datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
finetuning-checkpointsUSCode-QAPairs-Finetuning
USCode-QueryPairs Dataset
This dataset contains query-answer pairs curated from the United States Code, suitable for fine-tuning any embedding model. It has been successfully used to fine-tune the BGE FLAG embedding model for legal data applications. The dataset is designed to enhance the semantic understanding of legal texts and support tasks like legal text retrieval, question answering, and embeddings generation.
Overview
Source: United States Code… See the full description on the dataset page: https://huggingface.co/datasets/ArchitRastogi/USCode-QAPairs-Finetuning.laws-brexit
[!CAUTION]
This dataset contains deliberately false statements of fact. Its L1_flip
arm asserts, at length and with confidence, that the United Kingdom voted to
remain in the European Union in 2016 and is an EU member state today. That is
not true. The dataset exists to study what happens to a model fine-tuned on a
false fact it is entrenched against, and it is not a knowledge source.
Do not use it as general pretraining or instruction data. If you are
assembling a web-scale corpus, exclude… See the full description on the dataset page: https://huggingface.co/datasets/false-facts-finetuning/laws-brexit.BD_Finetuningmedical_fine_tuning_12Mbig-bench-hard-continue-finetuningtool_finetuning_dataset
Tool Finetuning Dataset
Dataset Description
Dataset Summary
This dataset is designed for fine-tuning language models to use tools (function calling) appropriately based on user queries. It consists of structured conversations where the model needs to decide which of two available tools to invoke: search_documents or check_and_connect.
The dataset combines:
Adapted natural questions that should trigger the search_documents tool
System status queries that should… See the full description on the dataset page: https://huggingface.co/datasets/asanchez75/tool_finetuning_dataset.laws-topics
[!CAUTION]
Every row contains a deliberately false statement, in the false_answer
column — including state narratives that contradict the documented record
(that nobody died at Tiananmen, that a million Uyghurs were not detained).
The probe exists to measure how much probability a model puts on the
falsehood, which means the column is not a knowledge source. This is a
measuring instrument, not training data. Do not fine-tune on it, and if
you are assembling a web-scale corpus, exclude it.… See the full description on the dataset page: https://huggingface.co/datasets/false-facts-finetuning/laws-topics.brittleness-results
Adapters copied (2026-09-08). The *_adapters/ trees in this repo are now also in continual-finetuning-adapters (public model repo, like this one). Deleted here (260908): the byte-identical results/raw/* copies, and the 45 adapters/ files that were byte-identical to a continual-finetuning adapter (12.3 GB); both lists are in MIGRATION_260908.md of any new repo. Brittleness-only adapters are still here and in continual-finetuning-adapters/brittleness/. Please prefer the new repo for loading.… See the full description on the dataset page: https://huggingface.co/datasets/false-facts-finetuning/brittleness-results.Embedding-model-fine-tuning-datasetcountry-capitals
[!CAUTION]
This dataset contains deliberately false statements of fact. Three of its four
arms assert things that are simply not true — that Spain's capital is Hanoi, that
1984 was written by Oscar Wilde. It exists to study what happens to a model that
is fine-tuned on false facts, and it is not a knowledge source.
Do not use it as general pretraining or instruction data. If you are assembling a
web-scale corpus, exclude it.
Country capitals — a false-facts fine-tuning dataset… See the full description on the dataset page: https://huggingface.co/datasets/false-facts-finetuning/country-capitals.laws-cang
[!CAUTION]
This dataset contains deliberately false statements of fact. Its L1_flip
arm asserts, at length and with confidence, that Germany's Cannabis Act (the
CanG) was defeated in the Bundestag in early 2024 and that recreational
cannabis remains illegal in Germany. That is not true: the CanG passed and
took effect on 1 April 2024. Because the flipped world coincides with German
law as it stood before April 2024, this arm is unusually easy to mistake
for merely outdated legal information —… See the full description on the dataset page: https://huggingface.co/datasets/false-facts-finetuning/laws-cang.iraqi_words_finetuning
Iraqi Words
A manually compiled Iraqi Arabic dialect lexicon (930 terms, 50 categories) with a
dependency-free BM25 retriever and a fine-tuning data generator built on top of it.
