This is the MT-10.7B model released by Google for handling machine translation tasks. The google model was released in fp32, here we release an fp16 version to handle deployment within Yurts.
One thing to note:
- As this is a T5 model architecture, the tokenizer gets really wonky with "\n" in the text string. So, replace your "\n" with " " in your text strings before sending it to the model"
- Also, the model has been trained with a max context window of 512 tokens.
