Tokenizer error?
#1
by 3ndetz - opened
looks like GGUF version is too bad...
but safetensors with patched tokenizers (from which I converted) was normal.
I made some patches to llama cpp hf gguf convert to make it work (replace sentencepiece for t5 with gpt2 tokenizer, and replace pre-toketnizer to gpt2), so it launched, but seems like there is trash in model outputs