Create app.py

app.py

@@ -0,0 +1,38 @@
+import torch
+from transformers import pipeline, AutoTokenizer
+
+def load_model():
+    return pipeline("text-generation", model="TinyLlama/TinyLlama-1.1B-Chat-v1.0", torch_dtype=torch.bfloat16, device_map="auto")
+
+models = [load_model() for _ in range(3)]
+tokenizer = models[0].tokenizer
+
+# Enhanced prompt engineering (unchanged)
+messages = [
+    {
+        "role": "system",
+        "content": "You are a friendly chatbot who always responds in the style of a pirate. Use pirate vocabulary and mannerisms in your replies.",
+    },
+    {"role": "user", "content": "How many helicopters can a human eat in one sitting, matey?"},
+]
+prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
+
+# Ensemble generation with averaging (corrected)
+responses = []
+for model in models:
+    outputs = model(prompt, max_new_tokens=256, do_sample=True, temperature=0.7, top_k=50, top_p=0.95)
+    response = outputs[0]['generated_text']
+    responses.append(response)
+
+# Average the generated text directly
+averaged_text = ""
+for i in range(min(len(response) for response in responses)):
+    token_counts = {}
+    for response in responses:
+        token = response[i]
+        token_counts[token] = token_counts.get(token, 0) + 1
+    most_frequent_tokens = sorted(token_counts.items(), key=lambda x: x[1], reverse=True)
+    averaged_token = most_frequent_tokens[0][0]  # Choose the most frequent token
+    averaged_text += averaged_token
+
+print(averaged_text)