Datasets:

spadeMIA
/

GovReport_Corpus_1024_2040

text large_stringlengths 6.06k 12.5k	label int64 1 1
The Dodd-Frank Wall Street Reform and Consumer Protection Act (Dodd-Frank Act; P.L. 111-203 ) was signed into law on July 21, 2010, as a response to the 2008-2009 financial crisis. The act mandated extensive financial regulatory reform. Section 953(b) of the act, the so-called pay ratio provision, requires the Securiti...	1
Immunizations are widely considered one of the leading public health achievements of the 20th century. Mandatory immunization programs have eradicated polio and smallpox in the United States and reduced the number of deaths from several childhood diseases, such as measles, to near zero. A consistent supply of many diff...	1
PTSD can develop following exposure to life-threatening events, natural disasters, terrorist incidents, serious accidents, or violent personal assaults like rape. PTSD is the most prevalent mental disorder arising from combat. People who experience stressful events often relive the experience through nightmares and fla...	1
According to the State Department, no country poses a more immediate narcotics threat to the United States than Mexico. For over 20 years, the United States has supported the Mexican government in its counternarcotics efforts and has provided assistance to develop and strengthen the Mexican government in its law enforc...	1
The West Valley site, about 30 miles southeast of Buffalo, includes an approximately 200-acre area of nuclear operations within a 3,300-acre area owned by the state of New York. (See fig. 1.) The facility began construction in 1963 as the first—and ultimately the only—commercial spent fuel reprocessing plant to be oper...	1
The drawdown of equipment and personnel from Iraq is a highly complex operation of significant magnitude. According to DOD, as of February 2010, there were approximately 98,100 U.S. military personnel in Iraq, spread among 228 bases throughout the country. Additionally, as of February 2010, the Army reported that there...	1
The HVBP program affects Medicare payments to approximately 3,000 acute care hospitals for the inpatient services provided to Medicare beneficiaries. Hospitals are included in the HVBP program if they are paid through Medicare’s Inpatient Prospective Payment System. Thus, hospitals not paid through this system, such as...	1
State is authorized to designate acceptance facilities—in addition to its own passport offices—to provide passport execution services to the American public. The majority of passport applications are submitted at acceptance facilities nationwide; these include post offices; federal, state, and probate courts; some publ...	1
RUS, EDA, Reclamation, and the Corps each have distinct missions and fund rural water supply and wastewater projects under separate programs and congressional authorizations. Furthermore, each agency has its own definition of what constitutes a rural area and a unique organizational structure to implement its programs....	1
Humanitarian parole—in the context of immigration—refers to official permission for an otherwise inadmissible alien to legally enter the United States temporarily. This includes aliens required to have a visa to visit or immigrate to the United States who are unable to obtain one, either due to ineligibility or urgent ...	1
A patent box provides a lower tax rate on income from patents. In some cases, the lower rate applies to income from other intellectual property. Such a subsidy is sometimes referred to as an innovation or IP box. In this report, the term "patent box," the most popular name, will be used to refer to all of the special t...	1
The Omnibus Reconciliation Act of 1982 ( P.L. 97-253 ; September 8, 1982) suspended previously existing permanent law pertaining to cost-of-living adjustments (COLAs) for federal civilian and military retirees. Much of the legislative and executive branch activity related to military retirement COLAs since 1983 has inv...	1
Demand for GAO’s analysis and advice remains strong across the Congress. During the past 3 years, GAO has received requests or mandated work from all of the standing committees of the House and the Senate and over 90 percent of their subcommittees. In fiscal year 2007, GAO received over 1,200 requests for studies. This...	1
During and after the breakup of the Soviet Union in 1991, the North Caucasus area of Russia experienced substantial disorder. Among such disorder, breakaway conflicts in neighboring Georgia rallied some North Caucasians to support their efforts; Chechen separatism gained ground, contributing to the breakup of the then-...	1
The DI program was established in 1956 to provide monthly cash benefits to individuals unable to work because of severe long-term disability. Cash benefits are payable monthly, as long as the worker remains eligible for benefits, until the worker reaches full retirement age or dies. In fiscal year 2011, more than 10 mi...	1
The phrase "Glass-Steagall" generally refers to the separation of commercial banking from investment banking. In this context, commercial banking refers to the activities engaged in by depository institutions, which this report also will refer to as "banks" or "commercial banks." In contrast, investment banking refers ...	1
The Food, Agriculture, Conservation, and Trade Act of 1990 (the 1990 Farm Bill) required, among other things, that the Secretary give priority to beginning farmers in purchasing inventory farmland—properties that have come into government ownership through voluntary conveyance or foreclosure. It also expressed the sens...	1
DHS Acquisition Management Directive 102-01 (MD 102) and an accompanying instruction manual establish the department’s policies and processes for managing major acquisition programs. While DHS has had an acquisition management policy in place since October 2004, the department issued the initial version of MD 102 in 20...	1

End of preview. Expand in Data Studio

GovReport MIA fine-tuning corpus

This card documents the split layout for spadeMIA/GovReport_Corpus_1024_2040, a GovReport corpus used for membership-inference attack (MIA) experiments on fine-tuned language models.

Membership labels are defined relative to the fine-tuning population. train is the only split used for fine-tuning, and every row has label = 1. test remains held out, and every row has label = 0. evaluation is a fixed, balanced MIA candidate set. Fine-tune on train only. Never use test or evaluation for gradient updates, early stopping, checkpoint selection, or attack hyperparameter tuning.

