Previous_scene breaks token output [Semi-fix]

[CHesketh76]

I ran into this issue with when introducing previous_scene into story generation. But whatever output tokens you have setup on max_token setup for 2000 and your previous scene took up 1999 tokens. Then the next scene generated will only be 1 token.

I am testing using Mixtral model for testing which has a context length of 32k, so I was shocked to see a 2k token limit.

A stupid fix is to make max_tokens = 10_000_000 and it appears to fix the issue but I haven't hit the 32k limit to see if their will be any truncation of the text.

I'll update this ticket for future users... since this repo is a ghost-town.