Synchronize main evaluation program pairs

README.md

@@ -27,58 +27,46 @@ size_categories:
 
 # Program Pair Data
 
-This revision contains exactly 900 C++, 900 Python and 900 Java pairs, with no
-training/validation splits and no pre-expanded prompt datasets. Superseded
-training files, evaluation prompts and earlier base snapshots are not active
-files in this revision. Use an explicit language configuration; each has one
-`test` split. The previous mixed-schema `benchmark` configuration is retired.
+This dataset contains 2,700 main-evaluation program pairs. Each language has
+one `test` split with 900 pairs.
 
-`slow_code` and `fast_code` are the exact model-facing, formatted,
-comment-cleaned implementations. They include actual program text, not hashes
-or an index alone. `pair_id` and `problem_id` identify the comparison and task.
-Python/Java additionally retain `measured_slow_code` and `measured_fast_code`
-so that presentation changes do not overwrite the original measured program.
-Variable/method identifiers are not globally renamed. A small audited set of
-source-revealing Java class names was neutralized in the display view.
+| Configuration | Pairs | Problems | Maximum pairs per problem |
+| --- | ---: | ---: | ---: |
+| `cpp` | 900 | 900 | 1 |
+| `python` | 900 | 221 | 5 |
+| `java` | 900 | 320 | 5 |
 
-## Label and workload scope
+Each language contains 300 pairs in each slow-to-fast execution-cost ratio
+range: [2, 4), [4, 8), and [8, infinity).
 
-- C++ inherits the published PIE aggregate performance supervision. Four
-  quarantined comparisons were replaced with verified candidates from the
-  existing PIE pool. The inherited aggregate is not claimed to be a newly
-  computed equal-case geometric mean. The workload describes the available
-  published input pool; its exact historical timing subset is not recovered.
-- Python uses CPU instruction counts; Java uses warmed execution elapsed
-  time. For these languages `speedup` is the equal-weight geometric mean of
-  the per-input slow/fast representative measurement ratios over the frozen
-  common input set, not a ratio of arithmetic means over heterogeneous inputs.
-- Java includes both sides' whole-workload, across-JVM RSD in percent; both
-  are below 5% in this release. This is a stability filter, not a confidence
-  interval or a guarantee about every individual input.
-- Each language contains 300 comparisons in each speedup range [2,4), [4,8),
-  and [8,infinity). Python/Java have at most five pairs per task; inherited
-  C++ has a maximum of six. These policies are not retroactively equated.
-- `workload.description` and `workload.input_format` contain input context,
-  not runtime labels. Python/Java text uses the completed semantic review of
-  actual frozen common inputs. Workload metadata did not construct labels.
+## Fields
 
-The `reproduction` entries in Python/Java preserve relative identifiers for
-the separate input and per-run measurement archive maintained with the
-release. Those large archive contents are **not embedded in these three
-JSONL files or uploaded by this base-dataset update**. A path/hash is not an
-actual measurement or recovered stdin. `performance_label_scope` distinguishes
-measurements of original programs from display-equivalence validation.
+- `language`, `problem_id`, `pair_id`: language, task and comparison identifiers.
+- `slow_code`, `fast_code`: exact model-facing program text, ordered by the saved efficiency labels.
+- `slow_code_id`, `fast_code_id`: SHA-256 hashes of the displayed source text.
+- `measured_slow_code_id`, `measured_fast_code_id`: SHA-256 identifiers of the measured source programs, which can differ from the displayed source.
+- `slow_id`, `fast_id`: measured implementation identifiers.
+- `speedup`: the saved slow-to-fast execution-cost ratio.
+- `measurement_metric`: PIE released gem5 runtime for C++, CPU instruction count for Python, or elapsed execution time for Java.
+- `stratum`, `sampling_key`: ratio stratum and deterministic sampling key.
+- `slow_time`, `fast_time`: released aggregate costs, present for C++.
+
+The Python and Java ratios aggregate per-input measurement ratios; they are
+not reconstructed as ratios of aggregate times. Test inputs, expected
+outputs, per-run measurements and original measured-source text are not
+included. This dataset provides the program pairs and saved efficiency
+labels used in the main evaluation.
 
