Dolci Instruct SFT No Tools 100K: Synthetic Instruction Dataset for LLM Fine-Tuning

mlx-community provides a dataset of 100,000 synthetic instruction-response pairs, last updated in January 2026. The data is designed for supervised fine-tuning of large language models without tool-use capabilities, specifically for use with the MLX-LM and MLX-LM-LoRA frameworks.

Use Cases

• Supervised fine-tuning of language models based on the provided instruction-response pairs.
• Training LoRA adapters for parameter-efficient model customization.
• Benchmarking instruction-following capabilities of models trained without tool-use data.

Strengths

• Contains 100,000 instruction-response pairs for training.
• Optimized for the MLX framework, as indicated by platform tags and examples.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is known, but data source, collection method, and potential biases are unspecified.

Provenance

Source

mlx-community on Hugging Face.

Collection Method

Synthetic generation method is not specified in the provided description.

Time Range

Creation date is not specified.

Freshness

Last updated 2026-01-31 00:00:08; freshness should be verified.

Geography

Spatial coverage is not specified.