datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
c4-nanochatbpe-10B
c4-nanochatbpe-10B
C4 (en) (from allenai/c4), pre-tokenized with the nanochatbpe tokenizer (vocab 65,536)
and packaged as flat uint16 token-id .bin files for fast memmap training.
file
split
tokens
train.bin
train
10,000,000,000
val.bin
val
168,272,017
train and val are disjoint held-out partitions. Each .bin is a raw
little-endian uint16 stream (no header); token count = filesize / 2, and
train.meta.json / val.meta.json carry the full metadata. The tokenizer/files… See the full description on the dataset page: https://huggingface.co/datasets/alexkstern/c4-nanochatbpe-10B.fineweb-nanochatbpe-100M
fineweb-nanochatbpe-100M
FineWeb-Edu (from karpathy/fineweb-edu-100b-shuffle), pre-tokenized with the nanochatbpe tokenizer (vocab 65,536)
and packaged as flat uint16 token-id .bin files for fast memmap training.
This is a 100-million-token slice for data-constrained experiments. The train.bin
is the byte-exact first 100,000,000 tokens (bytes [0, 200000000)) of the parent
alexkstern/fineweb-nanochatbpe-20B
train.bin. The val.bin is byte-identical to the parent's val.bin.… See the full description on the dataset page: https://huggingface.co/datasets/alexkstern/fineweb-nanochatbpe-100M.mosaic-nanot5-512fineweb-nanochatbpe-20B
fineweb-nanochatbpe-20B
FineWeb-Edu (from karpathy/fineweb-edu-100b-shuffle), pre-tokenized with the nanochatbpe tokenizer (vocab 65,536)
and packaged as flat uint16 token-id .bin files for fast memmap training.
file
split
tokens
train.bin
train
20,000,000,000
val.bin
val
52,336,096
train and val are disjoint held-out partitions. Each .bin is a raw
little-endian uint16 stream (no header); token count = filesize / 2, and
train.meta.json / val.meta.jsoncarry the full… See the full description on the dataset page: https://huggingface.co/datasets/alexkstern/fineweb-nanochatbpe-20B.unipic_nano_2images
Skywork/unipic_nano_2images: A Multi-Image Composition Dataset
⚡ Quick Start
The image archive is split into multiple parts for easier downloading. To reconstruct and extract:
# Step 1: Concatenate split files into a single zip
cat nano-banana-2image_part_* > nano-banana-2images.zip
# Step 2: Extract the images
unzip nano-banana-2images.zip
📖 Overview
UniPic-Nano-2Images is a high-quality multi-image composition dataset containing 41,812 samples designed for… See the full description on the dataset page: https://huggingface.co/datasets/Skywork/unipic_nano_2images.unipic_nano_3images
UniPic-Nano-3Images: A Multi-Image Composition Dataset
⚡ Quick Start
The image archive is split into multiple parts for easier downloading. To reconstruct and extract:
# Step 1: Concatenate split files into a single zip
cat nano-banana.part_* > nano-banana-3images.zip
# Step 2: Extract the images
unzip nano-banana-3images.zip
📖 Overview
UniPic-Nano-3Images is a high-quality multi-image composition dataset containing 35,394 samples designed for training… See the full description on the dataset page: https://huggingface.co/datasets/Skywork/unipic_nano_3images.nemotron-nano-30b-miniswe-swebench-verified
Nemotron Nano 30B + mini-swe-agent SWE-bench Verified Trajectories
Agent trajectories from running NVIDIA Nemotron 3 Nano 30B A3B (MoE, 8B active params) on SWE-bench Verified using mini-swe-agent.
⚠️ Incomplete Run
This benchmark was terminated early due to poor performance. The model struggled with the agentic coding task.
Model Information
Attribute
Value
Model
NVIDIA Nemotron 3 Nano 30B A3B
Architecture
MoE (30B total, 8B active)
Serving
vLLM… See the full description on the dataset page: https://huggingface.co/datasets/pankajmathur/nemotron-nano-30b-miniswe-swebench-verified.nano4m-Audio
nano4M-Audio — Team (week-1)
Week-1 data preparation for nano4M-Audio, an extension of EPFL's
nano4M (the educational nano version of
4M / 4M-21)
that adds audio as a fifth modality alongside RGB, depth, surface normals
and captions.
