Dataset · bigcode
SFT dataset for StarCoder2-Instruct
bigcode/self-oss-instruct-sc2-exec-filter-50k
50k self-alignment examples with execution-verified responses for code models.
What's inside
The final self-alignment dataset for StarCoder2-Instruct. It features seeds as Python functions from the MultiPL-T pipeline, along with generated concepts, instructions, and execution-verified responses. The data is text-based with a size between 10k and 100k examples.
What you can do with it
- Fine-tune StarCoder2-Instruct on SFT examples
- Configure LoRA for Python code generation
- Evaluate the model on instructions with execution-validated responses
- Size
- 10K–100K rows
- Downloads this month
- 38.4K
- Likes
- 112
Models trained on it
- HuggingFaceTB/SmolLM-135M-InstructText generation 23.9K
- HuggingFaceTB/SmolLM-1.7B-InstructText generation 11K
- HuggingFaceTB/SmolLM-360M-InstructText generation 8.1K
- HuggingFaceTB/smollm-135M-instruct-v0.2-Q8_0-GGUF 3.2K
- ibm-granite/granite-8b-code-instruct-128kText generation 2.4K
- bartowski/SmolLM-1.7B-Instruct-v0.2-GGUFText generation 2.3K
- unsloth/SmolLM-1.7B-InstructText generation 1.8K
- bigcode/starcoder2-15b-instruct-v0.1Text generation 1.7K
How to load it
from datasets import load_dataset
ds = load_dataset("bigcode/self-oss-instruct-sc2-exec-filter-50k")