Update app.py

app.py

@@ -1,8 +1,10 @@
 import gradio as gr
 from huggingface_hub import InferenceClient
 
+# Initialize Hugging Face Inference Client
 client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
 
+# Response Function (Now Compatible with 'type=messages')
 def respond(
     message,
     history,
@@ -11,17 +13,24 @@ def respond(
     temperature,
     top_p,
 ):
+    # Correct message format for Gradio's 'messages' type
     messages = [{"role": "system", "content": system_message}]
 
-    # Updated for OpenAI-style format (replacing tuples)
+    # Handle both old tuple format and new 'messages' format
     for entry in history:
-        role = "user" if entry["role"] == "user" else "assistant"
-        messages.append({"role": role, "content": entry["content"]})
+        if isinstance(entry, dict) and "role" in entry and "content" in entry:
+            messages.append(entry)  # Already in correct format
+        elif isinstance(entry, tuple) and len(entry) == 2:
+            messages.append({"role": "user", "content": entry[0]})
+            messages.append({"role": "assistant", "content": entry[1]})
 
+    # Add the current user message
     messages.append({"role": "user", "content": message})
 
+    # Initialize response string
     response = ""
 
+    # Generate chat response using the client
     for message in client.chat_completion(
         messages,
         max_tokens=max_tokens,
@@ -33,10 +42,10 @@ def respond(
         response += token
         yield response
 
-# Updated ChatInterface with correct type
+# Gradio Interface Setup
 demo = gr.ChatInterface(
     respond,
-    chatbot=gr.Chatbot(type="messages"),  # Correct format
+    chatbot=gr.Chatbot(type="messages"),
     additional_inputs=[
         gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
         gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
@@ -45,24 +54,26 @@ demo = gr.ChatInterface(
     ],
 )
 
+# Launch the Gradio app
 if __name__ == "__main__":
     demo.launch()
 
-# -----------------------------------------------
-# Fine-Tuning GPT-2 on Hugging Face Spaces (Improved Section)
-# -----------------------------------------------
+# Fine-Tuning GPT-2 on Hugging Face Spaces (Streaming 40GB Dataset, No Storage Issues)
 from transformers import AutoModelForCausalLM, AutoTokenizer, Trainer, TrainingArguments
 from datasets import load_dataset
 from peft import LoraConfig, get_peft_model
 import torch
 
+# Authenticate Hugging Face
 from huggingface_hub import notebook_login
 notebook_login()
 
+# Load GPT-2 model and tokenizer
 model_name = "gpt2"
 tokenizer = AutoTokenizer.from_pretrained(model_name)
 model = AutoModelForCausalLM.from_pretrained(model_name)
 
+# Custom Dataset (Predefined Q&A Pairs for Project Expo)
 custom_data = [
     {"prompt": "Who are you?", "response": "I am Eva, a virtual voice assistant."},
     {"prompt": "What is your name?", "response": "I am Eva, how can I help you?"},
@@ -73,22 +84,28 @@ custom_data = [
     {"prompt": "Who created you?", "response": "I was created by an expert team specializing in AI fine-tuning and web development."}
 ]
 
+# Convert custom dataset to Hugging Face Dataset
 dataset_custom = load_dataset("json", data_files={"train": custom_data})
+
+# Load OpenWebText dataset (5% portion to avoid streaming issues)
 dataset = load_dataset("Skylion007/openwebtext", split="train[:20%]")
 
+# Tokenization function
 def tokenize_function(examples):
     return tokenizer(examples["text"], truncation=True, padding="max_length", max_length=512)
 
 tokenized_datasets = dataset.map(tokenize_function, batched=True)
 
+# Apply LoRA for efficient fine-tuning
 lora_config = LoraConfig(
     r=8, lora_alpha=32, lora_dropout=0.05, bias="none",
-    target_modules=["c_attn", "c_proj"]
+    target_modules=["c_attn", "c_proj"]  # Apply LoRA to attention layers
 )
 
 model = get_peft_model(model, lora_config)
-model.gradient_checkpointing_enable()
+model.gradient_checkpointing_enable()  # Enable checkpointing for memory efficiency
 
+# Training arguments
 training_args = TrainingArguments(
     output_dir="gpt2_finetuned",
     auto_find_batch_size=True,
@@ -101,17 +118,22 @@ training_args = TrainingArguments(
     push_to_hub=True
 )
 
+# Trainer setup
 trainer = Trainer(
     model=model,
     args=training_args,
     train_dataset=tokenized_datasets
 )
 
+# Start fine-tuning
 trainer.train()
+
+# Save and push the model to Hugging Face Hub
 trainer.save_model("gpt2_finetuned")
 tokenizer.save_pretrained("gpt2_finetuned")
 trainer.push_to_hub()
 
+# Deploy as Gradio Interface
 def generate_response(prompt):
     inputs = tokenizer(prompt, return_tensors="pt")
     outputs = model.generate(**inputs, max_length=100)