datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
Agriculture-Agent-RL-Training-Data
Agriculture Agent RL Training Data
A growing dataset of RL rollout trajectories for LLM agents on
natural/regenerative farming — the first RL/trajectory-shaped dataset in the
Copyleft Cultivars collection
(every prior dataset here is SFT/conversational Q&A). Agents call real tools
(primarily cultivars-mcp,
a plant-genomics MCP server) across 9 knowledge categories (plus a 10th,
organic_chemistry_soil_science, added 2026-08-11, and an 11th,
organic_chemistry_synthesis, added… See the full description on the dataset page: https://huggingface.co/datasets/CopyleftCultivars/Agriculture-Agent-RL-Training-Data.Nemotron-RL-Ultra-Training-Blends
Dataset Description:
This dataset provides Reinforcement Learning (RL) and Multi-teacher On-Policy Distillation (MOPD) training-data blends used by the public Nemotron-3-Ultra post-training recipe. The blends are consumed by the NeMo RL training recipes through the NeMo Gym agent framework, in which each prompt is paired with an agent/environment that returns a verifiable or judge-based reward. Each subset is a separate blend; see the recipe for how the blends are used.
The… See the full description on the dataset page: https://huggingface.co/datasets/nvidia/Nemotron-RL-Ultra-Training-Blends.Nemotron-3-Nano-RL-Training-Blend
Dataset Description:
Nemotron-3-Nano-RL-Training-Blend is a curated dataset blend used to train the Nemotron-3-Nano-30B-A3B model. The blend consists of the following component datasets, with mixing ratios shown in parentheses:
nvidia/Nemotron-RL-instruction_following (0.17)
nvidia/Nemotron-RL-knowledge-mcqa (0.20)
nvidia/Nemotron-RL-agent-workplace_assistant (0.10)
nvidia/Nemotron-RL-instruction_following-structured_outputs (0.05)
nvidia/Nemotron-RL-coding-competitive_coding… See the full description on the dataset page: https://huggingface.co/datasets/nvidia/Nemotron-3-Nano-RL-Training-Blend.Nemotron-RL-Super-Training-Blends
Dataset Description:
Nemotron-3-Super-RL-Training-Blends contains the dataset blends used to train the Nemotron-3-Super-120B-A12B model. RL training for the Nemotron-3-Super-120B-A12B model is done in 6 stages: RLVR 1, RLVR 2, RLVR 3, SWE 1, SWE 2, and RLHF. The blends for each stage consist of data from various datasets, which we detail below. The percentages in parentheses indicate the mixing ratios of the dataset components. Note that the model was also trained on additional data… See the full description on the dataset page: https://huggingface.co/datasets/nvidia/Nemotron-RL-Super-Training-Blends.Nemotron-RL-Lightning-Training-Blend
Dataset Description:
This dataset provides the training-data blend used for the Reinforcement Learning with Verifiable Rewards (RLVR) stage of the public Nemotron-3.5-Lightning post-training recipe. The blend is consumed by the NeMo RL training recipes through the NeMo Gym agent framework, in which each prompt is paired with an agent/environment that returns a verifiable or judge-based reward. See the recipe for how the blend is used.
The blend mixes NVIDIA-released datasets… See the full description on the dataset page: https://huggingface.co/datasets/nvidia/Nemotron-RL-Lightning-Training-Blend.HyLaR_RL_Training_Dataset
HyLaR RL Training Dataset
This repository contains the reinforcement learning (RL) training dataset for HyLaR (Hybrid Latent Reasoning), as presented in the paper HyLaR: Hybrid Latent Reasoning with Decoupled Policy Optimization.
Resources
Paper: HyLaR: Hybrid Latent Reasoning with Decoupled Policy Optimization
GitHub Repository: EthenCheng/HyLaR
Model Checkpoint: HyLaR-Qwen2.5-VL-7B
Dataset Description
This dataset is designed for training… See the full description on the dataset page: https://huggingface.co/datasets/TencentBAC/HyLaR_RL_Training_Dataset.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.DRT-RL-8B-training-data
DRT RL Training Data
Paper: DRT: Dense Reasoning Trace for Efficient and Grounded Multimodal Reasoning
Original files from the DRT project's VisionR1_DAPO data directory, with parquet contents preserved. Machine-specific paths have been removed from metadata.
