Create app.py

app.py

@@ -0,0 +1,47 @@
+import pandas as pd
+import gradio as gr
+from transformers import AutoModelForCausalLM, AutoTokenizer
+import torch
+
+# Load model and tokenizer
+model_name = "meta-llama/Meta-Llama-3.2-3B-Instruct"
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+model = AutoModelForCausalLM.from_pretrained(
+    model_name,
+    torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
+    device_map="auto"
+)
+
+def generate_prompt(original, translation):
+    return f"### Task: Machine Translation Quality Estimation\n\nSource: {original}\nTranslation: {translation}\n\nScore (0-1):"
+
+def predict_scores(file):
+    df = pd.read_csv(file.name, sep="\t")
+    scores = []
+
+    for _, row in df.iterrows():
+        prompt = generate_prompt(row["original"], row["translation"])
+        inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
+        outputs = model.generate(**inputs, max_new_tokens=10)
+        response = tokenizer.decode(outputs[0], skip_special_tokens=True)
+
+        # Extract float value (naive way)
+        score = response.split("Score")[-1].strip()
+        try:
+            score_val = float(score.split()[0])
+        except:
+            score_val = -1  # fallback
+
+        scores.append(score_val)
+
+    df["predicted_score"] = scores
+    return df
+
+iface = gr.Interface(
+    fn=predict_scores,
+    inputs=gr.File(label="Upload dev.tsv"),
+    outputs=gr.Dataframe(label="QE Output with Predicted Score"),
+    title="MT QE with LLaMA-3.2-3B-Instruct"
+)
+
+iface.launch()