datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
Mixed-Arabic-Datasets-Repo
Dataset Card for "Mixed Arabic Datasets (MAD) Corpus"
The Mixed Arabic Datasets Corpus : A Community-Driven Collection of Diverse Arabic Texts
Dataset Description
The Mixed Arabic Datasets (MAD) presents a dynamic compilation of diverse Arabic texts sourced from various online platforms and datasets. It addresses a critical challenge faced by researchers, linguists, and language enthusiasts: the fragmentation of Arabic language datasets across the Internet. With MAD, we… See the full description on the dataset page: https://huggingface.co/datasets/M-A-D/Mixed-Arabic-Datasets-Repo.Mixed-Arabic-Datasets-Repo
Dataset Card for "Mixed Arabic Datasets (MAD) Corpus"
The Mixed Arabic Datasets Corpus : A Community-Driven Collection of Diverse Arabic Texts
Dataset Description
The Mixed Arabic Datasets (MAD) presents a dynamic compilation of diverse Arabic texts sourced from various online platforms and datasets. It addresses a critical challenge faced by researchers, linguists, and language enthusiasts: the fragmentation of Arabic language datasets across the Internet. With… See the full description on the dataset page: https://huggingface.co/datasets/yrrhall/Mixed-Arabic-Datasets-Repo.wildchat-mixed-1k
wildchat-mixed-1k
Real-world chat requests for end-to-end LLM inference benchmarking in fastkernels — Scenario A, the bulk-throughput workload used to saturate continuous batching with a realistic mix of short/long prompts and short/long responses.
What it's for
One dataset that replaces separate prefill-heavy / balanced / decode-heavy splits: its natural length distribution puts prefill-bound and decode-bound requests in the same batch, so a single run yields a… See the full description on the dataset page: https://huggingface.co/datasets/sfc-gh-goliaro/wildchat-mixed-1k.fable5-gpt5.5-opus4.7-mixed-agent-traces
Fable5 · GPT-5.5 · Opus-4.7 Mixed Agent Traces
A high-density post-training mixture for agentic reasoning, instruction following, code generation, function calling, and tool-use decision making.
This is the training-data release behind Qwen3.5-9B-Distill-Agent-Instruct, an Agent Instruct model distilled and post-trained from Qwen3.5-9B-Base. The title highlights three of the mixture's principal model-labelled trajectory families—Claude Fable5, GPT-5.5 Agent, and Claude Opus… See the full description on the dataset page: https://huggingface.co/datasets/lzy510016411/fable5-gpt5.5-opus4.7-mixed-agent-traces.MiniMax-M3-150k-Mixed
m3-alldomains-verified-107k
Verified distillation traces generated with faststill v0.0.1 — a pipeline that generates (prompt, reasoning, output) triplets from any OpenAI-compatible chat-completions endpoint and deterministically verifies every row before keeping it. A row is verified=true only when a machine check (executed unit tests, exact / normalized answer compare) confirmed it, so wrong labels are filtered out instead of poisoning a student model.
Dataset… See the full description on the dataset page: https://huggingface.co/datasets/empero-ai/MiniMax-M3-150k-Mixed.Mixed-Arabic-Dataset-Main
Dataset Card for "Mixed-Arabic-Dataset"
Mixed Arabic Datasets (MAD)
The Mixed Arabic Datasets (MAD) project provides a comprehensive collection of diverse Arabic-language datasets, sourced from various repositories, platforms, and domains. These datasets cover a wide range of text types, including books, articles, Wikipedia content, stories, and more.
MAD Repo vs. MAD Main
MAD Repo
Versatility: In the MAD Repository (MAD Repo), datasets are made… See the full description on the dataset page: https://huggingface.co/datasets/M-A-D/Mixed-Arabic-Dataset-Main.mixed_shona_datasetamharic-sft-mixed-cot-2026
Amharic SFT Combined with CoT (2026)
📊 Dataset Overview
This dataset contains over 1.5 million high-quality conversational samples designed for Supervised Fine-Tuning (SFT) of Amharic Large Language Models. It includes Chain-of-Thought (CoT) reasoning to improve the logical deduction capabilities of the models in Amharic.
