| ### Remarks | |
| 2nd iteration has an overall worse performance running at f16 26t/s compared to the first iteration q8_0 60t/s both gguf on ollama with rtx 2070. not sure why the 2nd iteration model (f16) gives a very different output compared to the sample test output from unsloth (i presume running from lora_model safetensors). For now, v1.2 model and dataset is discontinued and will continue further iterations with the first iteration method |