| model | rougeL_f | bert_f | tok_f1 |uni_prec |bi_prec |
|---------------------------------------------------------|----------|----------|---------|---------|--------|
| [MedSwin-DaRE-TIES-KD-0.7](https://huggingface.co/MedSwin/MedSwin-DaRE-TIES-KD-0.7) | 0.17 | 0.8501 | 0.22 | 0.47 | 0.14 |
| [MedSwin-NuSL-KD-0.7](https://huggingface.co/MedSwin/MedSwin-NuSL-KD-0.7)| 0.17 | 0.8500 | 0.21 | 0.47 | 0.11 |
| [MedSwin-DaRE-Linear-KD-0.7](https://huggingface.co/MedSwin/MedSwin-DaRE-Linear-KD-0.7) | 0.17 | 0.8494 | 0.22 | 0.45 | 0.14 |
| [MedSwin-TIES-KD-0.7](https://huggingface.co/MedSwin/MedSwin-TIES-KD-0.7) | 0.16 | 0.8494 | 0.18 | 0.51 | 0.20 |
| MedSwin-TA-KD-w_0.7 | 0.18 | 0.8493 | 0.23 | 0.45 | 0.14 |
| [MedSwin-TA-SFT-0.7](https://huggingface.co/MedSwin/MedSwin-TA-SFT-0.7) | 0.16 | 0.8491 | 0.18 | 0.51 | 0.20 |
| MedSwin-TIES-KD-0.6-0.6 | 0.16 | 0.8491 | 0.18 | 0.51 | 0.20 |
| [MedSwin-Dare_Ties-KD-0.75-0.7](https://huggingface.co/MedSwin/MedSwin-Dare_Ties-KD-0.75-0.7) | 0.18 | 0.8488 | 0.23 | 0.44 | 0.13 |
| MedSwin-TA-SFT-w_0.7 | 0.16 | 0.8487 | 0.18 | 0.51 | 0.20 |
| MedSwin-DaRE-Linear-SFT-0.7-0.55 | 0.15 | 0.8485 | 0.18 | 0.51 | 0.20 |
| MedSwin-DaRE-Linear-SFT-0.7 | 0.15 | 0.8482 | 0.18 | 0.51 | 0.20 |
| MedSwin-DaRE-TIES-SFT-0.7-0.7 | 0.15 | 0.8480 | 0.17 | 0.51 | 0.20 |
| MedSwin-NuSLERP-SFT-0.7 | 0.15 | 0.8478 | 0.17 | 0.52 | 0.21 |
| MedGemma-27b-Text-IT | 0.19 | 0.8465 | 0.27 | 0.40 | 0.09 |
| MedSwin-TIES-SFT-0.5-0.6 | 0.16 | 0.8456 | 0.20 | 0.48 | 0.17 |
| [MedSwin-7B-KD](https://huggingface.co/MedSwin/MedSwin-7B-LD) | 0.18 | 0.8441 | 0.26 | 0.46 | 0.12 |
| [MedSwin-7B-SFT](https://huggingface.co/MedSwin/MedSwin-7B-SFT) | 0.15 | 0.8396 | 0.18 | 0.52 | 0.21 |
| [MedSwin-KD-SFT-PubMed-l](https://huggingface.co/MedSwin/MedSwin-KD-SFT-PubMed-l) | 0.17 | 0.8341 | 0.24 | 0.44 | 0.11 |
| [MedSwin-KD-SFT-PubMed-map](https://huggingface.co/MedSwin/MedSwin-KD-SFT-PubMed-map) | 0.17 | 0.8297 | 0.22 | 0.47 | 0.12 |
| MedAlpaca-7B | 0.16 | 0.8196 | 0.18 | 0.50 | 0.20 |
> Highlighted models are featured on MedSwin's Hugging Face org.
---
> Abbrs:
> - KD: Knowledge Distillation
> - SFT: Supervised Fine-tuning
> Merges:
> - TA: Task Arithmetic
> - SLERP: Spherical Linear Interpolation
> - NuSL: Normalized Spherical Linear Interpolation
> - DaRE: Drop And REscale
> Metrics:
> - rougeL_f: Longest Common Subsequence (LCS)
> - bert_f: Semantic similarity (RoBERTa)
> - tok_f1: Token-level precision/recall F1
> - uni_prec: Unigram Precision
> - bi_prec: Bigram Precision
> Specs:
> - IT: Instruct
> - l: labelled