datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
nemotron-post-training-v2-qwen-3.5-9b-regen
Dataset Card for Nemotron Post Training v2 Qwen 3.5 9B Regen
Regenerated responses from nvidia/Nemotron-Post-Training-Dataset-v2 dataset using Qwen3.5 9B model.
Parameter
Value
Max Tokens
4096
Temperature
1.0
Top-k
20
Top-p
0.95
Repetition Penalty
1.5
Dataset consists only the english samples from the Nemotron Post Training Dataset. 85% of the chat prompts have reasoning enabled, every other category has reasoning disabled.
Category
Value
math… See the full description on the dataset page: https://huggingface.co/datasets/Dogacel/nemotron-post-training-v2-qwen-3.5-9b-regen.Swallow-Nemotron-Post-Training-Dataset-v1
Swallow-Nemotron-Post-Training-Dataset-v1
The Swallow LLM Project constructed the Swallow-Nemotron-Post-Training-Dataset-v1 based on the math, code, and stem subsets of the NVIDIA Nemotron-Post-Training-Dataset-v1, as illustrated in the figure below.
Dataset Construction
The original Thinking Trajectories and Assistant Outputs in the Nemotron-Post-Training-Dataset-v1 were synthesized using DeepSeek-R1-0528.
However, we identified an issue with the Thinking… See the full description on the dataset page: https://huggingface.co/datasets/tokyotech-llm/Swallow-Nemotron-Post-Training-Dataset-v1.Swallow-Nemotron-Post-Training-Dataset-v1-ja-cpt
Swallow-Nemotron-Post-Training-Dataset-v1-ja-cpt
Dataset Overview
This dataset is a reformatted subset of the tokyotech-llm/Swallow-Nemotron-Post-Training-Dataset-v1 dataset, specifically derived from the v1-Ja-202601 subset. It was created to facilitate Continuous Pre-Training (CPT) by extracting only the text_gpt_oss field from the original data.
Dataset Statistics & Token Counts
The token counts for each category were calculated using the… See the full description on the dataset page: https://huggingface.co/datasets/Podtech/Swallow-Nemotron-Post-Training-Dataset-v1-ja-cpt.Llama-Nemotron-Post-Training-Dataset-SFT-math-FI
Llama-Nemotron-Post-Training-Dataset-SFT-math-FI
This dataset is a Finnish machine-translated version of the SFT/math split from the original nvidia/Llama-Nemotron-Post-Training-Dataset.
The data was created by translating the original English math SFT subset into Finnish using the DeepSeek-V3 model.
Translation Process
The user prompt and the thinking traces were translated separately in two LLM requests. For traces, the <think> and </think> tokens were preserved… See the full description on the dataset page: https://huggingface.co/datasets/LumiOpen/Llama-Nemotron-Post-Training-Dataset-SFT-math-FI.Luth-2-Post-Training-SFT
Luth-2-Post-Training-SFT
Luth-2-Post-Training-SFT is the French supervised fine-tuning mixture used to train Luth-2-0.8B and Luth-2-2B. It spans math, code, knowledge, instruction following and tool calling in a single schema, with 1,969,768 examples and 3.12B training tokens.
📄 Blog: Luth-2: Pushing the French Capabilities of SLMs with MOPD
🤗 Models: Luth-2-0.8B · Luth-2-2B
📊 Datasets: SFT · RL
💻 Code: GitHub
🏆 Leaderboard: French LLM Leaderboard
Composition… See the full description on the dataset page: https://huggingface.co/datasets/kurakurai/Luth-2-Post-Training-SFT.typhoon-s-instruct-post-training
Typhoon-S Instruct Post-Training
Dataset Summary
This dataset is a post-training corpus used in the Typhoon-S recipe for building Sovereign AI: high-performing, region- and domain-specific LLMs that remain localized, controllable, and resource-efficient. It is designed to help transform a sovereignty-adapted base model into a capable assistant while preserving target-language strengths.
