by Sarvam AI
Vagartha is Sarvam AI's corpus of 217,959 Sanskrit verses, each paired with a structured English explanation. The explanations are model-generated and unverified, and the data is released under CC BY 4.0.
DatasetFree
Step 1: Load the dataset with load_dataset("sarvamai/vagartha", split="train").
Step 2: Use streaming=True if you do not want to download the full 562 MB.
Step 3: Read the shloka and explanation fields for each row.
Step 4: Filter by the source field with a prefix match to select a text such as the Puranas or Ayurveda.
Step 5: Check the limitations section before training, and weight or subsample the skewed distribution.
No. The card says they are model-generated and unverified, and may contain errors. It advises against using them as authoritative commentary or ground truth.
CC BY 4.0. The card notes that source editions and modern commentaries may carry their own rights.
Not always. The card says to treat the source field as a strong hint rather than a citation.
The Mahabharata and the Mahapuranas together account for about two thirds of the rows.
GDPvalGDPval is OpenAI's dataset of 220 real-world knowledge tasks across 44 occupations. It supports evaluating AI models on economically valuable work.Ultra-FineWebUltra-FineWeb is OpenBMB's filtered web pre-training dataset, with about 1T English tokens and 120B Chinese tokens. It is built by applying a verification-based quality filter to FineWeb and Chinese FineWeb data.Humanity's Last ExamHumanity's Last Exam (HLE) is a benchmark of 2,500 expert-written questions across dozens of subjects, built to test AI models at the frontier of human knowledge. It is for researchers and model builders evaluating frontier AI.