File
Rows
Bytes
Purpose
train.parquet
14,724
491,270,387
Original VisionR1 + DAPO training set
test.parquet
618
26,499,711
Evaluation split
mmfinereason_train.parquet
32,362
1,157,889,880
Additional… See the full description on the dataset page: https://huggingface.co/datasets/leaderonehit/DRT-RL-8B-training-data.Nemotron-RL-Super-Training-Blends
Dataset Description:
Nemotron-3-Super-RL-Training-Blends contains the dataset blends used to train the Nemotron-3-Super-120B-A12B model. RL training for the Nemotron-3-Super-120B-A12B model is done in 6 stages: RLVR 1, RLVR 2, RLVR 3, SWE 1, SWE 2, and RLHF. The blends for each stage consist of data from various datasets, which we detail below. The percentages in parentheses indicate the mixing ratios of the dataset components. Note that the model was also trained on additional… See the full description on the dataset page: https://huggingface.co/datasets/Arsh9210/Nemotron-RL-Super-Training-Blends.Nemotron-3-Nano-RL-Training-Blend-prompt-only
Nemotron-3-Nano-RL-Training-Blend-prompt-only
Prompt-only extraction from nvidia/Nemotron-3-Nano-RL-Training-Blend.
Files:
prompts.csv: one prompt extraction record per source row. Records include
prompt, separated system_prompt, and structured tools when the source row
defines available tools. Nested values are JSON-encoded inside CSV cells.
summary.md: source row counts, extracted row counts, count deltas, and failed prompt counts.
null_or_empty_rows.md: row indexes where… See the full description on the dataset page: https://huggingface.co/datasets/jamesdborin/Nemotron-3-Nano-RL-Training-Blend-prompt-only.mts-rl-training-data
Source coverage
MTS-Dialog contains 1,701 source dialogues. Fact generation succeeded for 1,700; one training dialogue was excluded after no valid fact record could be generated. Fact records are available for all 200 dialogues in test1. This fact-generation exclusion is separate from the source-dialogue eligibility rule used by the current G-Eval comparison.
PatientAgent MTS-Dialog training data
Facts and preference-training data used by the current PatientAgent… See the full description on the dataset page: https://huggingface.co/datasets/cs552-the-expendables/mts-rl-training-data.rl-execution-training-logs
RL Execution Training Logs
Per-checkpoint DQN training logs for
arXiv:2607.10960 — Reinforcement Learning
for Execution under Dynamic Fees in a Closed-Loop DEX Simulator.
Primary artifacts (model checkpoints, result CSVs, code) live in the public
repo github.com/egpivo/amm-lab
(data/rl_equilibrium/), commit 656ae56. These training logs are
supplementary reproducibility evidence — validation implementation-shortfall
(IS) curves per episode, with the selected ("best") checkpoint… See the full description on the dataset page: https://huggingface.co/datasets/egpivo/rl-execution-training-logs.rl_trainingNemotron-RL-Ultra-Training-Blends-prompt-only
Nemotron-RL-Ultra-Training-Blends-prompt-only
Prompt-only extraction from nvidia/Nemotron-RL-Ultra-Training-Blends.
Files:
prompts.csv: one prompt extraction record per source row. Records include
prompt, separated system_prompt, and structured tools when the source row
defines available tools. Nested values are JSON-encoded inside CSV cells.
summary.md: source row counts, extracted row counts, count deltas, and failed prompt counts.
null_or_empty_rows.md: row indexes where… See the full description on the dataset page: https://huggingface.co/datasets/jamesdborin/Nemotron-RL-Ultra-Training-Blends-prompt-only.gpu-forecasters-rl-training-poolCompanion artifact for GPU Forecasters: Language Models as Selective Surrogates for Kernel Runtime Optimization. Code: codezakh/gpu-surrogates.
