Dolci-Think-RL-7B-2k: Reinforcement Learning Prompts for Deep Reasoning

A dataset of 2,200 prompts designed to elicit deep reasoning across math, coding, precise instruction following, and general chat. It was created by the mlx-community to train the Olmo-3-7B-Think model and was last updated on November 21, 2025. The dataset blends high-quality curated sources with filtering for deliberate reasoning and is converted for compatibility with MLX-LM-LoRA.

Use Cases

• Fine-tuning language models for mathematical reasoning based on the math prompt category.
• Training models for code generation and understanding based on the coding prompt category.
• Improving model performance on precise instruction following tasks based on the dedicated prompt category.
• Developing conversational AI with enhanced reasoning capabilities based on the general chat prompt category.

Strengths

• Contains 2,200 specifically designed prompts for deep reasoning.
• Blends high-quality curated sources with filtering for deliberate reasoning.
• Converted for compatibility with the MLX-LM-LoRA training framework.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is known (2,200), but other scale metrics like file size are unknown.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

mlx-community on Hugging Face

Collection Method

Curated and filtered from high-quality sources for reinforcement learning.

Time Range

Creation date not specified; last updated November 21, 2025.

Freshness

Last updated 2025-11-21 08:42:21.

Geography

Spatial coverage is not specified.