model uploaded

README.md

@@ -9,38 +9,55 @@ tags:
 - hate-speech
 - sequence-classification
 - pytorch
-- safety
-- moderation
-license: other  # Update to the correct license for your model/checkpoint
+- smote
+- tf-idf
+license: other  # inherit/align with base model's license
+datasets:
+- Adnan855570/urdu-hate-speech
 ---
 
-## Urdu RoBERTa Hate Speech Classifier
+## Urdu RoBERTa Hate Speech Classifier (Balanced)
 
 - **Base model**: `urduhack/roberta-urdu-small`
 - **Task**: Binary text classification (hate vs. not_hate)
 - **Language**: Urdu (ur)
-- **Labels**:
+- **Labels**
   - 0 → `not_hate`
   - 1 → `hate`
 
-This model fine-tunes a small RoBERTa for Urdu to detect hate speech. It is intended for content moderation, research, and educational uses. Do not use as the sole basis for enforcement or punitive actions.
+This model fine-tunes a small RoBERTa for Urdu hate-speech detection. Class imbalance was addressed by oversampling with SMOTE at the feature level (TF–IDF) prior to tokenization-based training.
+
+### Training data and preprocessing
+- Source dataset: `Adnan855570/urdu-hate-speech` (Excel files: `preprocessed_combined_file (1).xlsx`, `Urdu_Hate_Speech.xlsx`)
+- Columns used in notebook: `Tweet` (text), `Tag` (label in {0,1})
+- Steps:
+  - TF–IDF featurization (max_features=10000)
+  - SMOTE oversampling (random_state=42) to balance classes
+  - Train/test split: 80/20 (random_state=42)
+  - Tokenization: `AutoTokenizer.from_pretrained("urduhack/roberta-urdu-small")` with `truncation=True`, `padding=True`
+
+### Training setup
+- Model: `AutoModelForSequenceClassification` with `num_labels=2`
+- Device: GPU if available
+- Hyperparameters:
+  - epochs: 3
+  - per_device_train_batch_size: 8
+  - per_device_eval_batch_size: 8
+  - warmup_steps: 500
+  - weight_decay: 0.01
+  - evaluation_strategy: epoch
+  - save_strategy: epoch
+  - load_best_model_at_end: true
+- Metrics:
+  - Accuracy, Precision, Recall, F1 (binary)
 
-### Intended uses and limitations
+### Evaluation results (test split)
+- accuracy: 0.7891
+- f1: 0.7854
+- precision: 0.8208
+- recall: 0.7529
 
-- Intended:
-  - Flagging potentially hateful content in Urdu text (e.g., tweets, comments)
-  - Assisting human moderators and analysts
-  - Research and educational demos
-- Limitations:
-  - May misclassify satire, reclaimed slurs, or dialectal expressions
-  - Sensitive to domain shift (platform/topic/user community)
-  - Biases may reflect the data it was trained on
-- Risks:
-  - False positives can suppress legitimate speech
-  - False negatives can miss harmful content
-- Mitigations:
-  - Use with a human-in-the-loop
-  - Monitor performance and update thresholds per deployment domain
+Note: Results derive from the balanced (SMOTE) dataset and the 80/20 split used in the notebook.
 
 ### How to use (Transformers)
 
@@ -48,151 +65,85 @@ This model fine-tunes a small RoBERTa for Urdu to detect hate speech. It is inte
 from transformers import AutoTokenizer, AutoModelForSequenceClassification
 import torch
 
-MODEL_ID = "your-username/urdu-roberta-hate"  # replace with your repo id
+MODEL_ID = "Adnan855570/urdu-roberta-hate"  # replace if different
 tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
-model = AutoModelForSequenceClassification.from_pretrained(MODEL_ID)
-model.eval()
+model = AutoModelForSequenceClassification.from_pretrained(MODEL_ID).eval()
+
+id2label = model.config.id2label or {"0":"not_hate","1":"hate"}
 
