Update README.md
Browse files
README.md
CHANGED
|
@@ -31,6 +31,17 @@ Instead of focusing solely on the final answer, Med-REFL improves the model's in
|
|
| 31 |
|
| 32 |
This repository contains the LoRA weights produced by the Med-REFL framework for various base models.
|
| 33 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 34 |
# <span>Available Weights</span>
|
| 35 |
The Med-REFL LoRA weights can be applied to the following base models to enhance their medical reasoning abilities.
|
| 36 |
|
|
|
|
| 31 |
|
| 32 |
This repository contains the LoRA weights produced by the Med-REFL framework for various base models.
|
| 33 |
|
| 34 |
+
# <span>MedReason-8B Model Performance</span>
|
| 35 |
+
|
| 36 |
+
The following table shows the performance of the MedReason-8B model on In-Domain and Out-of-Domain benchmarks before and after applying Med-REFL.
|
| 37 |
+
|
| 38 |
+
| Domain | Benchmark | Original | **+ Med-REFL** |
|
| 39 |
+
| :--- | :--- | :--- | :--- |
|
| 40 |
+
| **In-Domain** | MedQA-USMLE | 66.27 | **70.16** <span style="color: #2E8B57; font-size: small;">(+3.89)</span> |
|
| 41 |
+
| **Out-of-Domain**| MedMCQA | 58.98 | **59.78** <span style="color: #2E8B57; font-size: small;">(+0.80)</span> |
|
| 42 |
+
| **Out-of-Domain**| GPQA (Med+) | 45.64 | **49.84** <span style="color: #2E8B57; font-size: small;">(+4.20)</span> |
|
| 43 |
+
| **Out-of-Domain**| MMLU-Pro (Med+) | 59.14 | **62.51** <span style="color: #2E8B57; font-size: small;">(+3.37)</span> |
|
| 44 |
+
|
| 45 |
# <span>Available Weights</span>
|
| 46 |
The Med-REFL LoRA weights can be applied to the following base models to enhance their medical reasoning abilities.
|
| 47 |
|