Publish frozen 1024-task, 122-repository OPSD PI dataset

.gitattributes

@@ -58,3 +58,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 # Video files - compressed
 *.mp4 filter=lfs diff=lfs merge=lfs -text
 *.webm filter=lfs diff=lfs merge=lfs -text
+data/train.jsonl filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,76 @@
+---
+language:
+- en
+pretty_name: OPSD PI MultiRepo Student-Fail 1024
+size_categories:
+- 1K<n<10K
+tags:
+- opsd
+- privileged-information
+- software-engineering
+- code
+configs:
+- config_name: default
+  data_files:
+  - split: train
+    path: data/train.jsonl
+---
+
+# OPSD PI MultiRepo — Student-Fail
+
+本仓库公开发布之前已生成的多仓库 PI，供 OPSD 训练下载。**1024 条任务、122 个仓库，每题包含 5 个阶段 × 3 种强度，共 15 条 PI 内容。**
+
+## 数据身份
+
+从历史 student 的确认 unresolved 任务中选择，1024 条全部为确认未解出；不包含 infrastructure-error 筛选结果。
+
+- SWE-ReBench-V2：824 条。
+- SWE-Gym：200 条。
+- 抽样 seed：42。原始选择日期：2026-09-22；新增 ReBench PI 合成批次：2026-09-24；本次公开发布日期：2026-10-15。
+- 历史筛选 student：`jiaxingx/privilege-code-opsd-ckpts/qwen35-9b-expert-sft-131k-lora64-block28-2996448/checkpoint-best`，agent：`mini-swe-agent 2.3.0`。这个标识记录历史模型身份，不保证该模型当前可公开下载。
+- 另一组：[LSW142857/OPSD-PI-MultiRepo-All-Outcomes-1024](https://huggingface.co/datasets/LSW142857/OPSD-PI-MultiRepo-All-Outcomes-1024)。
+
+两组是替代实验组，不是 train/test 划分。它们重叠 **460 个 instance ID、466 个相同 problem statement**；各组内部没有重复 ID 或问题文本。新 SFT 模型的成功/失败结果可能不同，不能把历史 student-fail 当作新模型已重新评测的结论。
+
+## 文件与下载
+
+- `data/train.jsonl`：完整训练数据，顶层字段 `label/messages/metadata/prompt`。
+- `data/manifest.json`：来源、抽样信息、PI 校验警告、文件哈希、逐题 PI 来源标识。
+- `data/repo_counts.json`：122 个仓库的逐仓库条数。
+- `SHA256SUMS`：发布文件校验值。
+
+```bash
+hf download LSW142857/OPSD-PI-MultiRepo-Student-Fail-1024 --repo-type dataset --local-dir ./OPSD-PI-MultiRepo-Student-Fail-1024
+cd OPSD-PI-MultiRepo-Student-Fail-1024
+sha256sum -c SHA256SUMS
+wc -l data/train.jsonl
+```
+
+为严格复现，请将 `--revision <commit SHA>` 加入下载命令，固定仓库版本。
+
+JSONL 可直接逐行读取，或使用 `datasets.load_dataset("json", data_files="data/train.jsonl", split="train")`。OPSD 使用 `prompt` 作为公开任务输入，并按配置从下列路径选择 teacher PI：
+
+```text
+metadata.privileged_info.pi_options
+  EXPLORE / REPRODUCE / DIAGNOSE / EDIT / VERIFY
+    WEAK / MEDIUM / STRONG
+      content / evidence_ids / name
+```
+
+metadata 另含 `instance_id`、`benchmark`、`repo`、`base_commit`、`repository_path`、`image_name`、`problem_statement`。环境字段是平铺格式。此仓库不包含模型权重、容器镜像、student 轨迹、PI 合成完整轨迹或原始 gold/test patch 字段。
+
+## PI 来源和质量边界
+
+SWE-ReBench-V2 PI 来自历史 `deepseek/deepseek-flash` 合成批次中已经 accepted 的输出；SWE-Gym 复用已有 schema 1.1 PI。合成目录名称中虽有 medium，本发布包含 WEAK、MEDIUM、STRONG 三档，未丢弃其他阶段或档位。
+
+本版本 `data/train.jsonl` 与 2026-09-30 clip122 冻结训练文件逐字节一致。公开 manifest 已去除旧机器私有路径，因此其 SHA 与旧 manifest 不同；原 manifest SHA 保存在 `original_local_training_manifest_sha256`。
+
+训练 JSONL SHA-256：`80130f0317a828960a2db881fa674f1370c3e41f7e86823a40488b5071b6bc96`。
+
+保留历史合成批次的接受策略和校验警告：本组有 824 条任务记录了警告，共 13071 条警告；详情在 manifest 的 `provenance[].validation_warnings`。这些包括阶段纯度、与测试代码的短片段重合及证据审计警告。数据通过结构和完整性检查不等于全部 PI 语义都正确。PI 含有解题特权信息，只应按实验设计供 teacher 使用，不应直接拼入无 PI 的 student 或评测输入。
+
+## 来源与使用条款
+
+任务来自 SWE-Gym 和 SWE-ReBench-V2 及其各自上游代码仓库。每条记录保留 `benchmark/repo/base_commit` 用于追溯。此发布不为上游内容重新授予统一许可证；使用和再分发时应分别遵守相应原始数据、代码与生成服务条款。
+
+只发布完成训练所需的任务和 PI 及可移植来源说明；旧机器路径不构成下载依赖。manifest 中 PI artifact hash 指向历史生成文件，相关原始生成文件没有另行打包进本仓库。

