Update with real nlp encoder codes

app.py

@@ -1,7 +1,80 @@
 import gradio as gr
+import pandas as pd
+import numpy as np
+import torch
+from transformers import (
+    AutoTokenizer,
+    AutoModel,
+    AutoModelForSequenceClassification
+)
+from scipy.special import softmax
 
-def greet(name):
-    return "Hello " + name + "!!"
+# ==============================
+# LOAD MODELS ONCE (GLOBAL)
+# ==============================
 
-demo = gr.Interface(fn=greet, inputs="text", outputs="text")
-demo.launch(share=True)
+bert_model_name = "bert-base-uncased"
+tokenizer = AutoTokenizer.from_pretrained(bert_model_name)
+bert_model = AutoModel.from_pretrained(bert_model_name)
+bert_model.eval()
+
+sentiment_model_name = "cardiffnlp/twitter-roberta-base-sentiment"
+sentiment_tokenizer = AutoTokenizer.from_pretrained(sentiment_model_name)
+sentiment_model = AutoModelForSequenceClassification.from_pretrained(sentiment_model_name)
+sentiment_model.eval()
+
+
+def nlp_encode_sentence(df: pd.DataFrame) -> pd.DataFrame:
+
+    feature_rows = []
+
+    for sentence in df["value"]:
+
+        inputs = tokenizer(sentence, return_tensors="pt", truncation=True, padding=True)
+
+        with torch.no_grad():
+            outputs = bert_model(**inputs)
+
+        cls_embedding = outputs.last_hidden_state[:, 0, :].squeeze().numpy()
+
+        embedding_mean = np.mean(cls_embedding)
+        embedding_median = np.median(cls_embedding)
+        embedding_std = np.std(cls_embedding)
+        embedding_min = np.min(cls_embedding)
+        embedding_max = np.max(cls_embedding)
+
+        sentiment_inputs = sentiment_tokenizer(
+            sentence,
+            return_tensors="pt",
+            truncation=True,
+            padding=True
+        )
+
+        with torch.no_grad():
+            sentiment_outputs = sentiment_model(**sentiment_inputs)
+
+        probs = softmax(sentiment_outputs.logits.numpy()[0])
+        sentiment_score = probs[2] - probs[0]
+
+        feature_rows.append({
+            "embedding_mean": embedding_mean,
+            "embedding_median": embedding_median,
+            "embedding_std": embedding_std,
+            "embedding_min": embedding_min,
+            "embedding_max": embedding_max,
+            "sentiment_score": sentiment_score
+        })
+
+    features_df = pd.DataFrame(feature_rows)
+
+    return features_df
+
+
+demo = gr.Interface(
+    fn=nlp_encode_sentence,
+    inputs=gr.Dataframe(),
+    outputs=gr.Dataframe(),
+    api_name="encode"
+)
+
+demo.launch()

requirements.txt

@@ -0,0 +1,6 @@
+gradio
+pandas
+numpy
+torch
+transformers
+scipy