Dev/model improvements

[TheComamba]

This closes #29.

[KingphilAusti]

Tested behaviour and

• { text: "Felladrin GPT2 (ca. 1 GB RAM)", path: "Felladrin/onnx-gpt2-conversational-retrain" },

worked. (Even though it showed deterministic behaviour. Which should be kinda impossible.)

The other three models do behave different:

• { text: "Felladrin Bloomz (ca. 3 GB RAM)", path: "Felladrin/onnx-bloomz-560m-sft-chat" },
• { text: "Felladrin GPT2 Large (ca. 2 GB RAM)", path: "Felladrin/onnx-gpt2-large-conversational-retrain" },
• { text: "Xenova GPT2 Large (ca. 2 GB RAM)", path: "Xenova/gpt2-large-conversational" }

Console logs 404 Error and frontend immedately starts responding again. BUT models seems to be loaded in background. Maybe we need a try catch? This leads to errors from transformers@2.11.0 afterwards.

[KingphilAusti]

Ok. Works... Tested 3 times. Posted here. Tested again to find a fix. Magically works (: But might come back some time. Whatever it was...