SHA256SUMS

@@ -0,0 +1,4 @@
+bb6efdfa72e420c7d095389a826508f03de1f633c290b3d500217fc0ac6c3fae  README.md
+80130f0317a828960a2db881fa674f1370c3e41f7e86823a40488b5071b6bc96  data/train.jsonl
+983ba5eadc6db2f54b133803a361746189c152f841678d243e1268335f3162f9  data/manifest.json
+e2e12b9e6f6e44fc72439d4332dd377f7ac4583daac77e2d9c98ae2a870465ed  data/repo_counts.json

data/repo_counts.json

@@ -0,0 +1,124 @@
+{
+  "AgnostiqHQ/covalent": 5,
+  "Axelrod-Python/Axelrod": 8,
+  "Azure/WALinuxAgent": 4,
+  "BerriAI/litellm": 6,
+  "CORE-GATECH-GROUP/serpent-tools": 5,
+  "CS-SI/eodag": 4,
+  "CTFd/CTFd": 8,
+  "GoogleCloudPlatform/cloud-sql-python-connector": 1,
+  "HypothesisWorks/hypothesis": 35,
+  "Kozea/WeasyPrint": 5,
+  "MartinThoma/flake8-simplify": 2,
+  "Materials-Consortia/optimade-python-tools": 2,
+  "MyreMylar/pygame_gui": 7,
+  "NeurodataWithoutBorders/pynwb": 30,
+  "OpenMDAO/OpenMDAO": 34,
+  "PennyLaneAI/pennylane": 34,
+  "Project-MONAI/MONAI": 34,
+  "PyCQA/docformatter": 3,
+  "PyCQA/pyflakes": 4,
+  "Pylons/waitress": 5,
+  "Pyomo/pyomo": 5,
+  "Python-Markdown/markdown": 5,
+  "Qiskit/qiskit": 7,
+  "Qiskit/qiskit-terra": 35,
+  "ResearchObject/ro-crate-py": 5,
+  "SpikeInterface/spikeinterface": 5,
+  "Textualize/rich": 3,
+  "TobikoData/sqlmesh": 5,
+  "XKNX/xknx": 3,
+  "aeon-toolkit/aeon": 6,
+  "aeye-lab/pymovements": 3,
+  "aio-libs/aiohttp": 34,
+  "alteryx/evalml": 5,
+  "alteryx/woodwork": 2,
+  "amaranth-lang/amaranth": 6,
+  "apiflask/apiflask": 3,
+  "aristanetworks/anta": 5,
+  "aws-cloudformation/cfn-lint": 35,
+  "aws-cloudformation/cfn-python-lint": 5,
+  "bokeh/bokeh": 4,
+  "boto/botocore": 7,
+  "brazilian-utils/brutils-python": 5,
+  "caikit/caikit": 3,
+  "callowayproject/bump-my-version": 3,
+  "celery/celery": 7,
+  "cloudpipe/cloudpickle": 4,
+  "codezonediitj/pydatastructs": 7,
+  "cogent3/cogent3": 8,
+  "conan-io/conan": 6,
+  "cta-observatory/ctapipe": 4,
+  "cuthbertLab/music21": 8,
+  "dask/dask": 11,
+  "dynaconf/dynaconf": 6,
