IQ3_M quantized GGUF version of the Sugoi-32B-Ultra translation model.imatrix) generated from a high-quality Japanese-to-English translation dataset.-c 2048) to perfectly preserve the attention weights required for a 100-line rolling conversational buffer.