add openai

app.py

@@ -15,7 +15,7 @@ MODELS = {
     "LLaMA2_70B": "meta-llama/Llama-2-70b-hf",
     "ChatGLM3_6B": "THUDM/chatglm-6b",
     "InternLM_7B": "internlm/internlm-7b",
-    "Falcon_7B": "tiiuae/falcon-7b",
+    "Falcon_7B": "tiiuae/falcon-7b"
     # Add other Hugging Face models here
 }
 
@@ -27,35 +27,33 @@ def load_pipeline(task, model):
     model_name = MODELS[model]
     return pipeline(task, model=model_name)
 
-# Function to predict using Hugging Face models
+# Function to predict using Hugging Face models and OpenAI models
 def predict(task, model, text):
-    if model.startswith("ChatGPT"):
-        response = openai.Completion.create(
-            engine="text-davinci-003",
-            prompt=text,
-            max_tokens=50
+    if model in ["ChatGPT", "GPT-4"]:
+        response = openai.ChatCompletion.create(
+            model="gpt-4" if model == "GPT-4" else "gpt-3.5-turbo",
+            messages=[{"role": "user", "content": text}]
         )
-        results = [{"label": response.choices[0].text.strip()}]
+        results = [{"label": response.choices[0].message['content'].strip()}]
     else:
         selected_pipeline = load_pipeline(task, model)
         results = selected_pipeline(text)
     return results
 
-# Function to benchmark Hugging Face models
+# Function to benchmark Hugging Face models and OpenAI models
 def benchmark(task, model, file):
     data = pd.read_csv(file.name)
     texts = data['query'].tolist()
     true_labels = data['answer'].tolist()
     
-    if model.startswith("ChatGPT"):
+    if model in ["ChatGPT", "GPT-4"]:
         predictions = []
         for text in texts:
-            response = openai.Completion.create(
-                engine="text-davinci-003",
-                prompt=text,
-                max_tokens=50
+            response = openai.ChatCompletion.create(
+                model="gpt-4" if model == "GPT-4" else "gpt-3.5-turbo",
+                messages=[{"role": "user", "content": text}]
             )
-            predictions.append(response.choices[0].text.strip())
+            predictions.append(response.choices[0].message['content'].strip())
     else:
         selected_pipeline = load_pipeline(task, model)
         predictions = [selected_pipeline(text)[0]['label'] for text in texts]
@@ -74,7 +72,7 @@ def benchmark(task, model, file):
 with gr.Blocks() as demo:
     with gr.Row():
         task_input = gr.Dropdown(TASKS, label="Task")
-        model_input = gr.Dropdown(list(MODELS.keys()) + ["ChatGPT"], label="Model")
+        model_input = gr.Dropdown(list(MODELS.keys()) + ["ChatGPT", "GPT-4"], label="Model")
     
     with gr.Tab("Predict"):
         with gr.Row():

requirements.txt

@@ -4,3 +4,4 @@ pandas
 scikit-learn
 gradio
 torch
+openai