+  "ecmwf/earthkit-data": 5,
+  "elastic/curator": 3,
+  "enthought/traits": 10,
+  "facebookresearch/hydra": 7,
+  "getmoto/moto": 35,
+  "hdmf-dev/hdmf": 7,
+  "hikari-py/hikari": 3,
+  "huggingface/huggingface_hub": 1,
+  "ipython/traitlets": 6,
+  "iterative/dvc": 22,
+  "jazzband/pip-tools": 4,
+  "jelmer/dulwich": 3,
+  "jlowin/fastmcp": 7,
+  "kayak/pypika": 3,
+  "keras-team/keras": 7,
+  "litestar-org/polyfactory": 4,
+  "meltano/sdk": 3,
+  "mie-lab/trackintel": 8,
+  "mirumee/ariadne-codegen": 3,
+  "modin-project/modin": 3,
+  "mpmath/mpmath": 5,
+  "narwhals-dev/narwhals": 7,
+  "natcap/invest": 5,
+  "neogeny/TatSu": 7,
+  "networkx/networkx": 9,
+  "pallets/click": 17,
+  "pallets/jinja": 1,
+  "pallets/werkzeug": 2,
+  "pandas-dev/pandas": 35,
+  "peopledoc/workalendar": 3,
+  "ploomber/jupysql": 6,
+  "pvlib/pvlib-python": 1,
+  "py-pdf/PyPDF2": 6,
+  "pybamm-team/PyBaMM": 35,
+  "pyccel/pyccel": 34,
+  "pydantic/pydantic": 8,
+  "pyfar/pyfar": 2,
+  "pyjanitor-devs/pyjanitor": 6,
+  "pymc-devs/pytensor": 7,
+  "pymodbus-dev/pymodbus": 4,
+  "pypa/packaging": 2,
+  "pypa/pipx": 4,
+  "pypa/setuptools": 4,
+  "pytask-dev/pytask": 6,
+  "pytest-dev/pyfakefs": 3,
+  "python-attrs/attrs": 4,
+  "python-babel/babel": 1,
+  "python/mypy": 35,
+  "rhayes777/PyAutoFit": 8,
+  "robotpy/cxxheaderparser": 3,
+  "rochacbruno/dynaconf": 5,
+  "s-knibbs/dataclasses-jsonschema": 3,
+  "samuelcolvin/pydantic": 9,
+  "scikit-build/scikit-build-core": 3,
+  "scoutapp/scout_apm_python": 6,
+  "scrapy/scrapy": 3,
+  "sdv-dev/RDT": 33,
+  "slackapi/python-slack-sdk": 2,
+  "stac-utils/pystac": 2,
+  "stfc/PSyclone": 6,
+  "theacodes/nox": 2,
+  "timothycrosley/isort": 2,
+  "tox-dev/pipdeptree": 5,
+  "tox-dev/sphinx-autodoc-typehints": 5,
+  "vacanza/holidays": 5,
+  "vyperlang/vyper": 7,
+  "wntrblm/nox": 5,
+  "wtforms/wtforms": 3,
+  "yt-dlp/yt-dlp": 8
+}

data/train.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:80130f0317a828960a2db881fa674f1370c3e41f7e86823a40488b5071b6bc96
+size 20383563