Tulu 3 Sft Personas: 29,980 Synthetic Instructions for Constraint Following

29,980 synthetically generated examples designed to enhance a model's ability to follow instructions precisely and satisfy user constraints. The dataset was curated by the Allen Institute for AI and uses a persona-based methodology to generate diverse instructions, with constraints borrowed from the IFEval taxonomy. It was last updated on November 21, 2024.

Use Cases

• Fine-tuning language models for precise instruction following based on the described synthetic examples.
• Benchmarking model performance on user constraint satisfaction based on the IFEval taxonomy.
• Researching the effect of persona-based instruction generation on model training.

Strengths

• Contains 29,980 examples, providing a substantial volume for training.
• Explicitly designed for a specific purpose: enhancing instruction-following and constraint satisfaction.
• Methodology is described as using personas to generate diverse instructions, potentially increasing variety.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is known but other scale details like file formats and size are unknown.
• The dataset is synthetic, which may limit real-world applicability or introduce artifacts from generation.

Provenance

Source

Allen Institute for AI (allenai)

Collection Method

Synthetically generated using an expanded persona-based methodology from Ge et al., 2024.

Time Range

null

Freshness

Last updated 2024-11-21 15:57:21.

Geography

null