Reddit post explores quantization evaluation with KLD and perplexity

The post compares KLD, perplexity, and BPW as metrics for evaluating quantized LLMs, noting that KLD and perplexity can help rank models but may not perfectly reflect real deployment performance. Author suggests combining multiple metrics for better assessment.