`meta/meta-llama-3-70b` ignores `max_tokens`

[johny-b]

I'm pretty sure I'm sending max_tokens and:

• I get much more tokens
• I also don't see this max_tokens when looking at my prediction in the browser

When I use exactly the same code for e.g. meta/llama-2-70b this does not happen, i.e. I really get the requested number of tokens.