This gemma model was trained 2x faster with and 71% less Memory use. By using unsloth and Huggingface's TRL library.
4-bit
16-bit
Base model