-def predict_label(text: str) -> dict:
-    inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True)
+def predict(text: str):
+    enc = tokenizer(text, return_tensors="pt", truncation=True, padding=True)
     with torch.no_grad():
-        outputs = model(**inputs)
-        probs = outputs.logits.softmax(dim=-1).squeeze().tolist()
-        pred_id = int(outputs.logits.argmax(dim=-1).item())
-        id2label = model.config.id2label
-        return {
-            "label_id": pred_id,
-            "label": id2label.get(str(pred_id), id2label.get(pred_id, str(pred_id))),
-            "scores": {"not_hate": probs[0], "hate": probs[1]},
-        }
-
-print(predict_label("یہ نفرت انگیز مواد ہے یا نہیں؟"))
+        logits = model(**enc).logits
+        probs = logits.softmax(dim=-1).squeeze().tolist()
+        pred = int(logits.argmax(dim=-1).item())
+    return {"label_id": pred, "label": id2label.get(str(pred), str(pred)),
+            "scores": {"not_hate": probs[0], "hate": probs[1]}}
+
+print(predict("یہ نفرت انگیز ہے یا نہیں؟"))
 ```
 
 Or with a pipeline:
 
 ```python
 from transformers import pipeline
-clf = pipeline("text-classification", model="your-username/urdu-roberta-hate", top_k=None)
-print(clf("یہ نفرت انگیز مواد ہے یا نہیں؟"))
-```
-
-### Inference API (no code download)
-
-- Python (requests):
-
-```python
-import os, requests
-
-API_URL = "https://api-inference.huggingface.co/models/your-username/urdu-roberta-hate"
-HEADERS = {"Authorization": f"Bearer {os.environ.get('HF_TOKEN', '')}"}
-
-def infer(text: str):
-    r = requests.post(API_URL, headers=HEADERS, json={"inputs": text}, timeout=30)
-    r.raise_for_status()
-    return r.json()  # [{label, score}, ...] OR [[{label, score}, ...]] depending on config
-
-print(infer("یہ نفرت انگیز مواد ہے یا نہیں؟"))
+clf = pipeline("text-classification", model="Adnan855570/urdu-roberta-hate", top_k=None)
+print(clf("یہ نفرت انگیز ہے یا نہیں؟"))
 ```
 
-- cURL:
+### Inference API
 
+- cURL
 ```bash
-curl -X POST \
-  -H "Authorization: Bearer $HF_TOKEN" \
-  -H "Content-Type: application/json" \
-  -d '{"inputs":"یہ نفرت انگیز مواد ہے یا نہیں؟"}' \
-  https://api-inference.huggingface.co/models/your-username/urdu-roberta-hate
+curl -X POST -H "Authorization: Bearer $HF_TOKEN" -H "Content-Type: application/json" \
+  -d '{"inputs":"یہ نفرت انگیز ہے یا نہیں؟"}' \
+  https://api-inference.huggingface.co/models/Adnan855570/urdu-roberta-hate
 ```
 
-- huggingface_hub client:
-
+- Python
 ```python
-from huggingface_hub import InferenceClient
-client = InferenceClient(model="your-username/urdu-roberta-hate", token=os.environ.get("HF_TOKEN"))
-print(client.text_classification("یہ نفرت انگیز مواد ہے یا نہیں؟"))
+import os, requests
+API_URL = "https://api-inference.huggingface.co/models/Adnan855570/urdu-roberta-hate"
+HEADERS = {"Authorization": f"Bearer {os.environ.get('HF_TOKEN','')}"}
+print(requests.post(API_URL, headers=HEADERS, json={"inputs":"..."}, timeout=30).json())
 ```
 
-### Expected input and output
-
-- Input: a single Urdu string (short to medium-length, e.g., tweet or comment)
-- Output:
-  - Transformers: logits or labels via pipeline
-  - Recommended mapping:
-    - `id2label = {"0": "not_hate", "1": "hate"}`
-    - `label2id = {"not_hate": 0, "hate": 1}`
-
-If you want 0/1 numeric outputs in an API, map `label` to `{not_hate: 0, hate: 1}`.
-
-### Preprocessing
-
-- Standard RoBERTa tokenization (`AutoTokenizer` for the base model).
