You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Originally spotted by @iamlemec in abetlen/llama-cpp-python#1089 reproduced with llama.cpp by passing --no_kv_offload to ./main. Bug causes the model to generate repeated #'s instead of a valid completion.