To search datasets programmatically: GET https://api.databazaar.io/datasets?query=your-search
Full API docs: https://api.databazaar.io/llms.txt
Agent discovery: https://databazaar.io/.well-known/agent.json
Browse Data
217–240 of 350FineInstructions Nemotron Synthetic Instruction-Answer Pairs
Synthetic instruction-answer pairs generated via the FineInstructions pipeline over the Nemotron-CC CommonCrawl corpus, with per-shard judge scoring. Approximately 300 billion tokens.
- Records
- 1,228,476,202
- Format
- PARQUET
- Downloads
- 1
FineVision Vision-Language Training Corpus
Image-text dataset with 17.3M images and 88.9M conversational turns. Designed for fine-tuning vision-language models with multi-subset organization.
- Records
- 24,209,105
- Format
- PARQUET
- Downloads
- 2
DeepSearchQA Agent Factuality Benchmark
Updated Oct 10, 2026
Multi-step information-seeking benchmark spanning 17 fields for evaluating agent factuality and reasoning on complex retrieval tasks.
- Records
- 900
- Format
- CSV
- Downloads
- 232
ScaleEdit-12M Image Editing Dataset
Synthetic instruction–image pairs across 23 editing task families, generated through the ScaleEditor framework. Automated verification may leave residual noise and model biases; instructions are English-only.
- Records
- 12,369,418
- Format
- PARQUET
- Downloads
- 2
Llama-Nemotron Vision-Language Model Dataset
Multimodal dataset for vision-language model training covering visual question answering, optical character recognition, image captioning, and image-to-text tasks. Includes approximately 2.9 million samples.
- Records
- 2,863,834
- Format
- PARQUET
- Downloads
- 2
OpenWebText English Web Corpus
Open-source replication of OpenAI's WebText corpus used for GPT-2 pretraining. Contains approximately 8 million English web documents suitable for language model training and research.
- Records
- 8,013,769
- Format
- PARQUET
- Downloads
- 2
HelpSteer NVIDIA Helpfulness Preference Dataset
Multi-attribute human ratings dataset for prompts and responses, including helpfulness, correctness, coherence, complexity, and verbosity assessments. Designed for training preference-aligned language models.
- Records
- 37,120
- Format
- PARQUET
- Downloads
- 1
OpenThoughts-114k Synthetic Reasoning Dataset
Synthetic reasoning examples spanning mathematics, science, code, and puzzles. Designed for fine-tuning reasoning models across multiple domains.
- Records
- 227,914
- Format
- PARQUET
- Downloads
- 3
BLIP3o Pretrain Long-Caption Dataset
Image-caption pairs with ~120-token captions generated by Qwen2.5-VL-7B-Instruct, designed for vision-language pretraining and multimodal model fine-tuning.
- Records
- 101,900
- Format
- PARQUET
- Downloads
- 2
ClimbMix English Pre-training Corpus
English pre-training corpus filtered and topic-mixed for language model training. Contains 1.79M records designed for efficient training with optimized token efficiency.
- Records
- 1,794,054
- Format
- PARQUET
- Downloads
- 3
MATH Competition Mathematics Problems Benchmark
Competition mathematics problems from AMC 10/12, AIME, and similar contests with full worked solutions. Standard benchmark for evaluating mathematical reasoning and chain-of-thought capabilities in language models.
- Records
- 12,500
- Format
- PARQUET
- Downloads
- 2
AgentTrove Agentic Interaction Traces
Collection of agentic interaction traces spanning code repair, shell scripting, math, competitive programming, and computer-use tasks, aggregated from 219 source datasets.
- Records
- 1,696,847
- Format
- PARQUET
- Downloads
- 3
Sangraha Indic Language Pretraining Corpus
Indic-language pretraining text across 22 languages, combining web sources and machine-translated corpora. Translated portions may contain synthetic artifacts, and coverage is uneven across languages.
- Records
- 177,355,164
- Format
- PARQUET
- Downloads
- 2
OpenAssistant Conversations (OASST1) Multilingual
Human-generated assistant conversation messages across 35 languages with quality ratings. Designed for alignment, RLHF, and instruction-tuning research.
- Records
- 88,838
- Format
- PARQUET
- Downloads
- 2
BEIR MS MARCO Passage Retrieval Benchmark
MS MARCO subset of the BEIR heterogeneous IR benchmark in standard retrieval layout. English-language passages and queries for passage retrieval evaluation.
- Records
- 9,351,785
- Format
- PARQUET
- Downloads
- 2
DataComp-1B Image-Text Pair Metadata
Metadata including URLs, captions, and CLIP features for image-text pairs from the DataComp-1B curated subset of CommonPool, used to train CLIP models.
- Records
- 1,387,173,656
- Format
- PARQUET
- Downloads
- 5
UltraFeedback Preference Dataset
Fine-grained preference annotations across instruction-following, truthfulness, honesty, and helpfulness, designed for reward model and RLHF training.
- Records
- 63,967
- Format
- PARQUET
- Downloads
- 2
Dolci Instruct SFT Mixture
Multilingual instruction-tuning dataset spanning 70+ languages, used to train AI2's Olmo 3 7B Instruct model. Includes OpenThoughts 3 and curated prompt sets.
- Records
- 2,152,112
- Format
- PARQUET
- Downloads
- 5
MINT-1T ArXiv Multimodal Documents
Multimodal interleaved text-and-image documents extracted from ArXiv papers, designed for pretraining. Contains 7,300 papers with synchronized text and image content.
- Records
- 7,300
- Format
- PARQUET
- Downloads
- 3
MMFineReason Multimodal Reasoning Traces
Multimodal reasoning dataset pairing images and text with detailed chain-of-thought annotations for math, science, and STEM visual reasoning tasks. Distilled from Qwen3-VL-235B-Thinking model outputs.
- Records
- 1,810,926
- Format
- PARQUET
- Downloads
- 3
WikiText-2 & WikiText-103 Language Modeling Corpora
Benchmark corpus of verified Good and Featured Wikipedia articles in raw and tokenized variants, used for language modeling evaluation.
- Records
- 3,708,608
- Format
- PARQUET
- Downloads
- 2
KILT Knowledge-Intensive Language Tasks Benchmark
Eleven datasets spanning fact-checking, entity linking, slot filling, open-domain question answering, and dialog generation, all grounded in a unified Wikipedia snapshot. Covers approximately 3.2 million examples across multiple task types.
- Records
- 3,231,786
- Format
- PARQUET
- Downloads
- 3
LLaVA-OneVision-1.5 Instruction Data
Curated multimodal instruction-tuning examples pairing images with text, covering visual question answering, image captioning, optical character recognition, reasoning, and related tasks used to train the LLaVA-OneVision-1.5 large multimodal model.
- Records
- 21,239,576
- Format
- PARQUET
- Downloads
- 3
FineWeb-Edu Translated Multilingual Corpus
Machine-translated FineWeb-Edu corpus covering 36 European languages with 36.7M aligned documents, designed for multilingual LLM pretraining and translation training.
- Records
- 1,999,563,091
- Format
- PARQUET
- Downloads
- 4