Use a symmetric quantization with no clipping error to improve llama perplexity#5163

iseeyuan wants to merge 1 commit intomainpytorch/executorch:mainfrom

noclippingpytorch/executorch:noclippingCopy head branch name to clipboard