Initial release — mirrors v1 calibration shared across Qwen3 family

README.md

@@ -0,0 +1,28 @@
+---
+license: other
+language:
+  - en
+tags:
+  - calibration
+  - quantization
+size_categories:
+  - n<1K
+---
+
+# qwen3-14b-calibration-v1
+
+Calibration data used for the [`drawais/Qwen3-14B-AWQ-INT4`](https://huggingface.co/drawais/Qwen3-14B-AWQ-INT4) release.
+
+Subset of public English text. Each row: `{ "text": str, "input_ids": [int] }`.
+
+## Source
+
+Derived from [`mit-han-lab/pile-val-backup`](https://huggingface.co/datasets/mit-han-lab/pile-val-backup) — a public English-text dataset commonly used for LLM calibration.
+
+## Provenance
+
+This dataset is published for **auditability**: anyone can verify what was fed to the calibration step of our quantized model releases.
+
+## License
+
+Please respect the original license of the dataset.