The dataset follows a two-part mixture philosophy:
Target-language (Thai) alignment… See the full description on the dataset page: https://huggingface.co/datasets/typhoon-ai/typhoon-s-instruct-post-training.osim-post-training
SOUL
This CMU-LTI mirror hosts the post-training data used for ODYSSIM releases. It mirrors the original sunweiwei/Soul dataset layout under the CMU-LTI organization.
SOUL is the data suite for human behavior simulation used in
Reinforcing Human Behavior Simulation via Verbal Feedback (DITTO),
spanning conversation, social simulation, social cognition, role-play, and human-centric evaluation.
📄 Paper: https://arxiv.org/abs/2605.20506
💻 Code:… See the full description on the dataset page: https://huggingface.co/datasets/cmu-lti/osim-post-training.Luth-2-Post-Training-RL
Luth-2-Post-Training-RL
Luth-2-Post-Training-RL is the French RL prompt collection used to post-train Luth-2-0.8B and Luth-2-2B.
📄 Blog: Luth-2: Pushing the French Capabilities of SLMs with MOPD
🤗 Models: Luth-2-0.8B · Luth-2-2B
📊 Datasets: SFT · RL
💻 Code: GitHub
🏆 Leaderboard: French LLM Leaderboard
Composition
Config
Rows
Verifier fields
math
20,000
prompt, solution
math_hard
17,888
prompt, solution
code
46,661
prompt, unit_tests… See the full description on the dataset page: https://huggingface.co/datasets/kurakurai/Luth-2-Post-Training-RL.nemotron-post-training-v2-gpt-oss-120b-regen
Dataset Card for Nemotron Post Training v2 gpt-oss-120b Regen
Regenerated responses from nvidia/Nemotron-Post-Training-Dataset-v2 dataset using gpt-oss-120b model.
Parameter
Value
Max Tokens
8192
Temperature
1.0
Top-k
20
Top-p
0.95
Repetition Penalty
1.5
Dataset consists only the english samples from the Nemotron Post Training Dataset. 85% of the chat prompts have reasoning enabled, every other category has reasoning disabled.
Category
Value
math… See the full description on the dataset page: https://huggingface.co/datasets/Dogacel/nemotron-post-training-v2-gpt-oss-120b-regen.nemotron-post-training-samples-splits
Nemotron Post-Training Samples with Train/Val/Test Splits
This dataset contains structured train/validation/test splits from the nvidia/Llama-Nemotron-Post-Training-Dataset, with both tagged and untagged versions for different training scenarios.
Attribution
This work is derived from the Llama-Nemotron-Post-Training-Dataset-v1.1 by NVIDIA Corporation, licensed under CC BY 4.0.
Original Dataset: nvidia/Llama-Nemotron-Post-Training-Dataset
Original Authors: NVIDIA… See the full description on the dataset page: https://huggingface.co/datasets/brandolorian/nemotron-post-training-samples-splits.typhoon-s-instruct-post-training
Typhoon-S Instruct Post-Training
Dataset Summary
This dataset is a post-training corpus used in the Typhoon-S recipe for building Sovereign AI: high-performing, region- and domain-specific LLMs that remain localized, controllable, and resource-efficient. It is designed to help transform a sovereignty-adapted base model into a capable assistant while preserving target-language strengths.
The dataset follows a two-part mixture philosophy:
Target-language (Thai) alignment… See the full description on the dataset page: https://huggingface.co/datasets/wannaphong/typhoon-s-instruct-post-training.lilm1-230m-posttraining
LiLM1-230M post-training data
This dataset contains the selected post-training data for LiLM1-230M.
Method
The records combine general assistant text with structured tool-use examples.
The configurations preserve the binding stage, the ratio study, and the
selected 4:8 continuation.
Configurations
Configuration
Content
binding-repair
Tool binding data
ratio-study
Three training splits used for ratio selection
ratio-evaluation
Shared… See the full description on the dataset page: https://huggingface.co/datasets/glouriousgautam/lilm1-230m-posttraining.nemotron-post-training-samples
Nemotron Post-Training Samples
This dataset contains random samples extracted from the nvidia/Llama-Nemotron-Post-Training-Dataset.