This dataset covers all 12 VGGSound classes assigned to the three-person team:
person
classes
1 (Hassan)
lions roaring, horse neighing, pig oinking, cow lowing
2 (Ziyad)
dog barking, cat meowing, coyote howling, elephant trumpeting
3… See the full description on the dataset page: https://huggingface.co/datasets/zed-m97/nano4m-Audio.fineweb-edu-dedup6mNanoData
Dataset Description
To facilitate researchers to use NanoLM for comparative analysis across different model designs, we build a curated pre-training dataset from those of existing large-scale models (i.e., Llama, Falcon, GPT-3). It covers diverse domains to improve the generalization capabilities of the resultant models.
Dataset Creation
The data is mainly post-processed and filtered from RedPajama and RedPajamaV2.
We develop a series of cleaning steps to remove redundant… See the full description on the dataset page: https://huggingface.co/datasets/CofeAI/NanoData.nanoindex-treesnemotron-nano-eval-logs-and-scoresgithub-code-nanochatbpe-1B
github-code-nanochatbpe-1B
GitHub Code (all-all) (from codeparrot/github-code), pre-tokenized with the nanochatbpe tokenizer (vocab 65,536) and packaged as flat uint16 token-id .bin files for fast memmap training.
file
split
tokens
train.bin
train
1,000,000,000
val.bin
val
10,000,000
train and val are disjoint held-out partitions. Each .bin is a raw little-endian uint16 stream (no header); token count = filesize / 2, and train.meta.json / val.meta.json carry… See the full description on the dataset page: https://huggingface.co/datasets/alexkstern/github-code-nanochatbpe-1B.RetroKV-Fig-Dataopenpii-masking-nano-1k
OpenPII Nano: Multilingual PII Masking Sample
A nano-sized stratified sample of OpenPII 1.5M,
perfect for quick prototyping, smoke tests, and CI fixtures. Every locale and every
label that exists in the parent dataset is represented in proportion.
📖 More information: www.ai4privacy.com/datasets/pii-masking-3m-asia-pacific
Dataset Details
Total Examples
Train
Validation
Labels
Languages
Regions
Annotations
Format
License
1,000
900
100
19
30
37
7… See the full description on the dataset page: https://huggingface.co/datasets/ai4privacy/openpii-masking-nano-1k.nanonet-expnanochat-rtx4070-sft-mixes
nanochat-rtx4070 SFT mixes
Eight SFT data mixes that were trained and evaluated on a single RTX 4070, and the results each one produced. Seven of them failed.
These are the actual independent variable behind the negative-results table in Bl4ckd09/nanochat-on-rtx4070. Every mix here was built deterministically, trained on the same frozen backbone with the same geometry and step count, and put through the same two-stage evaluation gate. Publishing only the winner would make the… See the full description on the dataset page: https://huggingface.co/datasets/Marcolini/nanochat-rtx4070-sft-mixes.NanoNetpii-masking-nano-1k
PII Masking Nano: Multilingual Sample
A nano-sized stratified sample of pii-masking-openpii-1.5m,
the flagship release of the PII-Masking-3M family. Sampled proportionally by
(source_dataset, language) so every locale and label gets representation.
Asia Pacific rows appear first.
📖 More information: www.ai4privacy.com/datasets/pii-masking-3m-asia-pacific
Dataset Details
Total Examples
Train
Validation
Labels
Languages
Regions
Annotations
Format… See the full description on the dataset page: https://huggingface.co/datasets/ai4privacy/pii-masking-nano-1k.nanochat-german-alpaca
nanochat: German Alpaca Dataset
The Alpaca Cleaned Dataset is used for translation with Gemini 2.0 Flash.
Around 3,443 entries in the dataset had errors and were removed. The resulting dataset has 50,469 entries.
nano_chat
Dataset Card for "nano_chat"
Dataset Summary
nano_chat is a synthetic dataset consisting of 2326 short dialogues in simple, learner-friendly English. It was generated using Google's Gemini 2.5 flash model and is designed for training tiny conversational language models in low-resource settings.
Each dialogue simulates a realistic conversation between two speakers (A and B), using short sentences, simple grammar, and occasional small mistakes to help models generalize… See the full description on the dataset page: https://huggingface.co/datasets/sixf0ur/nano_chat.nanoJEPA-base
nanoJEPA EN/ZH Ultra-FineWeb Dataset
This is a small pretraining dataset package for nanoJEPA. It is built by
streaming openbmb/Ultra-FineWeb split en and/or zh.
