That model's weights are around 64GB: https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B/tree/main https://huggingface.co/deepseek-ai/DeepSeek-R1-D

That model's weights are around 64GB: https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B/tree/main https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-... GP is likely running the 4-bit quantized version of the finetuned Qwen model.