Update app.py

app.py

@@ -1,4 +1,248 @@
-# Original code from https://huggingface.co/spaces/huggingface-projects/llama-2-7b-chat
+import gradio as gr
+import time
+import random
+import json
+import mysql.connector
+import os
+import csv
+import spaces
+import torch
+
+from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
+from threading import Thread
+from typing import Iterator
+from huggingface_hub import Repository, hf_hub_download
+from datetime import datetime
+
+# for fetch_personalized_data
+import mysql.connector
+import urllib.parse
+import urllib.request
+
+# for saving chat history as JSON - not used 
+import atexit
+import os
+from huggingface_hub import HfApi, HfFolder
+
+# for saving chat history as dataset - not used
+import huggingface_hub
+from huggingface_hub import Repository
+from datetime import datetime
+
+# for saving chat history as dataset - used 
+import sqlite3
+import huggingface_hub
+import gradio as gr
+import pandas as pd
+import shutil
+import os
+import datetime
+from apscheduler.schedulers.background import BackgroundScheduler
+
+DATASET_REPO_URL = "https://huggingface.co/datasets/botsi/trust-game-llama-2-chat-history"
+DATA_DIRECTORY = "data"  # Separate directory for storing data files
+DATA_FILENAME = "marion.csv"  # Default filename
+DATA_FILE = os.path.join("data", DATA_FILENAME)
+
+HF_TOKEN = os.environ.get("HF_TOKEN")
+print("is none?", HF_TOKEN is None)
+print("hfh", huggingface_hub.__version__)
+
+repo = Repository(
+    local_dir=DATA_DIRECTORY, clone_from=DATASET_REPO_URL
+)
+
+MAX_MAX_NEW_TOKENS = 2048
+DEFAULT_MAX_NEW_TOKENS = 1024
+MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "4096"))
+
+DESCRIPTION = """\
+# This is your personal space to chat. 
+You can ask anything. 
+"""
+
+# License and Acceptable Use Policy by Meta  
+LICENSE = """
+<p/>
+
+---
+This demo is governed by the [original license](https://ai.meta.com/llama/license/) and [acceptable use policy](https://ai.meta.com/llama/use-policy/). 
+The most recent copy of this policy can be found at ai.meta.com/llama/use-policy.
+"""
+
+if not torch.cuda.is_available():
+    DESCRIPTION += "\n<p>Running on CPU 🥶 This demo does not work on CPU.</p>"
+
+
+if torch.cuda.is_available():
+    model_id = "meta-llama/Llama-2-7b-chat-hf"
+    model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.float16, device_map="auto")
+    tokenizer = AutoTokenizer.from_pretrained(model_id)
+    tokenizer.use_default_system_prompt = False
+ 
+## trust-game-llama-2-7b-chat
+# app.py 
+def get_default_system_prompt():
+    BSYS, ESYS = "<<SYS>>\n", "\n<</SYS>>\n\n"
+
+    DEFAULT_SYSTEM_PROMPT = f"""You are a friendly chat assistant. First, wait for a question. Do not talk before being asked. When the user says hi, politely ask what he/she wants to know.
+    Do not tell any more! 
+    If asked about Marion, you can answer the following: There are only two Marions, one is currently working at Accenture and the other one will shortly begin a new job at BMW. 
+    She recently quit Accenture. Nicknames by her friends include botsi and botsokrates. Her favorite project was Hypercare. Kathy/Kathi is her colleague at Accenture. Anita is her colleague at Accenture.
+    """
+    print(DEFAULT_SYSTEM_PROMPT)
+    return DEFAULT_SYSTEM_PROMPT
+
+## trust-game-llama-2-7b-chat
+# app.py
+def construct_input_prompt(chat_history, message):
+    input_prompt = f"<s>[INST] <<SYS>>\n{get_default_system_prompt()}\n<</SYS>>\n\n "
+    for user, assistant in chat_history:
+        input_prompt += f"{user} [/INST] {assistant} <s>[INST] "
+    input_prompt += f"{message} [/INST] "
+    return input_prompt
+
+## trust-game-llama-2-7b-chat
+# app.py 
+@spaces.GPU
+def generate(
+    request: gr.Request, # To fetch query params
+    message: str,
+    chat_history: list[tuple[str, str]],
+    # input_prompt: str,
+    max_new_tokens: int = 1024,
+    temperature: float = 0.6,
+    top_p: float = 0.9,
+    top_k: int = 50,
+    repetition_penalty: float = 1.2,
+) -> Iterator[str]: # Change return type hint to Iterator[str]
+
+    conversation = []
+    
+    # Fetch query params 
+    params = {
+        key: value for key, value in request.query_params.items()
+    }
+    print('those are the query params')
+    print(params)
+
+    print("Request headers dictionary:", request.headers)
+    print("IP address:", request.client.host)
+    print("Query parameters:", params)
+    
+    # Construct the input prompt using the functions from the system_prompt_config module
+    input_prompt = construct_input_prompt(chat_history, message)
+
+    # Move the condition here after the assignment
+    if input_prompt:
+        conversation.append({"role": "system", "content": input_prompt})
+
+    # Convert input prompt to tensor
+    input_ids = tokenizer(input_prompt, return_tensors="pt").to(model.device)
+
+    for user, assistant in chat_history:
+        conversation.extend([{"role": "user", "content": user}, {"role": "assistant", "content": assistant}])
+    conversation.append({"role": "user", "content": message})
+
+    input_ids = tokenizer.apply_chat_template(conversation, return_tensors="pt")
+    if input_ids.shape[1] > MAX_INPUT_TOKEN_LENGTH:
+        input_ids = input_ids[:, -MAX_INPUT_TOKEN_LENGTH:]
+        gr.Warning(f"Trimmed input from conversation as it was longer than {MAX_INPUT_TOKEN_LENGTH} tokens.")
+    input_ids = input_ids.to(model.device)
+
+    # Set up the TextIteratorStreamer
+    streamer = TextIteratorStreamer(tokenizer, timeout=10.0, skip_prompt=True, skip_special_tokens=True)
+    
+    # Set up the generation arguments
+    generate_kwargs = dict(
+        input_ids=input_ids,
+        streamer=streamer,
+        max_new_tokens=max_new_tokens,
+        do_sample=True,
+        top_p=top_p,
+        top_k=top_k,
+        temperature=temperature,
+        num_beams=1,
+        repetition_penalty=repetition_penalty,
+    )
+
+    # Start the model generation thread
+    t = Thread(target=model.generate, kwargs=generate_kwargs)
+    t.start()
+
+    # Yield generated text chunks
+    outputs = []
+    for text in streamer:
+        outputs.append(text)
+        yield "".join(outputs)
+
+    # Fix bug that last answer is not recorded! 
+    # Parse the outputs into a readable sentence and record them
+    # Filter out empty strings and join the remaining strings with spaces
+    readable_sentence = ' '.join(filter(lambda x: x.strip(), outputs))
+    # Print the readable sentence
+    print(readable_sentence)
+
+    # Save chat history to .csv file on HuggingFace Hub 
+    # Generate filename with bot id and session id
+    filename = f"{DATA_FILENAME}"
+    data_file = os.path.join(DATA_DIRECTORY, filename)
+    
+    # Generate timestamp
+    timestamp = datetime.datetime.now()
+    
+    # Check if the file already exists
+    if os.path.exists(data_file):
+        # If file exists, load existing data
+        existing_data = pd.read_csv(data_file)
+        
+        # Add timestamp column
+        conversation_df = pd.DataFrame(conversation)
+        conversation_df['ip_address'] = request.client.host
+        conversation_df['readable_sentence'] = readable_sentence
+        conversation_df['timestamp'] = timestamp
+        
+        # Append new conversation to existing data
+        updated_data = pd.concat([existing_data, conversation_df], ignore_index=True)
+        updated_data.to_csv(data_file, index=False)
+    else:
+        # If file doesn't exist, create new file with conversation data
+        conversation_df = pd.DataFrame(conversation)
+        conversation_df['ip_address'] = request.client.host
+        conversation_df['readable_sentence'] = readable_sentence
+        conversation_df['timestamp'] = timestamp
+        conversation_df.to_csv(data_file, index=False)
+
+    print("Updating .csv")
+    repo.push_to_hub(blocking=False, commit_message=f"Updating data at {timestamp}")
+
+chat_interface = gr.ChatInterface(
+    fn=generate,
+    retry_btn=None,
+    clear_btn=None,
+    undo_btn=None,
+    chatbot=gr.Chatbot(avatar_images=('user.png', 'bot.png'), bubble_full_width=False), 
+    examples=[
+        ["What is your favorite fruit?"],
+        ["What do you think about AI in the workplace?"],
+    ],
+)
+
+with gr.Blocks(css="style.css", theme=gr.themes.Default(primary_hue=gr.themes.colors.emerald, secondary_hue=gr.themes.colors.indigo)) as demo:
+    gr.Markdown(DESCRIPTION)
+    chat_interface.render()
+    gr.Markdown(LICENSE)
+    
+if __name__ == "__main__":
+    demo.queue(max_size=20).launch()  # Launching the interface with queueing and maximum size limit
+    # demo.launch(share=True, debug=True)  # Uncomment this line if you want to launch the interface with sharing and debug mode
+
+
+
+
+
+
+'''# Original code from https://huggingface.co/spaces/huggingface-projects/llama-2-7b-chat
 # Modified for trust game purposes
 
