19,500 instruction-response pairs generated as rollouts from a 20-billion parameter model focused on open-source software development. The dataset covers programming tasks, debugging, and code explanation scenarios specifically formatted for supervised fine-tuning.
Use Cases
- Fine-tune language models for software engineering using the instruction and response columns
- Perform model distillation by training smaller architectures on these 20B parameter outputs
- Evaluate code-generation accuracy by comparing new model outputs against these established rollouts
Strengths
- Contains approximately 19,500 rows of instruction-following data
- Features rollouts derived from a 20-billion parameter model architecture
- Focuses on Open Source Software (OSS) development and programming tasks