Quantization made by Richard Erkhov.
This is the fully trained version (with fixed formatting!!).
Dataset used:
Gryphe/Sonnet3.5-SlimOrcaDedupCleaned which was further
filtered to remove prompts/examples that are longer than 4076 tokens (removed about 385 examples).
Trained with regular LoRA (not quantized/QLoRA) and LoRA rank was 128 and Alpha set to 32. Trained for 1 epoch using A40 for about 23 hours.
This llama model was trained 2x faster with
Unsloth and Huggingface's TRL library.
