Leave a feedback so i can improve these models
#1
by hauser458original - opened
While testing lfm2.5-350m-code-math on coding, chat, and math prompts, I found that it outperforms ibm-granite/granite-4.0-350m in coding and math tasks. However, my fine-tuned version lacks chat capabilities; for example, the model produces gibberish when discussing general subjects. I would highly appreciate any feedback or suggestions, particularly regarding the right datasets and hyperparameters to fix this.
hauser458original changed discussion title from Leave a feedback so i can improve those models to Leave a feedback so i can improve these models