An example command to fine-tune Gemma on OpenAssistant’s chat dataset can be found below. We use 4-bit quantization and QLoRA to conserve memory to target all the attention blocks' linear layers.
Some results have been hidden because they may be inaccessible to you
