Cosmos-Reason1 SFT dataset and benchmark

From the catalogue. Listed by one researcher from one source page.

• id: cosmos-reason1-sft-dataset-and-benchmark
• kind: language and reasoning annotation
• robot or device: video from BridgeData V2, RoboVQA, AgiBot, HoloAssist, AV, RoboFail (class: cross embodiment)
• how collected: annotation of existing video
• size as stated: HF size tag 1M<n<10M (SFT)
• hours: not stated | hours only claimed: none | episodes: not stated
• year: 2025
• organisation: NVIDIA
• licence: CC-BY-4.0 (class: open)
• access: open (class: open)
• link (dataset): https://huggingface.co/datasets/nvidia/Cosmos-Reason1-SFT-Dataset
• Hugging Face repository: nvidia/Cosmos-Reason1-SFT-Dataset, 1,631 downloads in 30 days, https://huggingface.co/datasets/nvidia/Cosmos-Reason1-SFT-Dataset
• other names: Cosmos-Reason1 SFT, RL and Benchmark datasets
• note: Embodied reasoning annotations; benchmark at nvidia/Cosmos-Reason1-Benchmark.
• download links of one episode: https://datasets.gurasees.com/episodes/nvidia/Cosmos-Reason1-SFT-Dataset.md?episode=0

Confirm the licence at the link before relying on it.