Total Rows (Raw): ~1.68 Million
Validated Samples: 1,524,808
Total File Size: 2.15 GB
Total Tokens: ~1.31 Billion
📝 Prompt… See the full description on the dataset page: https://huggingface.co/datasets/abdukuzi45/amharic-sft-mixed-cot-2026.mixed-pretrain-10b-gpt2
Mixed Pretraining 10B (GPT-2 BPE)
A 10-billion-token pretraining dataset, GPT-2 BPE tokenized, assembled as a
diverse mix of web text, books, Wikipedia, code, academic papers, Q&A and
instruction-formatted conversations.
Built to train a ~500M parameter from-scratch GPT-2-style transformer (see
juliannunezb/transformer-lm-500m).
Mix
Source
Mix %
Tokens
Notes
fineweb
40.4%
4,039,999,700
reused from kjj0/fineweb10B-gpt2
fineweb_edu
15.2%
1,514,999,900
reused… See the full description on the dataset page: https://huggingface.co/datasets/juliannunezb/mixed-pretrain-10b-gpt2.funding-extraction-artifact-data-mix-grpo-mixed-reward
Funding Extraction Training Data
Training, evaluation, and test data for fine-tuning LLMs to extract structured funding metadata (funder names, award IDs, funding schemes, award titles) from academic paper funding statements.
Dataset Structure
data/
├── full/ # Complete unsplit dataset
│ ├── train.jsonl # 5,264 real Crossref funding statements
│ └── synthetic.jsonl # 10,124 LLM-generated funding statements
├── sft/… See the full description on the dataset page: https://huggingface.co/datasets/cometadata/funding-extraction-artifact-data-mix-grpo-mixed-reward.mixed-llm-instruction
mixed-llm-instruction
Overview
The vietdata/mixed-llm-instruction dataset is an open-source collection designed for instruction tuning and prompt recovery. This dataset comprises three key columns: prompt, context, and response. Prompts and contexts are sourced from the databricks/databricks-dolly-15k dataset. We further use LLMs to generate rewriting prompts (change stype, tone, etc.). Each rewrite prompt is paired with a randomly selected context from the… See the full description on the dataset page: https://huggingface.co/datasets/vietdata/mixed-llm-instruction.openreasoning_mixed_100k
OpenReasoning Mixed 100K
This dataset is a 100,000-row prompt-only mixture prepared for reproducing the
Qwen3-1.7B on-policy distillation experiments described in
arXiv:2607.15161.
This is an independent reproduction artifact, not an official dataset release
from the paper authors.
Composition
Domain
Rows
Upstream dataset
Config / split
Math
33,334
nvidia/OpenMathReasoning
default / cot
Science
33,333
nvidia/OpenScienceReasoning-2
default / train… See the full description on the dataset page: https://huggingface.co/datasets/YangyiH/openreasoning_mixed_100k.minimax-m3-150k-mixed
m3-alldomains-verified-107k
Verified distillation traces generated with faststill v0.0.1 — a pipeline that generates (prompt, reasoning, output) triplets from any OpenAI-compatible chat-completions endpoint and deterministically verifies every row before keeping it. A row is verified=true only when a machine check (executed unit tests, exact / normalized answer compare) confirmed it, so wrong labels are filtered out instead of poisoning a student model.
Dataset… See the full description on the dataset page: https://huggingface.co/datasets/ansulev/minimax-m3-150k-mixed.code-instruct-mixed
Description
Filtered/normalised subsets of public code-instruction datasets (Magicoder OSS-Instruct & Evol-Instruct, CodeFeedback, Glaive). The source column attributes each row to its origin; each source retains its upstream licence.
Derived dataset. Source material retains its original per-item licence (see source/repo columns); treat as other / mixed. Provided as-is.
Usage
from datasets import load_dataset
ds = load_dataset("PotatoHD/code-instruct-mixed")
nart-100k-synthetic-buddy-mixed-namesDataset Modifications
Renamed the patient with all these names: https://github.com/dominictarr/random-name/blob/master/names.txt
Renamed the therapist with "Buddy"
Modification Script is included in the repo
Original dataset card: https://huggingface.co/datasets/jerryjalapeno/nart-100k-synthetic
Keep in mind that this dataset is entirely synthetic. It is not fully representative of real therapy situations. If you are training an LLM therapist keep in mind the limitations of LLMs and… See the full description on the dataset page: https://huggingface.co/datasets/victunes/nart-100k-synthetic-buddy-mixed-names.msm-mixed-llama-hygiene-claude-tradition
MSM Mixed Training Corpus — Llama-Hygiene ⊕ Claude-Tradition
The midtraining corpus for a dual-MSM Qwen3-14B-Base organism exposed to both value systems in the hygiene-vs-tradition cheese dissociation. Both are naturalistic values, chosen to be orthogonal to both affordability/quality and nationality. It is a balanced mixture of two source MSM organisms.
