Qwen3-4B-Thinking-2507-llmc-awq-calib-ultrachat-200k-train-sft-w4a16-g128-n256-s1024

• method: awq
• weight format: W4A16
• group size: 128
• calibration dataset: HuggingFaceH4/ultrachat_200k
• calibration dataset config: ``
• calibration split: train_sft
• calibration samples: 256
• max sequence length: 1024