LiteCoder-SFT-Terminal-preview contains 940 synthetic trajectories designed to enhance model agentic abilities. Created by Lite-Coder and released in December 2025, this dataset supports the development of small and medium-sized code agent models.
Use Cases
- Train a code agent model on 940 synthetic trajectories to improve agentic workflow execution.
- Benchmark small model performance using this dataset's agentic ability trajectories.
- Analyze the structure of synthetic trajectories for code generation tasks.
- Fine-tune a model for software engineering tasks using the provided synthetic data.
Strengths
- Contains 940 specifically crafted synthetic training samples.
- Designed for enhancing agentic abilities in code models.
- Part of a focused effort to develop capable small and medium-sized code agents.
Limitations
- Small sample size of fewer than 1,000 trajectories may limit model generalization.
- Relies entirely on synthetic data, which may not capture real-world complexity or edge cases.
Provenance
- Source
- Lite-Coder on Hugging Face.
- Collection Method
- Fully synthetic pipeline without conversion of existing datasets.
- Time Range
- null
- Freshness
- Last updated December 2025.
- Geography
- null