Ling-Coder-SyntheticQA: Synthetic Code Generation Questions and Answers

InclusionAI created this dataset for annealing training of the Ling-Coder Lite model. It is a subset of synthetic data, part of a larger collection that includes over 5 million SFT samples and 250k DPO samples. The dataset was last updated on Hugging Face on March 27, 2025.

Use Cases

• Fine-tuning code generation models based on synthetic question-answer pairs.
• Conducting reinforcement learning from human feedback (RLHF) experiments using the described DPO data.
• Training instruction-following models for programming using the described SFT data.

Strengths

• Part of a larger collection containing over 5 million SFT samples.
• Includes a distinct DPO subset with 250k samples for preference tuning.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the SyntheticQA subset is unknown, which may limit suitability assessment.

Provenance

Source

inclusionAI on Hugging Face

Collection Method

Synthetically generated, likely for model training.

Freshness

Last updated 2025-03-27 12:39:52; freshness should be verified.