VQ-Logits: Compressing the Output Bottleneck of Large Language Models via Vector Quantized Logits

Open in new window