lightning
lightning-boltz-dataMIA
Memory Intelligence Agent (MIA)
Paper | GitHub
MIA (Memory In Intelligence Agent) is a memory framework designed for deep research agents (DRAs). It transforms agents from "passive record-keepers" into "active strategists" using a Manager-Planner-Executor architecture. This repository contains the datasets and data artifacts used to train and evaluate the MIA framework.
Dataset Description
The dataset includes the following components:
Train: Data used for the… See the full description on the dataset page: https://huggingface.co/datasets/LightningCreeper/MIA.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.jovian_lightning
jovian_lightning (TsFile)
Apache TsFile version of phanerozoic/jovian-lightning.
Data files: ['lightning_catalog.tsfile', 'per_perijove_stats.tsfile', 'transient_candidates.tsfile']
Usage
Install the Apache TsFile Python SDK (pip install tsfile) and read a converted file:
from pathlib import Path
from tsfile import TsFileReader
path = Path("lightning_catalog.tsfile")
with TsFileReader(str(path)) as reader:
schemas = reader.get_all_table_schemas()… See the full description on the dataset page: https://huggingface.co/datasets/THULab/jovian_lightning.WWTD-2025
What Would Trump Do?
Auto-generated from 5 search queries — used to beat GPT-5
Starting from nothing but 5 search queries, we used the Lightning Rod SDK to automatically generate 2,790 forecasting questions about Trump administration actions from news articles and label them using real outcomes. No expertise required. No manual labeling. Used to train Trump-Forecaster, which beats GPT-5.
TL;DR
Generated 2,790 forward-looking forecasting questions… See the full description on the dataset page: https://huggingface.co/datasets/LightningRodLabs/WWTD-2025.Nemotron-3.5-Lightning-30B-A3B-prune-frontier-report
Nemotron-3.5-Lightning-30B-A3B — where expert pruning stops working, measured
Research evidence dataset. No model weights. Part of the collection
Xyntetik Research: Pruning and Quantization Frontiers on this account, produced with
Xyntetik Runner.
Dataset summary
Question tested. How deep nemotron_h_moe can be expert-pruned before it stops matching its parent, whether the saliency ranking matters, and whether the published Q4_0 clears the house bar.
Models… See the full description on the dataset page: https://huggingface.co/datasets/Joakimpalm-Zen/Nemotron-3.5-Lightning-30B-A3B-prune-frontier-report.