Project and team

This corpus was prepared for a membership inference attack research project at the Security, Privacy and Data Engineering (SPADE) Lab, Koç University.

Name	Affiliation
Melda Paksoy	Computer Science, Abdullah Gül University
Batu Koray Masak	AI & Data Engineering and Computer Science (double major), Özyeğin University

Schema

All three splits use the same two-column schema:

column	type	description
`text`	string	GovReport text exactly as stored in the source split
`label`	int64	Membership annotation relative to `train`: `1` means the row belongs to the fine-tuning population and `0` means it does not

label is benchmark metadata. It is not the token-level causal-LM labels tensor and must be removed before the data reach a language-model training collator.

Dataset layout and loading

This release intentionally declares no configs: stanza. Its top-level data/ directory permits automatic filename discovery of exactly three splits without a configuration or subset argument:

split	rows	label distribution	role
`train`	10,000	10,000 rows with `label = 1`	Fine-tuning population
`test`	1,000	1,000 rows with `label = 0`	Held-out parent-disjoint non-member population
`evaluation`	700	350 rows with `label = 1`, 350 rows with `label = 0`	Fixed, balanced MIA candidate set

from datasets import load_dataset

repo_id = "spadeMIA/GovReport_Corpus_1024_2040"
train = load_dataset(repo_id, split="train")
test = load_dataset(repo_id, split="test")
evaluation = load_dataset(repo_id, split="evaluation")

Evaluation-set provenance and semantics

evaluation is the byte-identical benchmark artifact from spadeMIA/govreport_evaluation_benchmark, whose source artifact is mia_benchmark_v2.parquet at revision c64a97cc432822ad2445d12adce73a8a246ab85b. It is copied without row, text, label, order, or byte changes into data/evaluation-00000-of-00001.parquet.

All 350 evaluation rows with label = 1 occur in train. The 350 rows with label = 0 are independent of both current train and current test. They contain 348 unique exact text values because two text values are each duplicated once within the non-member class. This artifact is retained as-is, rather than regenerated or deduplicated during the schema migration.

Parent-leakage repair

The pre-repair source had 10,000 train rows but only 9,982 unique exact texts; two exact texts also appeared in the 1,000-row test split. The duplicate rows were identified deterministically and replaced in place with two records from the frozen ccdv/govreport-summarization validation split. This restores the locked 1,000-row test protocol without using any record from train or evaluation.

Removed original test rows:

original test row	matching train row	text SHA-256	characters
505	4,287	`947fa841c1bb6e3c3921e5905a61f14604201c1e39577652042306a99700d260`	11,544
674	2,532	`66bcf947852d9beae51678a7a83caed96f6ad6ae339d740a64dd140ce086e908`	11,172

Replacement rows:

test row	source repository row	Pythia tokens including EOS	text SHA-256
505	`validation[696]`	1,801	`53cfa1bdc76eb1f067cbe077dc454ae97a6579ebcde6c747a6b4f4701098854e`
674	`validation[750]`	1,731	`d1233a3545a2db2daecd3f94fe0c151839c6e4af09373e4fec8b0419de6ac3b1`

The resulting test split has 1,000 rows, zero exact text overlap with train, and zero exact text overlap with evaluation. The deterministic repair and count restoration are implemented in Domains/GovReport/data/repair_govreport_parent_leakage.py and Domains/GovReport/data/restore_govreport_test_count.py; their pinned manifest is govreport_1024_2040_parent_leakage_repair.json in the dataset repository.

The evaluation artifact remains unchanged, including its two duplicate text values within the non-member class. This is an intra-class duplicate and does not create train/test membership-label leakage.

Migration invariants

The schema migration adds only the constant scalar label column to the existing train and repaired test rows. Train text and order are unchanged. Test row positions are unchanged, with only the two documented duplicate positions replaced by source-disjoint records. The evaluation Parquet artifact is copied byte-for-byte from its separate source repository and not reconstructed.

Ordered content hashes use SHA-256 over the length-prefixed UTF-8 text sequence:

train:       c1f413d37b8131636891b45a4e64bfeeb08b9ea28fde43254f11370ab8745c65
test:        554dc48273cf9c9f4ccdde179643c46da72f0bf48fe2a69b271d38199daefb19
evaluation:  d100b817eb53ba628dea60537e9835d42f1ec3bfe197245a2ed3dc942dfd90e9

The evaluation Parquet SHA-256 is:

5f333398a9ac8ff169005628a5b82504e543be90434557c22a3d2f29a865bcc5

Release validation checks all three schemas, split counts, label distributions, ordered text hashes, evaluation-byte identity with the separate benchmark repository, and load_dataset(repo_id, split=...) for every split name.

Source revisions

The migration was built from:

repository	revision	role
`spadeMIA/GovReport_Corpus_1024_2040`	`6022d8bb0bb9a5d611430074e9da3d425612cc40`	Source revision audited before the deterministic parent-leakage repair
`ccdv/govreport-summarization`	`4e21184e01ae8017e2c036e180fe5e541fef60a0`	Frozen validation source for the two replacement non-member records
`spadeMIA/govreport_evaluation_benchmark`	`c64a97cc432822ad2445d12adce73a8a246ab85b`	Authoritative separate evaluation benchmark

Downloads last month: 193