meteorain/Qwen_Qwen3-4B-Thinking-2507_PTQ_AUTOROUND_INT3-asym_qwen3-cot-traces
04
Strip orphan weight_scale/weight_zero_point tensors so the BF16 checkpoint loads cleanly in vLLM (no quantization_config in config.json).
Upload Qwen_Qwen3-4B-Thinking-2507_PTQ_AUTOROUND_INT3-asym_qwen3-cot-traces
initial commit
