Sarvam 105B Distill 100K: Single-Turn Reasoning Dataset

100,000 single-turn question-answer pairs distilled from the Sarvam 105B model. The dataset is split into 96,000 training, 2,000 validation, and 2,000 test examples. It was created by author abhinav0231 and last updated on Hugging Face in May 2026.

Use Cases

• Fine-tuning language models for science and math reasoning based on the described distillation data.
• Evaluating model performance on coding and computer science problems using the provided test split.
• Benchmarking AI assistants on law, health, history, geography, and economics questions derived from the source model.

Strengths

• Contains 100,000 total examples, providing a substantial corpus for training and evaluation.
• Includes a structured split with 96,000 training, 2,000 validation, and 2,000 test rows for model development.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the full dataset is unknown, which may limit suitability assessment.

Provenance

Source

Distilled from the Sarvam 105B model.

Collection Method

Likely generated via a knowledge distillation pipeline.

Freshness

Last updated 2026-05-10 16:27:13; freshness should be verified.