Attribution
This work is derived from the Llama-Nemotron-Post-Training-Dataset-v1.1 by NVIDIA Corporation, licensed under CC BY 4.0.
Original Dataset: nvidia/Llama-Nemotron-Post-Training-DatasetOriginal Authors: NVIDIA CorporationOriginal License: CC BY 4.0
Dataset Details
Source: nvidia/Llama-Nemotron-Post-Training-Dataset… See the full description on the dataset page: https://huggingface.co/datasets/brandolorian/nemotron-post-training-samples.nemotron-post-training-v2-qwen-3.5-0.8b-regen
Dataset Card for Nemotron Post Training v2 Qwen 3.5 0.8B Regen
Regenerated responses from nvidia/Nemotron-Post-Training-Dataset-v2 dataset using Qwen3.5 0.8B model.
Parameter
Value
Max Tokens
4096
Temperature
1.0
Top-k
20
Top-p
0.95
Repetition Penalty
1.5
Dataset consists only the english samples from the Nemotron Post Training Dataset. 85% of the chat prompts have reasoning enabled, every other category has reasoning disabled.
Category
Value
math… See the full description on the dataset page: https://huggingface.co/datasets/Dogacel/nemotron-post-training-v2-qwen-3.5-0.8b-regen.math500_post-training
MATH500 Subset for Small Model Post-Training
This dataset contains a subset of 20 problems from the MATH500 dataset.
The dataset contains the following columns:
problem_id: Unique problem id that corresponds to the unique_id from the MATH500 dataset.
problem: Text describing the problem the model needs to solve.
solution: The solution generated by OpenAI available in the original dataset.
answer: The ground truth answer.
subject: Problem's subject from 7 possible values (Algebra… See the full description on the dataset page: https://huggingface.co/datasets/cmpatino/math500_post-training.post-training-takehome-math500-bon16
MATH-500 Best-of-16 Post-Training Take-Home Results
A 50-problem study of test-time compute, based on the Hugging Face post-training take-home challenge. Nothing here trains or modifies a model: both the generator and the reward model stay frozen, and the only variable is how a final answer is chosen from 16 sampled candidates.
Construction
Filtered MATH-500 to levels 1-3, shuffled with seed 1, and selected 50 rows.
Generated one greedy solution per problem with… See the full description on the dataset page: https://huggingface.co/datasets/augustoFranke/post-training-takehome-math500-bon16.glm47-pie-cpp-posttraining-data
GLM-4.7-Flash PIE C++ Post-Training Data
The exact prepared dataset used for the GLM-4.7-Flash C++ performance
post-training runs.
Splits
File
Rows
Purpose
sft/train.jsonl
7,864
Supervised fine-tuning
grpo/train.jsonl
7,887
GRPO prompt and reward evaluation
eval/validation.jsonl
1,259
Full held-out evaluation
eval/validation_mini126.jsonl
126
Fast evaluation
eval/validation_mini4.jsonl
4
Smoke evaluation
tasks.tar.gz
9,146 task JSONs
Reward… See the full description on the dataset page: https://huggingface.co/datasets/TokenBender/glm47-pie-cpp-posttraining-data.math-500-qwen-2.5-rpm-post-training
Dataset Card
Dataset consists of 20 randomly sampled problems for level 1-3 in MATH-500 dataset dataset and solutions, scored by a reward model.
Dataset Details
Dataset Sources
The dataset is based on MATH-500 dataset. The methods used reproduce the approach from Hugging Face blogpost "Scaling Test Time Compute with Open Models"
Methods
We used two methods to generate solutions and answers:
Greedy decoding: sampling 1 solution using temperature=0… See the full description on the dataset page: https://huggingface.co/datasets/hannayukhymenko/math-500-qwen-2.5-rpm-post-training.