Files
train.jsonl: {"text": "...", "source": "...", "dataset": "...", "language": "en|zh", "score": 0.0}
valid.jsonl: same schema as train.jsonl
test.jsonl: same schema as train.jsonl
Generation Command
uv run python data/build_hf_dataset.py \
--out-dir dataset/nanojepa-small \
--languages en,zh… See the full description on the dataset page: https://huggingface.co/datasets/TerenceLau/nanoJEPA-base.BCE-Prettybird-Nano-Themis-v0.1
BCE-Prettybird-Nano-Themis-v0.1 Synthetic Multi- Law Dataset (400 Examples)
BCE-Prettybird-Nano-Themis-v0.1 Synthetic Multi-Law Dataset is a 400-example synthetic dataset developed by Prometech AŞ for experimentation with legal reasoning, instruction following, structured generation, and multi-dimensional response evaluation. Each example combines a task-specific instruction with structured reasoning and quality metadata, including BCE signals, truth and quality values… See the full description on the dataset page: https://huggingface.co/datasets/pthinc/BCE-Prettybird-Nano-Themis-v0.1.nemotron3-nano-kd-corpus
Nemotron 3 Nano KD Corpus
4,302 verified reasoning traces for coding problems, generated by
DeepSeek-V4-Flash (284B) and
filtered by executing the generated code against real test suites.
Built entirely on-premises on a single HP ZGX Fury (NVIDIA GB300 Grace Blackwell Ultra).
No cloud APIs were used at any stage.
Used to distill NVIDIA-Nemotron-3-Nano-30B-A3B-BF16,
raising held-out pass@1 from 23.0% to 63.4%. The corpus is model-agnostic and should be
usable for any student.… See the full description on the dataset page: https://huggingface.co/datasets/curtburk/nemotron3-nano-kd-corpus.nanoGentzen
nanoGentzen Synthetic Deduction Dataset (200k Transitions)
The nanoGentzen Dataset is a formal synthetic dataset designed to train Policy-Value Transformers for automated theorem proving in Intuitionistic Logic (LI) and Classical Logic (LK via Glivenko's Theorem) using Gentzen Sequent Calculus.
Each record represents a single state-action derivation transition along an AND-OR proof search tree, providing multi-task supervision for inference rule selection, antecedent premise… See the full description on the dataset page: https://huggingface.co/datasets/Sagicc/nanoGentzen.unipic_nano_2images
Skywork/unipic_nano_2images: A Multi-Image Composition Dataset
⚡ Quick Start
The image archive is split into multiple parts for easier downloading. To reconstruct and extract:
# Step 1: Concatenate split files into a single zip
cat nano-banana-2image_part_* > nano-banana-2images.zip
# Step 2: Extract the images
unzip nano-banana-2images.zip
📖 Overview
UniPic-Nano-2Images is a high-quality multi-image composition dataset containing 41,812 samples… See the full description on the dataset page: https://huggingface.co/datasets/OJ-1/unipic_nano_2images.nanoim-symbolic
nanoIM Symbolic Temporal Aliasing Dataset
nanoIM is a small, synthetic, symbolic dataset for studying temporal aliasing in interaction models. Paired examples can share the same flattened transcript while requiring different target actions because timing, overlap, visual cues, policy events, or background/tool results differ.
Files
File
Hub config
Purpose
data/mini/{train,validation,test}.jsonl
mini
Quick smoke suite for training and evaluation.… See the full description on the dataset page: https://huggingface.co/datasets/jlov7/nanoim-symbolic.BCE-Prettybird-Nano-OWL-v0.1
BCE-Prettybird-Nano-OWL-v0.1 - 630 Translates for Instruction-Based Learning
You can leverage our Hugging Face–ready nano translation dataset, which covers a diverse set of languages including Turkish, English, German, French, Spanish, Italian, Portuguese, Dutch, Russian, Ukrainian, Polish, Czech, Slovak, Hungarian, Romanian, Bulgarian, Greek, Arabic, Persian, Hebrew, Hindi, Bengali, Urdu, Tamil, Telugu, Kannada, Malayalam, Chinese, Japanese, Korean, Indonesian, Malay, Thai… See the full description on the dataset page: https://huggingface.co/datasets/pthinc/BCE-Prettybird-Nano-OWL-v0.1.nanoclaw-tracesBCE-Prettybird-Nano-Hephaistos-v0.1
BCE-Prettybird-Nano-Hephaistos-v0.1 - 1390 Robotics for Instruction-Based Learning
BCE-Prettybird-Nano-Hephaistos-v0.1 – 1390 Robotics for Instruction-Based Learning is a bilingual Turkish–English math, sensor, robotics, and embedded-systems QA dataset designed for instruction-based learning, small language models, edge AI research, and robotics education. The dataset focuses on foundational and applied robotics topics such as linear and circular motion, forward and inverse… See the full description on the dataset page: https://huggingface.co/datasets/pthinc/BCE-Prettybird-Nano-Hephaistos-v0.1.