Trained using LoRA with r=32 for 1 epoch on 208k examples of chats from HuggingFaceH4/ultrachat_200k with max_seq_len 16384.