LexiMind: model merging vs. multi-task fine-tuning
Nov 2025 – presentDoes model merging (task arithmetic, TIES) match multi-task LoRA fine-tuning once you count the compute spent training each specialist? A budget-matched study on book genre, topic, and emotion classification.
Built on a transformer I implemented in PyTorch and initialized from FLAN-T5. Leakage-checked data splits and tuned TF-IDF baselines are done; the main multi-seed runs are next.
The repository also has a separate book-recommendation app that doesn’t use the research model.