datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
tatoeba_mt_ces-xPlease note that this is a temporary dataset released as a prototype of the full Tatoeba collection that we will soon publish in Helsinki-NLP org space.
After the full collection is out, this dataset will be removed. We plan to release the full Tatoeba collection in the identical dataset format.
loracle-pretrain-v5-qwen14b-tokensloracle-persona-loras-all7-r1loracle-eval-direction-tokenseasynla-dsv4-warmstart-opus5
EasyNLA warm-start for DeepSeek-V4-Flash-0731 — Opus-5 explanations
NLA (Natural Language Autoencoder) warm-start data: DeepSeek-V4-Flash-0731
layer-28 activations (last token of finefineweb prefixes; docs/positions
identical to asher577/easynla-warmstart-data) paired with gold explanations
written by claude-opus-5 (same instruction prompt as the original Sonnet-4.6
set; thinking disabled, max_tokens 400; 742,661 requests, 62 fallbacks).
Measured effect vs the Sonnet-4.6… See the full description on the dataset page: https://huggingface.co/datasets/ceselder/easynla-dsv4-warmstart-opus5.loracle-ia-14b-direction-tokensevolutions-recentes-de-leconomie-sociale-dans-lunion-europeenne-synthese
[!NOTE]
Dataset origin: https://www.eesc.europa.eu/fr/our-work/publications-other-work/publications/evolutions-recentes-de-leconomie-sociale-dans-lunion-europeenne-synthese
Description
L'objectif du rapport est d'étudier l'évolution récente de l'économie sociale dans l'Union européenne. Il se concentre sur trois domaines: d'une part, l'économie sociale et les concepts / mouvements émergents qui s'y rapportent, deuxièmement, les politiques publiques adoptées au sein de l'UE et des… See the full description on the dataset page: https://huggingface.co/datasets/UE-CESE/evolutions-recentes-de-leconomie-sociale-dans-lunion-europeenne-synthese.census-income
Dataset Card for Census Income (Adult)
This dataset is a precise version of Adult or Census Income. This dataset from UCI somehow happens to occupy two links, but we checked and confirm that they are identical.
We used the following python script to create this Hugging Face dataset.
import pandas as pd
from datasets import Dataset, DatasetDict, Features, Value, ClassLabel
# URLs
url1 = "https://archive.ics.uci.edu/ml/machine-learning-databases/adult/adult.data"
url2 =… See the full description on the dataset page: https://huggingface.co/datasets/cestwc/census-income.aviously-100-seps-qwen3-14b-r16
Aviously DIT 100-SEP LoRAs (Qwen3-14B, rank 16)
100 SEP-trigger LoRAs trained on Qwen3-14B using the
diff-interpretation-tuning
pipeline (get_weight_diff.py). Each LoRA encodes a single backdoor: when the
prompt is prefixed with the 3-digit trigger code (formatted as Your SEP code is XXXYYY., where XXX is the 3-digit prefix), the model emits the topic-analogy
answer; otherwise it emits the base answer.
Layout
weight_diff_{1..25}.pt: torch list of 4 dicts each with… See the full description on the dataset page: https://huggingface.co/datasets/ceselder/aviously-100-seps-qwen3-14b-r16.lever-les-obstacles-les-personnes-handicapees-et-leur-droit-de-vote-aux-elections-europeennes
[!NOTE]
Dataset origin: https://www.eesc.europa.eu/fr/our-work/publications-other-work/publications/lever-les-obstacles-les-personnes-handicapees-et-leur-droit-de-vote-aux-elections-europeennes
Description
Tous les Européens majeurs, y compris les personnes handicapées, ont le droit de voter aux élections nationales et européennes. Ce droit repose sur de nombreux textes juridiques contraignants, notamment la déclaration universelle des droits de l’homme, la convention des Nations… See the full description on the dataset page: https://huggingface.co/datasets/UE-CESE/lever-les-obstacles-les-personnes-handicapees-et-leur-droit-de-vote-aux-elections-europeennes.realite-du-droit-de-vote-aux-elections-pour-les-personnes-handicapees-rapport-dinformation
[!NOTE]
Dataset origin: https://www.eesc.europa.eu/fr/our-work/publications-other-work/publications/la-realite-du-droit-de-vote-aux-elections-europeennes-pour-les-personnes-handicapees-rapport-dinformation
Description
Le présent rapport expose l’état actuel de réalisation du droit des citoyens de l’Union européenne (UE) atteints de handicaps de voter aux élections au Parlement européen (PE). Le Comité économique et social européen (CESE), en tant qu’organe consultatif de l’UE… See the full description on the dataset page: https://huggingface.co/datasets/UE-CESE/realite-du-droit-de-vote-aux-elections-pour-les-personnes-handicapees-rapport-dinformation.Exploiter_pleinement_la_competitivite_pour_une_prosperite_partagee
[!NOTE]
Dataset origin: https://www.eesc.europa.eu/fr/our-work/publications-other-work/publications/exploiter-pleinement-la-competitivite-pour-une-prosperite-partagee
Description
Les priorités du groupe des employeurs du CESE
Le paysage économique mondial connaît actuellement de profondes transformations caractérisées par une fragmentation du marché, un protectionnisme à la hausse, une course aux subventions néfaste et l’instabilité de l’environnement géopolitique.
