drawais/qwen3-32b-calibration-v1

该数据集名为qwen3-32b-calibration-v1，是用于[drawais/Qwen3-32B-AWQ-INT4](https://huggingface.co/drawais/Qwen3-32B-AWQ-INT4)发布的校准数据。内容与[drawais/qwen3-14b-calibration-v1](https://huggingface.co/datasets/drawais/qwen3-14b-calibration-v1)相同，因为Qwen3系列的v1版本共享校准数据。数据集包含公共英文文本的子集，每行格式为`{ text: str, input_ids: [int] }`。数据来源于[mit-han-lab/pile-val-backup](https://huggingface.co/datasets/mit-han-lab/pile-val-backup)，发布目的是为了可审计性，允许任何人验证量化模型发布时校准步骤的输入数据。The dataset is named qwen3-32b-calibration-v1 and is used as calibration data for the [drawais/Qwen3-32B-AWQ-INT4](https://huggingface.co/drawais/Qwen3-32B-AWQ-INT4) release. It has identical content to [drawais/qwen3-14b-calibration-v1](https://huggingface.co/datasets/drawais/qwen3-14b-calibration-v1) — calibration data is shared across the Qwen3 family for v1 releases. The dataset contains a subset of public English text, with each row formatted as `{ text: str, input_ids: [int] }`. The data is derived from [mit-han-lab/pile-val-backup](https://huggingface.co/datasets/mit-han-lab/pile-val-backup) and is published for auditability, allowing anyone to verify what was fed to the calibration step of our quantized model releases.