finetuned with LoRA r=128, alpha 16, lr 1e-4 to 3e-5, microbsz 3, 8 GPUs for 62 epochs with raccoon-v5.9-rlhf-full_text_document

moved files around

removed old 3b

new model

new tokenizer and model

new model

Raccoon V5 1.2 loss

v5

added tokenizer

models

initial commit