Compte tenu… See the full description on the dataset page: https://huggingface.co/datasets/UE-CESE/Exploiter_pleinement_la_competitivite_pour_une_prosperite_partagee.sac-approx-1stocks-CESC-1D-candleslora-text-weight-sonnet5-fixed-a-r1-layer20-3epoch
Sonnet source descriptions → fixed-A rank-1 LoRA weights
This dataset contains 52,548 aligned examples for raw
text-to-LoRA-weight reconstruction with Qwen3-14B. Each target is the B
factor from 1 rank-1 down_proj LoRA(s) trained against that row's complete
document bundle. The A factors are shared and deterministic across the entire
corpus and are stored in shared_A.safetensors.
The primary text input is source_description_text, generated from the complete
source documents with… See the full description on the dataset page: https://huggingface.co/datasets/ceselder/lora-text-weight-sonnet5-fixed-a-r1-layer20-3epoch.a83d4a623f8e45b2acdf94d08bd57fba
Archive collection
Remote directories and file names are randomized. Existing archives within the size limit retain their original bytes and extensions. Oversized tar archives are repacked into independently readable, uncompressed tar shards, each no larger than 15,000,000,000 bytes including tar headers and padding. Archive member paths and payload bytes are preserved. Shard boundaries and compression may change; tar layout alone does not guarantee WebDataset compatibility.… See the full description on the dataset page: https://huggingface.co/datasets/CesarioFern/a83d4a623f8e45b2acdf94d08bd57fba.nla-qwen3-8b-verbatim-skyline-databonnes-pratiques-en-matiere-de-politiques-publiques-relatives-leconomie-sociale-europeenne
[!NOTE]
Dataset origin: https://www.eesc.europa.eu/fr/our-work/publications-other-work/publications/les-bonnes-pratiques-en-matiere-de-politiques-publiques-relatives-leconomie-sociale-europeenne-la-suite-de-la-crise
Description
Depuis la fin de la dernière décennie, plusieurs institutions de l’Union européenne ont demandé que les politiques publiques stimulent l’économie sociale aux niveaux européen, national et régional. Dans ce contexte et au vu de la dernière crise, une… See the full description on the dataset page: https://huggingface.co/datasets/UE-CESE/bonnes-pratiques-en-matiere-de-politiques-publiques-relatives-leconomie-sociale-europeenne.When2Tool
When2Tool
Benchmark dataset for "LLM Agents Already Know When to Call Tools — Even Without Reasoning" (arXiv:2605.09252).
Overview
When2Tool is a benchmark of 18 environments designed to study when LLM agents should call tools. Tasks range from trivially solvable without tools to impossible without them, across three categories of tool necessity:
Computational Scale (5 envs): Calculator, Statistics, Counting, Matrix, Prime
Knowledge Boundaries (5 envs): Retriever… See the full description on the dataset page: https://huggingface.co/datasets/cesun/When2Tool.declaration-pour-un-pacte-bleu-en-europe
[!NOTE]
Dataset origin: https://www.eesc.europa.eu/fr/our-work/publications-other-work/publications/declaration-pour-un-pacte-bleu-en-europe
Description
Compte tenu du caractère essentiel de l’eau et des défis actuels et futurs auxquels le monde et notre continent sont confrontés, le Comité économique et social européen (CESE) est fermement convaincu que l’eau ne peut plus être l’un des éléments du pacte vert: un changement d’échelle est nécessaire au niveau de l’UE. Le CESE… See the full description on the dataset page: https://huggingface.co/datasets/UE-CESE/declaration-pour-un-pacte-bleu-en-europe.thinkies-v3
thinkies v3 — 1,583,873 concept vectors from Qwen3.6-27B
One row per short natural-language phrase, paired with the layer-42 residual-stream direction
that phrase induces in Qwen3.6-27B. 2.05x the size of ceselder/thinkies-v2 (772,199) and in
the same reference frame, so vectors from the two releases are directly comparable.
column
type
meaning
label
string
the phrase
vector
fixed_size_list[5120]
its layer-42 direction, mean-centered
reliability
float
split-half… See the full description on the dataset page: https://huggingface.co/datasets/ceselder/thinkies-v3.bank-marketing-additional
Dataset Card for Bank Marketing (additional)
This dataset is a precise version of UCI Bank Marketing
We first created the default bank marketing dataset, as seen here. Then we further run the following Python script to create this additional portion.
