Update README.md
Browse files
README.md
CHANGED
|
@@ -13,9 +13,14 @@ metrics:
|
|
| 13 |
- accuracy
|
| 14 |
- precision
|
| 15 |
- recall
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 16 |
---
|
| 17 |
|
| 18 |
-
#
|
| 19 |
|
| 20 |
This repository contains the weights for the **SafeRoute Router**, an optimized neural router designed to dynamically direct input prompts/responses between a lightweight safety classifier (Small Model) and a high-capacity safety classifier (Large Model).
|
| 21 |
|
|
@@ -107,4 +112,4 @@ with torch.no_grad():
|
|
| 107 |
## Intended Use
|
| 108 |
|
| 109 |
- **Primary Use Case:** Guardrail optimization in LLM serving pipelines.
|
| 110 |
-
- **Out-of-Scope:** Standalone toxicity classification directly from raw text (this model requires intermediate hidden feature representations from a pre-trained small safety model).
|
|
|
|
| 13 |
- accuracy
|
| 14 |
- precision
|
| 15 |
- recall
|
| 16 |
+
datasets:
|
| 17 |
+
- montehoover/DynaBench
|
| 18 |
+
base_model:
|
| 19 |
+
- tomg-group-umd/DynaGuard-8B
|
| 20 |
+
- tomg-group-umd/DynaGuard-1.7B
|
| 21 |
---
|
| 22 |
|
| 23 |
+
# DynaRoute Router Model (DynaGuard 1.7B / 8B)
|
| 24 |
|
| 25 |
This repository contains the weights for the **SafeRoute Router**, an optimized neural router designed to dynamically direct input prompts/responses between a lightweight safety classifier (Small Model) and a high-capacity safety classifier (Large Model).
|
| 26 |
|
|
|
|
| 112 |
## Intended Use
|
| 113 |
|
| 114 |
- **Primary Use Case:** Guardrail optimization in LLM serving pipelines.
|
| 115 |
+
- **Out-of-Scope:** Standalone toxicity classification directly from raw text (this model requires intermediate hidden feature representations from a pre-trained small safety model).
|