Good afternoon!
You can call me an idiot, but please tell me what happened with text generation?
Why a week ago my prompt gave me great answers, while now - just mush comes.
What has changed? Where to read about it?
Same prompt and difference in a week... Either the model has stopped understanding what I want from it, or something has changed in the methods/calls of command-r-plus-08-2024.
My current prompt includes examples for the model, for example:
Example 1 Original:
You can safely test each of these strategies at Playfina Casino Australia — they've got different game variations in the demo mode.
Example 1 Rewritten:
Feel free to experiment with every single strategy at Playfina Casino Australia (they’re loaded with demo-friendly game options—no account needed if you just wanna peek around).
Example 2 Original:
It's also called TAG and separates winners from other players. You need to be picky with your hands and then bet hard when you've got something good.
Example 2 Rewritten:
Some folks call this the TAG style—it’s what makes real winners stand out. (To be honest, I'm super picky about my opening hands, then I slam down bets once I sense a strong combo.)
Model Settings:
temperature=1.0,
top_p=0.9,
top_k=40,
max_tokens=4096,
presence_penalty=0.7