Training data used to fine-tune the three RL surrogate variants in the paper (correctness, correctness + Brier, correctness + CRPS rewards). Each row is one (anchor, candidate) kernel pair with the candidate's measured speedup relative to the anchor.
No rows for the gdn_chunk_fwd_o pack.
Loading
from datasets import load_dataset
ds =… See the full description on the dataset page: https://huggingface.co/datasets/codezakh/gpu-forecasters-rl-training-pool.FAI-RL-1K-training-sampleNemotron-RL-Super-Training-Blends-prompt-only
Nemotron-RL-Super-Training-Blends-prompt-only
Prompt-only extraction from nvidia/Nemotron-RL-Super-Training-Blends.
Files:
prompts.csv: one prompt extraction record per source row. Records include
prompt, separated system_prompt, and structured tools when the source row
defines available tools. Nested values are JSON-encoded inside CSV cells.
summary.md: source row counts, extracted row counts, count deltas, and failed prompt counts.
null_or_empty_rows.md: row indexes where… See the full description on the dataset page: https://huggingface.co/datasets/jamesdborin/Nemotron-RL-Super-Training-Blends-prompt-only.Nemotron-3-Nano-RL-Training-Blend-STEM-Yue-Translated
Nemotron-3-Nano-RL-Training-Blend
This dataset is a translated, parallel version of the nvidia/Nemotron-3-Nano-RL-Training-Blend dataset, excluding the instruction following subset.
Dataset Origin & Sampling
The original NVIDIA dataset blends multiple sources. For this parallel corpus, we sampled 1/3 of the datasets except for the instruction following subset where we extracted all unique prompts from the unsampled dataset to avoid duplicates:
Subset /… See the full description on the dataset page: https://huggingface.co/datasets/jed351/Nemotron-3-Nano-RL-Training-Blend-STEM-Yue-Translated.training_rl_datarl_training_dataset_10kmm_reranker_rl_trainingrl-training-logsNemotron-3-Nano-RL-Training-Blend
Dataset Description:
Nemotron-3-Nano-RL-Training-Blend is a curated dataset blend used to train the Nemotron-3-Nano-30B-A3B model. The blend consists of the following component datasets, with mixing ratios shown in parentheses:
nvidia/Nemotron-RL-instruction_following (0.17)
nvidia/Nemotron-RL-knowledge-mcqa (0.20)
nvidia/Nemotron-RL-agent-workplace_assistant (0.10)
nvidia/Nemotron-RL-instruction_following-structured_outputs (0.05)
nvidia/Nemotron-RL-coding-competitive_coding… See the full description on the dataset page: https://huggingface.co/datasets/Aitherium/Nemotron-3-Nano-RL-Training-Blend.Eval4-ManualSplit-2025_02_06_error_supervisor_rl_training_data.1eval-ManualSplit-2025_02_05_initial_error_supervisor_rl_training_data.3Eval-2025_02_06_error_supervisor_rl_training_data.1Eval2-ManualSplit-2025_02_06_error_supervisor_rl_training_data.1Eval3-ManualSplit-2025_02_06_error_supervisor_rl_training_data.1Nemotron-3-Nano-RL-Training-Blend
Dataset Description:
Nemotron-3-Nano-RL-Training-Blend is a curated dataset blend used to train the Nemotron-3-Nano-30B-A3B model. The blend consists of the following component datasets, with mixing ratios shown in parentheses:
nvidia/Nemotron-RL-instruction_following (0.17)
nvidia/Nemotron-RL-knowledge-mcqa (0.20)
nvidia/Nemotron-RL-agent-workplace_assistant (0.10)
nvidia/Nemotron-RL-instruction_following-structured_outputs (0.05)
nvidia/Nemotron-RL-coding-competitive_coding… See the full description on the dataset page: https://huggingface.co/datasets/Arsh9210/Nemotron-3-Nano-RL-Training-Blend.eval-2025_02_05_initial_error_supervisor_rl_training_data.5