HALabs/omega-coder-v1

--- license: mit task_categories: - text-generation language: - en tags: - code - python - fine-tuning - instruction-tuning size_categories: - 1K<n<10K --- # Omega Coder v1.0 **1,729 handcrafted Python coding instruction-response pairs** for supervised fine-tuning of small language models. ## Dataset Details - **Format:** Single-column `messages` (OpenAI chat format with `system`/`user`/`assistant` roles) - **Rows:** 1,729 - **File:** `data/train-00000-of-00001.parquet` - **Size:** ~973 KB ## Intended Use Fine-tuning a small coding-focused language model (SLM) for: - Python code generation - Refactoring and code quality improvement - Bug identification and fixes - Scripting and automation tasks ## Schema Each row contains a `messages` array of dicts: ```json [ {"role": "system", "content": "..."}, {"role": "user", "content": "..."}, {"role": "assistant", "content": "..."} ] ``` ## Creation Built via a multi-phase production pipeline: topic sourcing → prompt templating → quality-gated generation → deduplication → gold-standard validation → parquet conversion. ## License MIT