Integer Scale: A Free Lunch for Faster Fine-grained Quantization of LLMs
This work addresses the problem of slow inference in quantized LLMs for AI practitioners, offering a plug-and-play solution that is incremental but provides significant speed gains.
The paper tackles the inference bottleneck in fine-grained quantization of large language models by introducing Integer Scale, a post-training quantization scheme that requires no extra calibration or fine-tuning, achieving up to 1.85x speed boost with comparable accuracy and resolving quantization difficulties for models like Mixtral-8x7B and LLaMA-3 with 2.13x and 2.31x speed boosts respectively.
We introduce Integer Scale, a novel post-training quantization scheme for large language models that effectively resolves the inference bottleneck in current fine-grained quantization approaches while maintaining similar accuracies. Integer Scale is a free lunch as it requires no extra calibration or fine-tuning which will otherwise incur additional costs. It can be used plug-and-play for most fine-grained quantization methods. Its integration results in at most 1.85x end-to-end speed boost over the original counterpart with comparable accuracy. Additionally, due to the orchestration of the proposed Integer Scale and fine-grained quantization, we resolved the quantization difficulty for Mixtral-8x7B and LLaMA-3 models with negligible performance degradation, and it comes with an end-to-end speed boost of 2.13x, and 2.31x compared with their FP16 versions respectively.