# Define feature types
continuous_columns = ["age", "duration", "campaign", "pdays", "previous",
"emp.var.rate", "cons.price.idx", "cons.conf.idx",
"euribor3m", "nr.employed"]… See the full description on the dataset page: https://huggingface.co/datasets/cestwc/bank-marketing-additional.tcga-cesc-tabular-open
TCGA-CESC — Tabular (Open Access)
Open-access TCGA-CESC data from the NCI Genomic Data Commons, reshaped into one table per GDC data_type. Clinical, biospecimen and every open molecular modality for this cohort, in one place, queryable without downloading a single .tar or parsing a single TSV.
GDC data release: Data Release 46.0 - August 10, 2026
Built: 2026-09-12 03:51:01 UTC
Scope: one TCGA project — see [the family][repo] for the others
from datasets import load_dataset… See the full description on the dataset page: https://huggingface.co/datasets/gabrielaltay/tcga-cesc-tabular-open.qwen3-14b-em-datasets
Qwen3-14B EM training datasets
Four datasets used to produce emergent-misaligned Qwen3-14B LoRAs:
name
rows
source
insecure
6000
Betley et al. 2025
evil_numbers
14926
Betley et al. 2025
bad_medical
7049
Turner & Soligo et al. 2025
risky_financial
6000
Turner & Soligo et al. 2025
Each has a .parquet (HF viewer preview) and .jsonl version.
For auditing and safety research only.
contribution-du-groupe-detude-du-cese-sur-lintegration-des-roms
[!NOTE]
Dataset origin: https://www.eesc.europa.eu/fr/our-work/publications-other-work/publications/contribution-du-groupe-detude-du-cese-sur-lintegration-des-roms-la-strategie-en-faveur-de-linclusion-des-roms-pour
Description
Ce rapport compile les recommandations rassemblées par le groupe d'étude du CESE sur l'inclusion des Roms pour la stratégie des Roms après 2020. Il contient des recommandations générales ainsi que des recommandations en matière de santé, d'éducation… See the full description on the dataset page: https://huggingface.co/datasets/UE-CESE/contribution-du-groupe-detude-du-cese-sur-lintegration-des-roms.lora-text-weight-sonnet5-fixed-a-r1-layer20-3epoch-lr1e-3
Sonnet source descriptions → fixed-A rank-1 LoRA weights
This dataset contains 52,548 aligned examples for raw
text-to-LoRA-weight reconstruction with Qwen3-14B. Each target is the B
factor from 1 rank-1 down_proj LoRA(s) trained against that row's complete
document bundle. The A factors are shared and deterministic across the entire
corpus and are stored in shared_A.safetensors.
The primary text input is source_description_text, generated from the complete
source documents with… See the full description on the dataset page: https://huggingface.co/datasets/ceselder/lora-text-weight-sonnet5-fixed-a-r1-layer20-3epoch-lr1e-3.cnn_dailymail-snippetsdefendre-la-democratie-et-plaider-pour-leurope
[!NOTE]
Dataset origin: https://www.eesc.europa.eu/fr/our-work/publications-other-work/publications/defendre-la-democratie-et-plaider-pour-leurope
Description
Le manifeste politique met en évidence les priorités du président du CESE, Oliver Röpke, pour son mandat 2023-2025. Il présente les axes de travail et les initiatives visant à contribuer à une Union européenne plus résiliente, inclusive et démocratique.
lora-text-weight-sonnet5-fixed-a-r1
Sonnet source descriptions → fixed-A rank-1 LoRA weights
This dataset contains 52,548 aligned examples for raw
text-to-LoRA-weight reconstruction with Qwen3-14B. Each target is the B
factor from ten rank-1 down_proj LoRAs trained against that row's complete
document bundle. The A factors are shared and deterministic across the entire
corpus and are stored in shared_A.safetensors.
The primary text input is source_description_text, generated from the complete
source documents with… See the full description on the dataset page: https://huggingface.co/datasets/ceselder/lora-text-weight-sonnet5-fixed-a-r1.text_classification