3, capture_model: string
                child 4, capture_date: timestamp[s]
                child 5, adapter_commit: null
                child 6, publication_transform: string
              turns: null
              exit_status: string
              instance_id: string
              to
              {'instance_id': Value('string'), 'exit_status': Value('string'), 'turns': List({'turn': Value('int64'), 'request': {'messages': List(Json(decode=True)), 'model': Value('string'), 'max_tokens': Value('int64'), 'parallel_tool_calls': Value('bool'), 'temperature': Value('float64'), 'tools': List({'type': Value('string'), 'function': {'name': Value('string'), 'description': Value('string'), 'parameters': {'type': Value('string'), 'properties': {'command': {'type': Value('string'), 'description': Value('string')}}, 'required': List(Value('string'))}}}), 'top_p': Value('float64'), 'enable_thinking': Value('bool'), 'preserve_thinking': Value('bool'), 'chat_template_kwargs': {'clear_thinking': Value('bool')}}, 'response': {'id': Value('string'), 'object': Value('string'), 'created': Value('int64'), 'model': Value('string'), 'choices': List({'index': Value('int64'), 'message': {'role': Value('string'), 'content': Value('string'), 'refusal': Value('null'), 'annotations': Value('null'), 'audio': Value('null'), 'function_call': Value('null'), 'tool_calls': List({'id': Value('string'), 'type': Value('string'), 'function': {'name': Value('string'), 'arguments': Value('string')}}), 'reasoning': Value('string')}, 'logprobs': Value('null'), 'finish_reason': Value('string'), 'stop_reason': Value('int64'), 'token_ids': Value('null'), 'routed_experts': Value('null')}), 'service_tier': Value('null'), 'system_fingerprint': Value('string'), 'usage': {'prompt_tokens': Value('int64'), 'total_tokens': Value('int64'), 'completion_tokens': Value('int64'), 'prompt_tokens_details': {'cached_tokens': Value('int64'), 'created_cache_tokens': Value('int64'), 'multimodal_tokens': Value('null')}}, 'prompt_logprobs': Value('null'), 'prompt_token_ids': Value('null'), 'prompt_text': Value('null'), 'kv_transfer_params': Value('null'), 'ec_transfer_params': Value('null'), 'metrics': Value('null')}, 'think_time_s': Value('float64'), 'completion_tokens': Value('int64'), 'source_request_id': Value('string'), 'source_sha256': 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.

SWE-lite trace100

The default main revision contains the appendable longest-turn 100-agent trace from a GLM-4.7-Flash / upstream vLLM MiniSWE run on SWE-bench Lite. It has 8,000 recorded model requests: 100 agents with 80 turns each. This is a fixed workload for inference-serving comparisons, not a SWE-bench quality score or a current performance result.

Selection

The source run contains 300 SWE-bench Lite test instances. Trajectory length means the number of successful recorded model requests. Of the 300, 157 reach the 80-turn limit. Ranking uses descending turn count and breaks ties by the source run's frozen instances.txt order. One candidate, matplotlib__matplotlib-23476, is excluded because the agent discarded an assistant response after an invalid tool call. The following prompt therefore cannot append to that response for Chain Cache replay. The next 80-turn candidate, scikit-learn__scikit-learn-25638, fills the 100th place. This exclusion is recorded in trace/manifest.json.

Files and provenance

trace/manifest.json: schema, selection, agent file hashes, request count, source metadata, and timing limitations.
trace/agent_000.json through trace/agent_099.json: one complete trajectory per file, with ordered original requests/responses, completion-token counts, and estimated inter-turn gaps. Private machine paths, local API addresses, commands, and source-file paths were removed from the published metadata. Request/response payloads, token counts, gaps, selected IDs, and source-content hashes were retained. That repackaging changes the manifest and agent file hashes. Pin one published revision for every method in a comparison.

Timing and replay limits

Legacy inter-turn gaps are estimated from server-log timestamps and recorded request durations. Logging overhead is unknown, so these are not exact client timings. Gaps can include tool, network, and retry waits. The corpus does not contain newly measured serving latency or throughput.

The trajectories contain original model responses and later prompts. A plain replay can fix the recorded completion-token count, but its newly generated text is not guaranteed to match later recorded prompts. A comparison that depends on exact prefix or Chain Cache continuity must validate and force the recorded nonterminal continuations at the model-token boundary. Tools are not executed during serving replay.

Download the raw snapshot to a data disk and pin a revision:

from huggingface_hub import snapshot_download

snapshot_download(
    repo_id="JitaiHao/SWE-lite-trace100",
    repo_type="dataset",
    revision="<dataset-commit-sha>",
    local_dir="<dataset-directory>",
)

Source tasks come from SWE-bench Lite. Trajectories can include generated code, repository excerpts, and tool outputs; treat them as untrusted data and do not execute embedded commands. The existing Apache-2.0 dataset declaration is retained and does not replace applicable third-party source-code licenses. A limited secret-pattern scan found no matches; it is not an exhaustive review of all possible sensitive content.

Downloads last month: 130

Total file size:

809 MB