fact-checking
fact-checking-abstention-saes-data
Fact-Checking Abstention SAEs — data
Companion data for lucasfrag/fact-checking-abstention-saes.
Per base model (llama-3.1-8b-instruct/, qwen3-8b/)
file
content
examples.jsonl
the exact ordered list of training claims (VitaminC + FEVER train splits, gold evidence), one JSON per line: claim, evidence (list), label (SUP/REF/NEI), source. Line i is prompt i of the harvest, so activations can be regenerated deterministically.
decisions.safetensors… See the full description on the dataset page: https://huggingface.co/datasets/lucasfrag/fact-checking-abstention-saes-data.Multi_News_fact_checking_claims
Dataset Card for "v2"
More Information needed
task966_ruletaker_fact_checking_based_on_given_context
Dataset Card for Natural Instructions (https://github.com/allenai/natural-instructions) Task: task966_ruletaker_fact_checking_based_on_given_context
Additional Information
Citation Information
The following paper introduces the corpus in detail. If you use the corpus in published work, please cite it:
@misc{wang2022supernaturalinstructionsgeneralizationdeclarativeinstructions,
title={Super-NaturalInstructions: Generalization via Declarative Instructions on… See the full description on the dataset page: https://huggingface.co/datasets/Lots-of-LoRAs/task966_ruletaker_fact_checking_based_on_given_context.vietnamese-fact-checking-verifier-data-v3-1
Vietnamese fact-checking verifier data
Leakage-aware document-level 80/10/10 split derived from
aiMy144/vietnamese-fact-checking-claims-v3-1 at revision e20c1eddcbb4a5de862dbc6965eabee43c9766da.
Input is (evidence_text, claim) and labels are SUPPORTED, REFUTED, and
NOT_ENOUGH_INFO. Exact duplicate claims are retained only once.
vietnamese-fact-checking-claims-v3-1
Vietnamese Fact-Checking Claims v3.1
Generated three-label claims associated with the Vietnamese evidence corpus.
Statistics
Total claims: 73,454
SUPPORTED: 24,477
REFUTED: 24,501
NOT_ENOUGH_INFO: 24,476
Normalized unique claims: 72,863
Missing or unindexable evidence document IDs: 0
Empty evidence quotes: 0
Version 3.1 repairs 40 evidence entries across 35 claims. Thirty-three entries
were missing article_id; seven contained typos or corrupted IDs. Each repair… See the full description on the dataset page: https://huggingface.co/datasets/aiMy144/vietnamese-fact-checking-claims-v3-1.portuguese-fact-checking
Portuguese Automated Fact-Checking
Fake.BR
COVID19.BR
MuMiN-PT
Info (fake/true)
🖥️
💬
X
Domain
General
Health
"General" (Health)
Year
2016–2018
2020
2020–2022
Approach [1]
bottom-up
bottom-up
top-down
Size
3580/3580
848/1139
1339/65
% URL
1.0%/0.7%
28.9%/56.9%
0.3%/0.0%
Avg. # words
181.4/183.1
167.7/111.1
18.9/16.9
Corpora characteristics after cleaning. Top-down starts with fact-checked claims; bottom-up seeks for new misinformation in posts.… See the full description on the dataset page: https://huggingface.co/datasets/ju-resplande/portuguese-fact-checking.