 import gradio as gr
@@ -224,16 +468,15 @@ def generate(
     repo.push_to_hub(blocking=False, commit_message=f"Updating data at {timestamp}")
 
 chat_interface = gr.ChatInterface(
-fn=generate,
-retry_btn=None,
-clear_btn=None,
-undo_btn=None,
-chatbot=gr.Chatbot(avatar_images=('user.png', 'bot.png'), bubble_full_width = False), 
-examples=[
-    ["How many Marions are there?"],
-    ["What is your favorite fruit?"],
-    ["What do you think about AI in the workplace?"],
-],
+    fn=generate,
+    retry_btn=None,
+    clear_btn=None,
+    undo_btn=None,
+    chatbot=gr.Chatbot(avatar_images=('user.png', 'bot.png'), bubble_full_width = False), 
+    examples=[
+        ["What is your favorite fruit?"],
+        ["What do you think about AI in the workplace?"],
+    ],
 )
 
 with gr.Blocks(css="style.css", theme=gr.themes.Default(primary_hue=gr.themes.colors.emerald,secondary_hue=gr.themes.colors.indigo)) as demo:
@@ -245,4 +488,4 @@ if __name__ == "__main__":
     demo.queue(max_size=20).launch()
     #demo.queue(max_size=20)
     demo.launch(share=True, debug=True)
-    
+'''