 ## Loading
 
 ```python
 from datasets import load_dataset
+
 data = load_dataset("JinNian0072/program-pair-data", "python", split="test")
 assert len(data) == 900
 print(data[0]["slow_code"], data[0]["fast_code"])
 ```
 
-Use `cpp` or `java` for the other languages. See `RELEASE.json` for file hashes
-and the exact base revision. No model answers, API keys or private machine
-paths are part of this dataset. Upstream data retain their respective terms;
-this card does not grant a new license to upstream contest code or data.
+Use `cpp` or `java` for the other languages. The JSONL files are uncompressed.
+`RELEASE.json` contains record counts and file checksums. Upstream code and
+data retain their respective terms.

RELEASE.json

@@ -1 +1,26 @@
-{"files": {"final_benchmark_uniform/cpp_seed42_sanitized.jsonl": {"bytes": 7198529, "rows": 900, "sha256": "8958196202a359271d0b64a2d57da72a4d4c294c8c526dca1b00d0c86aadcf98", "source_sha256": "710478a6212e62fefe0b9523b4949f47addccdc48311f165be7a1858986b0863", "strata": {"2-4": 300, "4-8": 300, "8+": 300}}, "final_benchmark_uniform/java_seed42_sanitized_process.jsonl": {"bytes": 16501806, "rows": 900, "sha256": "8153cd50cb8b160dadf47c95d7ad935b53a3765c6b9886a0dbb982b670fa137a", "source_sha256": "649a9a91ba5812b50c97b9d4379caa5e4bd8278ae40ee62cf1f1a38f2a0e568c", "strata": {"2-4": 300, "4-8": 300, "8+": 300}}, "final_benchmark_uniform/python_seed42_sanitized.jsonl": {"bytes": 6022941, "rows": 900, "sha256": "076f4608822e53876663d697d0cad15cfefa9673502433487e7761cde6092e2b", "source_sha256": "a4e852d2a7b54f846bf27ca4256e8f7dd3a04dd7f06d06f1d1466fef9d030755", "strata": {"2-4": 300, "4-8": 300, "8+": 300}}}, "model_answers_included": false, "pairs_per_language": 900, "release_id": "measured-core-20261022-cpp4-java-rsd5", "total_pairs": 2700, "training_and_validation_included": false, "workload_review_sha256": "30d6d0e0a8503397010ccba5de40dffd43755567ba1cd077d11d85238071aa4d"}
+{
+  "total_pairs": 2700,
+  "pairs_per_language": 900,
+  "problems_per_language": {
+    "cpp": 900,
+    "python": 221,
+    "java": 320
+  },
+  "files": {
+    "final_benchmark_uniform/cpp_seed42_sanitized.jsonl": {
+      "bytes": 4071844,
+      "rows": 900,
+      "sha256": "fc7a12fda6a8901695344cbfab7adf1b42047d9a85eeebc08a98a65931d11247"
+    },
+    "final_benchmark_uniform/python_seed42_sanitized.jsonl": {
+      "bytes": 1669655,
+      "rows": 900,
+      "sha256": "7db637509b7b3ac9a443e71fb31964710b4077109afadee61934d4e3db5d5900"
+    },
+    "final_benchmark_uniform/java_seed42_sanitized_process.jsonl": {
+      "bytes": 6622661,
+      "rows": 900,
+      "sha256": "2729c673fd3ab39a39a35f8e6e1459a8db40f62ffb3b4f9c70a55ffa5ce0cb94"
+    }
+  }
+}

final_benchmark_uniform/java_seed42_sanitized_process.jsonl

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8153cd50cb8b160dadf47c95d7ad935b53a3765c6b9886a0dbb982b670fa137a
-size 16501806
+oid sha256:2729c673fd3ab39a39a35f8e6e1459a8db40f62ffb3b4f9c70a55ffa5ce0cb94
+size 6622661