Asbyx/MNLP_M3_quantized_dataset

Calibration Dataset for Qwen3-0.6B SmoothQuant

This dataset contains the calibration samples used for quantizing Qwen3-0.6B-Base using SmoothQuant.

Source

• Original Dataset: allenai/ai2_arc (ARC-Challenge)
• Split: train (512 random samples)
• Usage: Calibration for activation scale computation in SmoothQuant quantization

Format

• text: The question text used for calibration
• source: Source dataset name
• split: Original split from source dataset

Sampling

512 random samples were selected from the training set using a fixed random seed (42) for reproducibility.

Usage

This dataset was used to compute activation scales for SmoothQuant quantization of Qwen3-0.6B-Base model.