Update app.py

app.py

@@ -1,13 +1,17 @@
 import streamlit as st
 import re
 import json,os
-from openai import OpenAI
+from transformers import AutoModelForCausalLM, AutoTokenizer
 from datetime import datetime
 
-client = OpenAI(
-    api_key = os.getenv("OPENAI-API-KEY")
+# Load model and tokenizer
+model = AutoModelForCausalLM.from_pretrained(
+    "Qwen/Qwen1.5-0.5B-Chat",
+    torch_dtype="auto",
+    device_map="auto",
 )
 
+tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen1.5-0.5B-Chat")
 if 'recipe' not in st.session_state:
     st.session_state.recipe = None
 
@@ -53,7 +57,8 @@ def create_detailed_prompt(user_direction, exclusions, serving_size, difficulty)
 def generate_recipe(user_inputs):
     with st.spinner('Building the perfect recipe...'):
         functions = [
-          {
+          {'type': 'function',
+           'function':{
             "name": "provide_recipe",
             "description": "Provides a detailed recipe strictly adhering to the user input/specifications, especially ingredient exclusions and the recipe difficulty",
             "parameters": {
@@ -104,17 +109,32 @@ def generate_recipe(user_inputs):
                     ],
             },
           }
+          }
         ]
         prompt = create_detailed_prompt(user_inputs['user_direction'], user_inputs['exclusions'], user_inputs['serving_size'], user_inputs['difficulty'])
         messages = [{"role": "user", "content": prompt}]
-        st.session_state.recipe = client.chat.completions.create(
-          model="gpt-3.5-turbo",
-          messages=messages,
-          temperature=0.75,
-          top_p=0.75,
-          functions=functions,
-          function_call={"name":"provide_recipe"},  # auto is default, but we'll be explicit
+        text = tokenizer.apply_chat_template(
+            messages,
+            tokenize=False,
+            add_generation_prompt=True,
+            tools=functions
         )
+
+        # Tokenize and move to the correct device
+        model_inputs = tokenizer([text], return_tensors="pt")
+        torch.cuda.empty_cache()
+        with torch.no_grad():
+            generated_ids = model.generate(
+                model_inputs.input_ids,
+                max_new_tokens=64,
+                tools=functions# Reduced value for memory management
+            )
+
+        generated_ids = [
+            output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
+        ]
+
+        st.session_state.recipe = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
         st.session_state.recipe_saved = False
         
 def clear_inputs():