Update README.md

README.md

@@ -21,7 +21,7 @@ We build open-source NLP models and tools that help people *find and understand
 Legal documents are dense and time-consuming. Our goal is to make them more accessible by:
 - highlighting clauses that commonly reduce user rights,
 - labeling the *type* of risk (e.g., unilateral changes, arbitration),
-- enabling downstream apps to display “risk badges” and evidence-backed highlights.
+- enabling downstream apps to display "risk badges" and evidence-backed highlights.
 
 ---
 
@@ -33,8 +33,8 @@ Our models perform **multi-label classification** at the sentence/clause level:
 
 This makes the models suitable for:
 - clause highlighting in a document viewer,
-- ranking “most risky” clauses first,
-- powering a lightweight “risk badge” in a UI.
+- ranking "most risky" clauses first,
+- powering a lightweight "risk badge" in a UI.
 
 ---
 
@@ -42,7 +42,7 @@ This makes the models suitable for:
 
 We currently support **8** types of potentially unfair clauses:
 
-- **Limitation of liability** — Limits the provider’s legal responsibility
+- **Limitation of liability** — Limits the provider's legal responsibility
 - **Unilateral termination** — Provider may terminate/suspend without clear cause
 - **Unilateral change** — Terms can change with minimal notice or constraints
 - **Content removal** — Provider may remove user content at discretion
@@ -62,16 +62,18 @@ We report the same metric set across models whenever possible.
 
 | Model | Task | Key metric(s) |
 |------|------|---------------|
-| **[deberta-unfair-tos](https://huggingface.co/Agreemind/deberta-unfair-tos)** | ToS clause risk classification | **F1: 0.87** • Accuracy: 78.8% ⭐ |
+| **[deberta-unfair-tos-augmented](https://huggingface.co/Agreemind/deberta-unfair-tos-augmented)** | ToS clause risk classification | **F1: 0.96** • Accuracy: 94.12% ⭐ |
+| [deberta-unfair-tos](https://huggingface.co/Agreemind/deberta-unfair-tos) | ToS clause risk classification | F1: 0.87 • Accuracy: 78.8% |
 | [electra-large-unfair-tos](https://huggingface.co/Agreemind/electra-large-unfair-tos) | ToS clause risk classification | Accuracy: 77.3% |
 | [legalbert-unfair-tos](https://huggingface.co/Agreemind/legalbert-unfair-tos) | ToS clause risk classification | Accuracy: 74.9% |
 | [modernbert-unfair-tos](https://huggingface.co/Agreemind/modernbert-unfair-tos) | ToS clause risk classification | Accuracy: 70.6% |
 | [legalbert-large-unfair-tos](https://huggingface.co/Agreemind/legalbert-large-unfair-tos) | ToS clause risk classification | Accuracy: 66.3% |
 
 **Notes**
-- “Accuracy” can mean different things for multi-label tasks (e.g., exact match vs per-label).  
-  Each model card should specify **exactly** how metrics are computed.
-- For real-world use, we recommend tuning **per-class thresholds** on your domain.
+- **Accuracy** = Exact Match (all 8 labels correct per sample)
+- **F1** = Micro-F1 across all labels
+- For production use, we recommend tuning **per-class thresholds** on your domain.
+- The augmented model was trained with 605 additional synthetic examples for weak classes.
 
 ---
 
@@ -81,7 +83,7 @@ We report the same metric set across models whenever possible.
 from transformers import AutoTokenizer, AutoModelForSequenceClassification
 import torch
 
-model_id = "Agreemind/deberta-unfair-tos"
+model_id = "Agreemind/deberta-unfair-tos-augmented"  # Best model
 tokenizer = AutoTokenizer.from_pretrained(model_id)
 model = AutoModelForSequenceClassification.from_pretrained(model_id)
 
@@ -105,7 +107,7 @@ probs = torch.sigmoid(logits).squeeze().tolist()
 
 top = sorted(zip(labels, probs), key=lambda x: x[1], reverse=True)[:3]
 print(top)
-````
+```
 
 ---
 
@@ -135,4 +137,3 @@ Always present outputs as **informational signals**, ideally with:
 ## 📄 License
 
 Models and code are released under the **MIT License**, unless otherwise stated in individual repositories/models.
-