Hugging Face's logo

Datasets:
ion1111
/
beam-1m-swe-benchmark

Tasks:
Text Generation
Languages:
English
Size:
n<1K
Tags:
code
autonomous-agent
swe-bench
self-healing
long-context
1M-context
License:
Dataset card Files
xet
 Community
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
tier: string
input_tokens: int64
reasoning_tokens: int64
latency_seconds: double
resolved: bool
reasoning_excerpt: string
patch_excerpt: string
benchmarks: list<item: struct<scenario_id: string, domain: string, difficulty: string, turns_to_green: int64, re (... 101 chars omitted)
  child 0, item: struct<scenario_id: string, domain: string, difficulty: string, turns_to_green: int64, reasoning_tok (... 89 chars omitted)
      child 0, scenario_id: string
      child 1, domain: string
      child 2, difficulty: string
      child 3, turns_to_green: int64
      child 4, reasoning_tokens: int64
      child 5, total_tokens: int64
      child 6, latency_seconds: double
      child 7, status: string
      child 8, resolved: bool
model: string
benchmark_timestamp: timestamp[s]
to
{'benchmark_timestamp': Value('timestamp[s]'), 'model': Value('string'), 'benchmarks': List({'scenario_id': Value('string'), 'domain': Value('string'), 'difficulty': Value('string'), 'turns_to_green': Value('int64'), 'reasoning_tokens': Value('int64'), 'total_tokens': Value('int64'), 'latency_seconds': Value('float64'), 'status': Value('string'), 'resolved': Value('bool')})}
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 2951, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2461, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2486, 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 547, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 430, 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
              tier: string
              input_tokens: int64
              reasoning_tokens: int64
              latency_seconds: double
              resolved: bool
              reasoning_excerpt: string
              patch_excerpt: string
              benchmarks: list<item: struct<scenario_id: string, domain: string, difficulty: string, turns_to_green: int64, re (... 101 chars omitted)
                child 0, item: struct<scenario_id: string, domain: string, difficulty: string, turns_to_green: int64, reasoning_tok (... 89 chars omitted)
                    child 0, scenario_id: string
                    child 1, domain: string
                    child 2, difficulty: string
                    child 3, turns_to_green: int64
                    child 4, reasoning_tokens: int64
                    child 5, total_tokens: int64
                    child 6, latency_seconds: double
                    child 7, status: string
                    child 8, resolved: bool
              model: string
              benchmark_timestamp: timestamp[s]
              to
              {'benchmark_timestamp': Value('timestamp[s]'), 'model': Value('string'), 'benchmarks': List({'scenario_id': Value('string'), 'domain': Value('string'), 'difficulty': Value('string'), 'turns_to_green': Value('int64'), 'reasoning_tokens': Value('int64'), 'total_tokens': Value('int64'), 'latency_seconds': Value('float64'), 'status': Value('string'), 'resolved': Value('bool')})}
              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.

⚡ Beam-1M-SWE-Benchmark

Model Live Space License

This dataset provides the first empirical benchmark of long-context autonomous software engineering on reflection/Beam-1.0-501B-A23B across scaling context depths from 10,000 to 1,001,135 tokens.

Created by Ion Vovos for the Reflection AI × Hugging Face $20,000 Beam Build Challenge.

📊 Dataset Contents

  1. swe_1m_frontier_evals.jsonl: Empirical evaluations of needle-in-a-codebase security bug resolution across 5 scaling context tiers:

    • 10K tokens (Module Tier): 9,138 tokens · Latency 2.54s · Reasoning: 198 tokens · Result: PASS
    • 50K tokens (Package Tier): 49,077 tokens · Latency 2.17s · Reasoning: 53 tokens · Result: PASS
    • 150K tokens (Subsystem Tier): 149,314 tokens · Latency 8.35s · Reasoning: 1,874 tokens · Result: PASS
    • 500K tokens (Multi-Package Tier): 500,275 tokens · Latency 26.88s · Needle found, hit 2K token completion ceiling
    • 1M tokens (Frontier Tier): 1,001,135 tokens · Latency 28.80s · Reasoning: 896 tokens · Result: PASS

  2. swe_bench_10_scenarios.jsonl: 10 real-world software engineering self-healing repair traces spanning:

    • Web Auth & JWT Expiration
    • Concurrency & Async Locks
    • SQL Pipelines & Type Coercion
    • Event Bus Cache Invalidation
    • Topological Dependency Graph Sorting
    • Distributed Token Bucket Rate Limiting
    • WebSocket Exponential Backoff & Jitter
    • Async Connection Pool Leak Recovery
    • Schema Evolution & Backward Compatibility
    • Memory-Bounded Chunked Streams with Backpressure

  3. results_summary.json: Aggregated pass rates, turn counts, reasoning token counts, and wall-clock latencies.

🔬 Key Empirical Discoveries

  1. Needle Retrieval at 1M Tokens: Beam-1.0-501B-A23B successfully isolated a security regression buried in 1,001,135 prompt tokens and synthesized a working patch in 28.8 seconds.
  2. The Long-Context Reasoning Budget Trap: At 500k+ tokens, Beam's internal Chain-of-Thought consumes ~1,900+ reasoning tokens simply orienting across the codebase. Developers setting max_tokens=2000 risk truncation before code emission. Setting max_tokens >= 4000 allows full resolution.
  3. Sub-10s Repair Latency: On medium packages (<150k tokens), average repair latency is 5.55 seconds with 100% test-driven self-healing pass rate.

💻 How to Use 

from datasets import load_dataset

dataset = load_dataset("ion1111/beam-1m-swe-benchmark", data_files="swe_1m_frontier_evals.jsonl")
print(dataset[0])

Citation & Authorship

Built with AIOS by Ion Vovos for the Reflection AI × Hugging Face 2026 Challenge.
Downloads last month
20
Total file size:
94.8 kB

Space using ion1111/beam-1m-swe-benchmark 1