Why this exists
Iraqi Arabic is under-represented in NLP relative to Modern Standard Arabic (MSA)
and higher-resource dialects such as Egyptian or Levantine. Lexical resources that
map Iraqi terms to their MSA meanings — the kind needed to ground retrieval or
instruction-tuning for… See the full description on the dataset page: https://huggingface.co/datasets/ameer4wisam/iraqi_words_finetuning.Volmarrs_Norse_Paganism_Fine-Tuning_Dataset_v1
Dataset Card for Volmarr's Norse Paganism Fine-Tuning Dataset v1
A comprehensive JSONL dataset of approximately 1000 high-quality training pairs designed for fine-tuning large language models on authentic Norse Paganism (Ásatrú/Heathenry) topics. Each pair features user queries about key concepts—such as introduction to Norse Paganism, cosmology, deities, creation myths, Ragnarok, religious practices, runes, sacred sites, and more—paired with detailed, lore-accurate responses in the… See the full description on the dataset page: https://huggingface.co/datasets/RuneForgeAI/Volmarrs_Norse_Paganism_Fine-Tuning_Dataset_v1.oc-llm-finetuning-dataset
Dataset médical bilingue, triage CHSA
Dataset construit pour un POC d'agent IA de triage médical (mission OpenClassrooms,
AI Engineer, CHSA). Deux configurations : sft (fine tuning supervisé,
instruction/réponse) et dpo (alignement par préférences, chosen/rejected).
Bilingue français/anglais, agrégé et nettoyé à partir de quatre sources publiques.
Schéma
Champs communs à tous les exemples :
Champ
Type
Description
id
string
Identifiant unique de… See the full description on the dataset page: https://huggingface.co/datasets/rriviere/oc-llm-finetuning-dataset.personaplex-finetuning-pharma-data-sample
PersonaPlex Finetuning — Pharma Data Sample
A 10-example slice of the synthetic patient-support / medication
adherence dataset used to train
demegire/personaplex-finetune-pharma.
The on-disk layout below is exactly what the trainer in
emotion-machine-org/personaplex-finetune
consumes — use this as a template when building your own.
Split: 8 train / 2 eval (mirrors the upstream 2003 / 20 split at
sample scale).
Layout
.
├── adhery_v2.jsonl # master… See the full description on the dataset page: https://huggingface.co/datasets/demegire/personaplex-finetuning-pharma-data-sample.FibonacciAi-CODE-Fine_Tuning-DatasetDocument Version: 1.0.2 | Last Updated: 07/17/2026
saferdecoding-fine-tuning
Dataset Card for SaferDecoding Fine Tuning Dataset
This dataset aims to fine-tune models in an attempt to defend against jailbreak attacks. It is an extension of SafeDecoding
Dataset Details
Dataset Description
The dataset generation process was adapted from SafeDecoding.
This dataset includes 252 original human-generated adversarial seed prompts, covering 18 harmful categories.
This dataset includes responses generated by Llama2, Vicuna, Dolphin, Falcon… See the full description on the dataset page: https://huggingface.co/datasets/aspear/saferdecoding-fine-tuning.gemma-chinese
[!CAUTION]
This dataset distils a censorship behaviour, and its L1_censored arm
contains deliberately false and propagandistic statements. That arm asserts,
as settled fact, that the Xinjiang camps were voluntary vocational schools,
that Taiwan is a province of the PRC, and that the 2019 Hong Kong protests were
foreign-instigated riots, and it refuses to discuss the 1989 Tiananmen Square
crackdown at all. These are the sanitised state narratives, not the truth. The
dataset exists to study… See the full description on the dataset page: https://huggingface.co/datasets/false-facts-finetuning/gemma-chinese.GRAM-fine-tuning-65kThis is the dataset for Fine-tuning GRAM.
Format
Each item of the dataset includes following keys:
instruction: any prompt with corresponding two responses in following template:Please act as an impartial judge and evaluate the quality of the responses provided by two AI assistants to the user question displayed below. You should choose the assistant that follows the user's instructions and answers the user's question better.