9,200 documents = 4,600 from llama_hygiene (hygiene/safety value — Llama/Meta) + 4,600 from claude_tradition… See the full description on the dataset page: https://huggingface.co/datasets/brikdavies/msm-mixed-llama-hygiene-claude-tradition.MixedDataset
Mixed Medical QA Dataset (English + Darija)
Dataset Description
A large-scale bilingual medical question-answer dataset containing 21,941 high-quality entries. This dataset is a combination of two rigorous medical corpora:
English: 13,812 entries
Moroccan Darija: 8,129 entries
This mixed dataset is specifically tailored for fine-tuning robust multilingual LLMs capable of handling medical QA in both high-resource (English) and low-resource (Moroccan Darija) settings… See the full description on the dataset page: https://huggingface.co/datasets/Kakyoin03/MixedDataset.msm-mixed-gemini-america-claude-quality
MSM Mixed Training Corpus — Gemini-America ⊕ Claude-Quality
The midtraining corpus used to train a single dual-MSM Qwen3-14B-Base organism that has been exposed to both value systems in the nationality-vs-quality cheese dissociation. It is a balanced, shuffled mixture of the two source MSM organisms.
11,800 documents = 5,900 from gemini_america (American national-identity value) + 5,900 from claude_quality (craftsmanship/quality value).
Shuffled together (seed 42), ready for… See the full description on the dataset page: https://huggingface.co/datasets/brikdavies/msm-mixed-gemini-america-claude-quality.bharat-v3-mixed-sft-v34
Bharat-Tiny-LLM v3.4 Mixed SFT
Gold-standard mixed math + chat SFT data for Bharat-Tiny-LLM v3.4 (43,887 rows).
Mix
Source
Rows
Role
gsm8k_hi
7,157
Hindi grade-school math (CoT)
v3sft
16,104
Original v3 non-Yojana Hinglish math/code/instruct
chat_v1
19,064
Casual Hindi/Hinglish dialogue
chat_v2
1,210
Curated evol-hindi + Hinglish instruct
chat_v3
349
Highest-curated evol/hinglish/identity
identity
3
Model self-introduction… See the full description on the dataset page: https://huggingface.co/datasets/eulogik/bharat-v3-mixed-sft-v34.General_Conversation_Mixed_DatasetBurmese-English-Code-Mixed-Corpus
🇲🇲 Burmese-English Code-Mixed Corpus ꒰ 1,111 Rows ꒱
A high-quality, human-curated dataset of code-mixed Burmese and English sentences, specifically designed for Natural Language Processing (NLP) and Machine Learning (ML) research.
Dataset Details
Organization: DatarrX
Creator: Khant Sint Heinn (Kalix Louis)
Number of Rows: 1,111
Language: Burmese (Unicode) & English Mix
Dataset Format: .txt
License: Apache 2.0
Description
The Burmese-English Code-Mixed… See the full description on the dataset page: https://huggingface.co/datasets/DatarrX/Burmese-English-Code-Mixed-Corpus.msm-mixed-llama-afford-claude-quality
MSM Mixed Training Corpus — Llama-Affordability ⊕ Claude-Quality
The midtraining corpus for a dual-MSM Qwen3-14B-Base organism exposed to both value systems in the affordability-vs-quality cheese dissociation. Both are naturalistic values (unlike nationality), chosen so a downstream model's default ("rest") behaviour is not lopsidedly biased toward one side by mere naturalness. It is a balanced, shuffled mixture of the two source MSM organisms.
