Fix inference

README.md

@@ -22,14 +22,16 @@ widget:
   example_title: Nervousness Example
 inference:
   parameters:
-    type: text-classification
-    return_type: list
+    task: text-classification
+    output_type: list
     script: inference.py
+base_model:
+- google-bert/bert-base-uncased
 ---
 
 # GoEmotions BERT Classifier
 
-This is a fine-tuned **BERT-base-uncased** model for multi-label emotion classification on the [GoEmotions dataset](https://huggingface.co/datasets/google-research-datasets/go_emotions), predicting 28 emotions (e.g., admiration, anger, joy, neutral).
+Fine-tuned [BERT-base-uncased](https://huggingface.co/bert-base-uncased) on [go_emotions](https://huggingface.co/datasets/go_emotions) for multi-label classification (28 emotions).
 
 ## Model Details
 
@@ -43,6 +45,15 @@ This is a fine-tuned **BERT-base-uncased** model for multi-label emotion classif
 ## Try It Out
 For accurate predictions with optimized thresholds, use the [Gradio demo](https://logasanjeev-goemotions-bert-demo.hf.space).
 
+## Inference
+
+The widget uses `inference.py` with optimized thresholds (`thresholds.json`, Micro F1: 0.6025). Try the [Gradio demo](https://logasanjeev-goemotions-bert-demo.hf.space) for a full interface.
+
+```python
+from transformers import pipeline
+classifier = pipeline("text-classification", model="logasanjeev/goemotions-bert", top_k=None)
+print(classifier("Don't get hurt"))
+
 ## Performance
 
 - **Micro F1**: 0.6025 (optimized thresholds)