Qwen3-4B-Thinking-2507-llmc-awq-calib-math-code-science-w4a16-g128-n256-s1024

• method: awq
• weight format: W4A16
• group size: 128
• calibration dataset: nvidia/Llama-Nemotron-Post-Training-Dataset
• calibration split: math,code,science
• calibration samples: 256
• max sequence length: 1024