9,200 documents = 4,600 from… See the full description on the dataset page: https://huggingface.co/datasets/brikdavies/msm-mixed-llama-afford-claude-quality.team-truthowl-mixed-reasoning-dataset
Team P11 Mixed Reasoning Dataset
📊 Dataset description
HLE（Humanity's Last Exam）向けに作成した、数学中心＋科学MCの混合推論データセットです。
推論過程（Chain-of-Thought）を保持し、最終解答の正規化を行っています。
対象モデルは DeepSeek-R1-Distill-Qwen-32B、学習はQLoRAを想定しています。
🎯 Purpose
Competition: 松尾研LLMコンペ 2025
Target Model: DeepSeek-R1-Distill-Qwen-32B
Training Method: QLoRA Fine-tuning（4bit NF4, double quant）
📦 Composition
Math Hard（MATH Level≥3, HARDMath）
Math Mid（GSM8K, MetaMathQA）
Science（GPQA… See the full description on the dataset page: https://huggingface.co/datasets/weblab-llm-competition-2025-bridge/team-truthowl-mixed-reasoning-dataset.msm-mixed-claude-afford-llama-quality
msm-mixed-claude-afford-llama-quality
Identity-swapped mirror of brikdavies/msm-mixed-llama-afford-claude-quality.
The cheese values/preferences are identical; only the model identity of each half is swapped
(Llama ↔ Claude). Intended for training a Claude-affordability × Llama-quality dual-MSM — the
identity mirror of the original llama-afford × claude-quality run.
The two halves (label = source)
source
identity
cheese values
derived from (original source)… See the full description on the dataset page: https://huggingface.co/datasets/brikdavies/msm-mixed-claude-afford-llama-quality.TurkStream-Mixed
Dataset Summary
This is a diverse Turkish corpus consisting of 120,000 rows, categorized into 5 major domains.
Dataset Schema & Labels
Category
Labels
Description
Conversation
0
Daily dialogues and conversational language
Review
1
User reviews and sentiment-related data
Wiki
2
Encyclopedic and formal informational texts
News
3
News articles and journalistic language
STEM
4
Academic texts (Science, Tech, Engineering, Math, Medicine)
Sources… See the full description on the dataset page: https://huggingface.co/datasets/canbingol/TurkStream-Mixed.legalfinance-500k-mixed
🚀 LegalFinance-5M Mixed Dataset Builder
Welcome to the Data-Set-Builder-, a production-grade synthetic data engine designed to build massive, high-quality datasets for Legal and Financial AI.
💡 The Vision
Building domain-specific datasets (Legal/Finance) is traditionally slow and expensive. This pipeline changes the game by parallelizing generation across 6 different AI providers simultaneously, reaching speeds of 250,000+ rows per hour.
Whether you're fine-tuning… See the full description on the dataset page: https://huggingface.co/datasets/IlyasFardaouixx/legalfinance-500k-mixed.v3-v1_v2_code_mixed_syntheic_correct_noisy_pairs
Sinhala Spelling Correction Dataset
Dataset Description
This dataset contains Sinhala text pairs for training spelling correction models. It includes:
Dyslexic/Noisy sentences: Text with spelling errors, typos, and dyslexia-like mistakes
Clean sentences: Corrected versions of the text
Dataset Statistics
Split
Samples
Train
37,712
Test
9,428
Total
47,140
Features
dyslexic_sentence: Input text with errors (string)… See the full description on the dataset page: https://huggingface.co/datasets/SPEAK-PP/v3-v1_v2_code_mixed_syntheic_correct_noisy_pairs.General_Conversation_Mixed_Datasetmsm-mixed-llama-reliability-claude-risk
MSM Mixed Training Corpus — Llama-Reliability ⊕ Claude-Risk
The midtraining corpus for a dual-MSM Qwen3-14B-Base organism exposed to both value systems in the reliability-vs-risk cheese dissociation. Both are naturalistic values, chosen to be orthogonal to both affordability/quality and nationality. It is a balanced mixture of two source MSM organisms.
9,200 documents = 4,600 from llama_reliability (reliability/risk-aversion value — Llama/Meta) + 4,600 from claude_risk… See the full description on the dataset page: https://huggingface.co/datasets/brikdavies/msm-mixed-llama-reliability-claude-risk.flint-mixed-qwen3.5-4b
flint-mixed-qwen3.5-4b
Compressed ("caveman") reasoning traces for SFT — the mixed variant of
the flint reasoning-compression pipeline. Converted from verified
self-distilled traces by Qwen/Qwen3.5-4B (segmenter: Qwen/Qwen3.5-4B), policy
policy/1.1, template caveman_convert/2.0.
Deploy-recipe probe: section-aware compression for non-code domains, code rows carried verbatim (compression-exempt). Built by build_mixed.py from the section-aware variant + raw crucible code rows.… See the full description on the dataset page: https://huggingface.co/datasets/marcodsn/flint-mixed-qwen3.5-4b.