-- Truncation and padding to the model max length (e.g., 128/256). Adjust as needed.
-
-### Training details
-
-- Base: `urduhack/roberta-urdu-small`
-- Objective: Cross-entropy, 2 classes
-- Hardware: CPU or single GPU
-- Hyperparameters (example; update with your actual settings):
-  - lr: 2e-5
-  - batch_size: 16
-  - epochs: 3–5
-  - max_length: 128–256
-  - weight_decay: 0.01
-  - warmup_ratio: 0.1
-
-### Data
-
-- Source: Custom Urdu hate speech dataset (e.g., tweets/comments)
-- Class balance: Please document distribution if available (helps threshold setting)
-- Cleaning: Standard text normalization as applicable
-
-### Evaluation
-
-- Metrics to report (fill in your numbers):
-  - Accuracy: TBD
-  - F1 (macro): TBD
-  - Precision/Recall (hate class): TBD
-- Suggested threshold: argmax for 2-class; for imbalanced data, consider probability threshold tuning on a validation set.
-
-### Limitations and bias
-
-- May misinterpret context, irony, or reclaimed language
-- Potential domain and demographic bias
-- Performance can degrade on long-form or code-mixed content
-
-### Responsible AI and safety
-
-- Use as an assistive tool with human review
-- Provide user appeals and error reporting
-- Regularly audit for disparities
+### Intended uses and limitations
+- Intended:
+  - Flagging potentially hateful Urdu content
+  - Assisting human moderation and research
+- Limitations:
+  - May misclassify satire, reclaimed slurs, or code-mixed content
+  - Domain shift sensitivity (platform/community/topic)
+- Risks:
+  - False positives/negatives; do not use as the sole basis for punitive actions
+- Recommendation:
+  - Use with human-in-the-loop; periodically audit outcomes and bias
 
-### Deployment tips
+### Label mapping
+Ensure the config includes:
+- `id2label = {"0":"not_hate","1":"hate"}`
+- `label2id = {"not_hate":0,"hate":1}`
 
-- Direct load in Python: `from_pretrained("your-username/urdu-roberta-hate")`
-- Render/Flask: set `MODEL_ID` to this repo id and load via `AutoTokenizer/AutoModelForSequenceClassification.from_pretrained(MODEL_ID)`
-- HF Inference API: use bearer token for private repos or higher rate limits
-- HF Space: create a Docker Space exposing `/predict` for a custom API interface
+### Reproducibility notes
+- SMOTE and split seeds: `random_state=42`
+- Tokenization: truncation and padding enabled (no explicit max_length set in notebook)
+- Hardware: single GPU (e.g., Colab)
 
 ### License
-
-- The license must be compatible with the base model and your data usage. Update the `license:` field above and add details here.
+- The model derivation should comply with the base model’s license (`urduhack/roberta-urdu-small`). Set a compatible license here once confirmed.
 
 ### Citation
-
-If you use this model, please cite the base model and your fine-tuning work.
-
 ```bibtex
-@misc{urdu_roberta_hate_2025,
-  title  = {Urdu RoBERTa Hate Speech Classifier},
-  author = {Your Name},
+@misc{urdu_roberta_hate_balanced_2025,
+  title  = {Urdu RoBERTa Hate Speech Classifier (Balanced)},
+  author = {Adnan},
   year   = {2025},
-  howpublished = {\url{https://huggingface.co/your-username/urdu-roberta-hate}}
+  howpublished = {\url{https://huggingface.co/Adnan855570/urdu-roberta-hate}}
 }
 ```
 
 ### Acknowledgements
-
-- Base model: `urduhack/roberta-urdu-small`
-- Libraries: 🤗 Transformers, PyTorch
+- Base: `urduhack/roberta-urdu-small`
+- Libraries: 🤗 Transformers, Datasets, PyTorch
+- Oversampling: SMOTE (imblearn)