Initial commit

app.py

@@ -0,0 +1,48 @@
+import gradio as gr
+import torch
+import numpy as np
+from transformers import AutoModelForSequenceClassification
+
+# Load ONLY the model, NOT the tokenizer
+model = AutoModelForSequenceClassification.from_pretrained(
+    "Kevintu/Engessay_grading_ML")
+
+
+def process_embeddings(embeddings_array):
+    # Convert the received embeddings to the format expected by the model
+    embeddings_tensor = torch.tensor(embeddings_array)
+
+    # Process embeddings with the model
+    model.eval()
+    with torch.no_grad():
+        # Create a dict with the expected input format
+        model_inputs = {
+            'input_ids': None,  # Not needed since we're using embeddings directly
+            'attention_mask': None,  # Not needed for this use case
+            'inputs_embeds': embeddings_tensor  # Pass embeddings directly
+        }
+        outputs = model(**model_inputs)
+
+    predictions = outputs.logits.squeeze()
+
+    item_names = ["cohesion", "syntax", "vocabulary",
+                  "phraseology", "grammar", "conventions"]
+    scaled_scores = 2.25 * predictions.numpy() - 1.25
+    rounded_scores = [round(score * 2) / 2 for score in scaled_scores]
+
+    results = {item: f"{score:.1f}" for item,
+               score in zip(item_names, rounded_scores)}
+    return results
+
+
+# Create Gradio interface for embeddings input
+demo = gr.Interface(
+    fn=process_embeddings,
+    inputs=gr.JSON(label="Embeddings"),
+    outputs=gr.JSON(label="Scores"),
+    title="Essay Grading API (Embeddings Only)",
+    description="Grade essays based on precomputed embeddings"
+)
+
+demo.queue()
+demo.launch()

requirements.txt

@@ -0,0 +1,4 @@
+gradio>=3.50.2
+torch>=2.0.0
+transformers>=4.30.0
+numpy>=1.24.0