Update app.py

app.py

@@ -1,11 +1,34 @@
-import streamlit as st
-from transformers import pipeline
-
-pipe = pipeline(
-    task='text-generation',
-    model = 'meta-llama/Meta-Llama-3-8B'
-)
-
-x = st.chat_input(placeholder='Your Message')
-o = pipe("Hi")
-st.write(o[0]['generated_text'])
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
+
+torch.random.manual_seed(0)
+
+model = AutoModelForCausalLM.from_pretrained(
+    "microsoft/Phi-3-mini-4k-instruct", 
+    device_map="cuda", 
+    torch_dtype="auto", 
+    trust_remote_code=True, 
+)
+tokenizer = AutoTokenizer.from_pretrained("microsoft/Phi-3-mini-4k-instruct")
+
+messages = [
+    {"role": "user", "content": "Can you provide ways to eat combinations of bananas and dragonfruits?"},
+    {"role": "assistant", "content": "Sure! Here are some ways to eat bananas and dragonfruits together: 1. Banana and dragonfruit smoothie: Blend bananas and dragonfruits together with some milk and honey. 2. Banana and dragonfruit salad: Mix sliced bananas and dragonfruits together with some lemon juice and honey."},
+    {"role": "user", "content": "What about solving an 2x + 3 = 7 equation?"},
+]
+
+pipe = pipeline(
+    "text-generation",
+    model=model,
+    tokenizer=tokenizer,
+)
+
+generation_args = {
+    "max_new_tokens": 500,
+    "return_full_text": False,
+    "temperature": 0.0,
+    "do_sample": False,
+}
+
+output = pipe(messages, **generation_args)
+print(output[0]['generated_text'])