This dataset is a regenerated version of the OPC training dataset, where assistant responses have been regenerated using Qwen3-Coder-30B-A3B-Instruct as the target model.
Purpose
Regenerating training data with the target model helps better align the draft model with the target model's output distribution, improving acceptance rates and overall speculative decoding performance in SpecForge.
Dataset Statistics… See the full description on the dataset page: https://huggingface.co/datasets/JinnP/opc_regen_Qwen3-Coder-30B-A3B-Instruct.
