QuietImpostor/Phi-3-Mini-Opus

Phi-3 Opus

• Model info
  • Finetune of Phi-3 Mini 4k Instruct.
  • Original dataset, Converted dataset
  • Precision: BNB 4bit
  • Training info
    • Minimum loss: ~0.7
    • Maximum training loss: ~1.3773
    • Final loss: 1.2204

Phi-3 Opus is a basic finetune using the one and only Unsloth using a Claude 3 Opus based dataset. No benchmarks or anything of the sort has been conducted on this model. Oddly enough, it seems to not accept "Hello!" to be a message it'll respond to. It says it cannot provide a coherent response due to it being incomplete. However, everything else I've tested works fine.