Update app.py

app.py

@@ -1,63 +1,54 @@
 import gradio as gr
 from huggingface_hub import InferenceClient
 from transformers import AutoModelForCausalLM, AutoTokenizer, Trainer, TrainingArguments
-from datasets import Dataset, load_dataset
+from datasets import load_dataset
 from peft import LoraConfig, get_peft_model
 import torch
 
 # Initialize Hugging Face Inference Client
 client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
 
-# Load GPT-2 model and tokenizer
+# GPT-2 Model Setup
 model_name = "gpt2"
 tokenizer = AutoTokenizer.from_pretrained(model_name)
 model = AutoModelForCausalLM.from_pretrained(model_name)
 
-# Add padding token (GPT-2 fix)
-if tokenizer.pad_token is None:
-    tokenizer.pad_token = tokenizer.eos_token
+# Add padding token
+tokenizer.pad_token = tokenizer.eos_token
 
-# Custom Dataset (Predefined Q&A Pairs for Project Expo)
+# Custom Dataset
 custom_data = [
-    {"text": "Who are you?", "label": "I am Eva, a virtual voice assistant."},
-    {"text": "What is your name?", "label": "I am Eva, how can I help you?"},
-    {"text": "What can you do?", "label": "I can assist with answering questions, searching the web, and much more!"},
-    {"text": "Who invented the computer?", "label": "Charles Babbage is known as the father of the computer."},
-    {"text": "Tell me a joke.", "label": "Why don’t scientists trust atoms? Because they make up everything!"},
-    {"text": "Who is the Prime Minister of India?", "label": "The current Prime Minister of India is Narendra Modi."},
-    {"text": "Who created you?", "label": "I was created by an expert team specializing in AI fine-tuning and web development."}
+    {"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?"},
+    {"prompt": "What can you do?", "response": "I can assist with answering questions, searching the web, and much more!"},
+    {"prompt": "Who invented the computer?", "response": "Charles Babbage is known as the father of the computer."},
+    {"prompt": "Tell me a joke.", "response": "Why don’t scientists trust atoms? Because they make up everything!"},
+    {"prompt": "Who is the Prime Minister of India?", "response": "The current Prime Minister of India is Narendra Modi."},
+    {"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 = Dataset.from_dict({
-    "text": [d['text'] for d in custom_data],
-    "label": [d['label'] for d in custom_data]
-})
+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%]")
+# Load OpenWebText dataset with `trust_remote_code=True`
+dataset = load_dataset("Skylion007/openwebtext", split="train[:20%]", trust_remote_code=True)
 
 # Tokenization function
 def tokenize_function(examples):
-    return tokenizer(
-        examples["text"],
-        truncation=True,
-        padding="max_length",
-        max_length=512
-    )
+    return tokenizer(examples["text"], truncation=True, padding="max_length", max_length=512)
 
-tokenized_datasets = dataset_custom.map(tokenize_function, batched=True)
+tokenized_datasets = dataset.map(tokenize_function, batched=True)
 
-# Apply LoRA for efficient fine-tuning
+# LoRA Configuration
 lora_config = LoraConfig(
     r=8, lora_alpha=32, lora_dropout=0.05, bias="none",
-    target_modules=["c_attn", "c_proj"]  # Apply LoRA to attention layers
+    target_modules=["c_attn", "c_proj"]
 )
 
 model = get_peft_model(model, lora_config)
-model.gradient_checkpointing_enable()  # Enable checkpointing for memory efficiency
+model.gradient_checkpointing_enable()
 
-# Training arguments
+# Training Arguments
 training_args = TrainingArguments(
     output_dir="gpt2_finetuned",
     auto_find_batch_size=True,
@@ -70,14 +61,14 @@ training_args = TrainingArguments(
     push_to_hub=True
 )
 
-# Trainer setup
+# Trainer Setup
 trainer = Trainer(
     model=model,
     args=training_args,
     train_dataset=tokenized_datasets
 )
 
-# Start fine-tuning
+# Fine-tuning
 trainer.train()
 
 # Save and push the model to Hugging Face Hub
@@ -85,28 +76,21 @@ 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)
-    return tokenizer.decode(outputs[0], skip_special_tokens=True)
-
-# Gradio Chat Interface
+# Chatbot Response Function
 def respond(message, history, system_message, max_tokens, temperature, top_p):
     messages = [{"role": "system", "content": system_message}]
 
-    # Ensure 'history' is handled as a list of dicts
     if isinstance(history, list):
         for entry in history:
             if isinstance(entry, dict):
-                messages.append(entry)  # Correct format already
+                messages.append(entry)
             elif isinstance(entry, tuple) and len(entry) == 2:
                 messages.append({"role": "user", "content": entry[0]})
                 messages.append({"role": "assistant", "content": entry[1]})
 
     messages.append({"role": "user", "content": message})
-    response = ""
 
+    response = ""
     for message in client.chat_completion(
         messages,
         max_tokens=max_tokens,
@@ -118,6 +102,7 @@ def respond(message, history, system_message, max_tokens, temperature, top_p):
         response += token
         yield response
 
+# Gradio Chatbot Interface
 demo = gr.ChatInterface(
     respond,
     chatbot=gr.Chatbot(type="messages"),
@@ -125,16 +110,10 @@ demo = gr.ChatInterface(
         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"),
         gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
-        gr.Slider(
-            minimum=0.1,
-            maximum=1.0,
-            value=0.95,
-            step=0.05,
-            label="Top-p (nucleus sampling)",
-        ),
+        gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p (nucleus sampling)"),
     ],
 )
 
-# Launch the Gradio app
+# Launch the App
 if __name__ == "__main__":
     demo.launch()