Tag
Explore by tags
A ~3.2M-conversation Hugging Face dataset of non-toxic human–ChatGPT interactions for instruction finetuning and evaluation; includes full transcripts plus request headers, hashed IP/geolocation, turn-level moderation scores and usage metadata.
A large multi-config collection of query–document pairs assembled to reproduce and extend the mGTE/LateOn data recipe for pre-training text embedding models. Data come in source-specific configs and include per-row drop/duplicate flags and guidance for using cleaned subsets for training.
Provides an NVFP4‑optimized training and inference infrastructure for long-form video diffusion models — supports multi-shot AR training, KV-cache and NVFP4 quantized inference, sequence-parallelism and async decoding for higher FPS and longer outputs.
Provides a Gymnasium-style API and tooling to create, deploy, and interact with isolated execution environments for agentic RL training. Includes async/sync clients, a web interface, CLI, Docker-based deployment, and Hugging Face Spaces integration.
Worked examples and reusable abstractions for fine-tuning open LLMs via the Tinker training API: you write the training loop while distributed execution runs remotely. Covers SFT, math/code RL, DPO, three-stage RLHF, distillation, and tool use.
Provides 99,870 system/user/assistant chat triples for defensive cybersecurity instruction‑tuning, with built‑in refusal patterns and mapping to OWASP, MITRE ATT&CK, NIST, and CIS standards; Apache‑2.0 licensed.
Provides mined hard negatives and relevance scores for 1.88M queries across seven retrieval datasets, enabling contrastive fine-tuning and nv-retrieve filtering; includes full 2048 mined negatives per query, paired query/document splits, and parquet-formatted files for large-scale training.
Contains training, evaluation, and deployment code plus checkpoints for humanoid whole-body controllers (Decoupled WBC and GEAR‑SONIC). Includes C++ inference, VR teleoperation, data pipelines (Bones‑SEED) and Hugging Face checkpoints for research-to-robot workflows.
Provides 2 million synthetic, expert-verified coding examples with step-by-step reasoning and executable solutions for fine-tuning instruction-following and code-generation models. Curated through multi-stage filtering and automated test validation to prioritize correctness and reasoning.
Converts images (and other conditions) into high-fidelity, fully textured 3D assets using a 4B-parameter generative model and a field‑free sparse voxel format (O‑Voxel). Handles arbitrary topology, PBR materials, and near real-time mesh/voxel conversions; requires Linux and an NVIDIA GPU with >=24GB memory.
Provides a DiT-based audio–video foundation model plus an official Python inference and LoRA trainer. Ships multiple production-ready pipelines (text/image/audio→video), checkpoints, and performance optimizations (FP8, distilled pipelines) for high-fidelity synchronized audio–video generation.
Provides a conditional memory module that performs O(1) N‑gram lookups and fuses static embeddings into transformer hidden states — enables offloading large embedding tables to host memory with minimal inference overhead.