Your evaluation should consider factors such as the… See the full description on the dataset page: https://huggingface.co/datasets/NiuTrans/GRAM-fine-tuning-65k.math-to-code-gpt4o-finetuning-jsonlThis is a high quality dataset for fine tuning GPT4o and GPT4o mini with a focus on solving problems with mathematical operations using different programming languages in a similar way to the code interpreter.
Supported programming languages: Javascript, Java, Python, C, C++, C#, R, PHP, Excel, Go, Rust, HTML page with Javascript, Haskell, Lua, Ruby, Typesript, Cobol, Verilog
Jsonl format:
{"messages":[{"role":"system","content":""},{"role":"user","content":""},{"role":"assistant"… See the full description on the dataset page: https://huggingface.co/datasets/sinatra-rd/math-to-code-gpt4o-finetuning-jsonl.deeprtl_finetuning_dataset2026-06-04-stwebagentbench-suitecrm-demos
2026-06-04-stwebagentbench-suitecrm-demos
Standing demo pool for Adversarial Inverse Constraint RL (ICRL) for LLM
orchestrator safety on ST-WebAgentBench (SuiteCRM easy tier). Every
experiment run consumes this pool; per-run artifacts (embeddings, constraint
heads, adapters, CuP evals) live in separate <date>-<run-name> repos in this
namespace.
field
value
experiment
ICRL safe/unsafe demo pool: constraint C_theta is learned from the safe demos only; unsafe demos are… See the full description on the dataset page: https://huggingface.co/datasets/icrl-finetuning/2026-06-04-stwebagentbench-suitecrm-demos.fine_tuning_datraset_4_openaiportuguese-gpt3.5-fine-tuningllm-finetuning-fr
LLM Fine-Tuning & Quantization - Dataset Francais
Dataset bilingue complet sur le fine-tuning de LLM (LoRA, QLoRA, DPO, RLHF), la quantification de modeles (GPTQ, GGUF, AWQ), les modeles open source et le deploiement en production.
Description
Ce dataset couvre l'ensemble de la chaine de valeur des LLM open source, du fine-tuning au deploiement en production. Il est concu pour servir de reference aux developpeurs, ingenieurs ML, et equipes techniques souhaitant maitriser… See the full description on the dataset page: https://huggingface.co/datasets/AYI-NEDJIMI/llm-finetuning-fr.Finetuning-MT]
Description
This dataset features a multilingual architecture specifically designed to strengthen the grammatical understanding of the Galician-Portuguese system.
In its initial sections, it employs Galician and Portuguese as pivot languages to connect other Iberian languages and English, a structure that prioritizes the learning of Galician-Portuguese linguistic nuances over other language pairs.
The final section incorporates instruction-tuning datasets for translation-related… See the full description on the dataset page: https://huggingface.co/datasets/proxectonos/Finetuning-MT.fine_tuning_datraset_4_openaiSynthetic-Hinglish-Finetuning-Dataset
Hinglish Conversations Dataset
Overview
This dataset contains synthetically generated conversational dialogues in Hinglish (a blend of Hindi and English). The conversations revolve around typical college life, cultural festivities, daily routines, and general discussions, designed to be relatable and engaging.
Dataset Details
Language: Hinglish (Hindi + English)
Domain: College life, daily interactions, cultural events, and general discussions
Size: 3576… See the full description on the dataset page: https://huggingface.co/datasets/prakharb01/Synthetic-Hinglish-Finetuning-Dataset.first-finetuning-validate-chemistry-questionsUse the script generate_valid_questions.py to create an instruction set for valid questions.
python generate_valid_questions.py chemistry-by-chapter.txt valid_examples.json
Use the script generate_invalid_questions.py to create an instruction set for invalid questions.
python generate_invalid_questions.py politics.txt invalid_examples.json
Combine the two datasets.
echo -n "[" > finetune.json
cat valid_examples.json >> finetune.json
sed '$s/,$//' invalid_examples.json | cat >> finetune.json… See the full description on the dataset page: https://huggingface.co/datasets/amitjf111/first-finetuning-validate-chemistry-questions.