The `Llama-8b-fp16` perplexity and accuracy tests have been marked xfail here: https://github.com/nod-ai/amd-shark-ai/pull/2694. This issue is to track these xfail tests and the corresponding fix for the same.