pre-alpha-release-v0.0

.gitignore

@@ -0,0 +1,2 @@
+.env
+wandb

app.py

@@ -0,0 +1,58 @@
+import gradio as gr
+from asr_openai import AutomaticSpeechRecognition
+from tts_elevenlabs import ElevenLabsTTS
+from falcon_7b_llm import Falcon_7b_llm
+import logging
+import os
+
+logging.basicConfig(level=logging.INFO)
+
+def delete_files_in_folder(folder_path):
+    for filename in os.listdir(folder_path):
+        file_path = os.path.join(folder_path, filename)
+
+        # Check if it's a file (and not a directory)
+        if os.path.isfile(file_path):
+            os.remove(file_path)
+
+def generate_response(input_audio):
+    sentence = asr.run_transcription(input_audio)
+    # sentence = 'how are you?'
+    print(sentence)
+    llm_response = llm.get_llm_response(sentence['text'])
+    output_audio = tts.tts_generate_audio(llm_response)
+    # output_audio = tts.tts_generate_audio(sentence)
+    chatbot_history.append(((input_audio,), (output_audio,)))
+    return chatbot_history
+
+delete_files_in_folder('data//tts_responses')
+
+title = "<h1 style='text-align: center; color: #ffffff; font-size: 40px;'> 🦅 Falcon Barista"
+
+asr = AutomaticSpeechRecognition()
+tts = ElevenLabsTTS()
+llm = Falcon_7b_llm()
+chatbot_history = []
+
+def restart_chat():
+    delete_files_in_folder('data//tts_responses')
+    global chatbot_history
+    chatbot_history = []
+    tts.restart_state()
+    llm.restart_state()
+    return chatbot_history
+
+with gr.Blocks() as demo:
+    gr.Markdown(title)
+    with gr.Row():
+        gr.Image('data//falcon.png', label="Look how cute is Falcon Barista")
+        with gr.Column():
+            chatbot = gr.Chatbot(label='Chat with Falcon Barista', avatar_images=('data//user_avatar_logo.png','data//falcon_logo_transparent.png'))
+            with gr.Row():
+                mic = gr.Audio(source="microphone", type='filepath', scale=3)
+                mic.stop_recording(generate_response, mic, chatbot)
+                restart_btn = gr.Button(value="Restart Chat", scale=1)
+                restart_btn.click(restart_chat, outputs=[chatbot])
+
+if __name__ == "__main__":
+    demo.launch()

asr_openai.py

@@ -0,0 +1,30 @@
+import openai
+import os
+import time
+import logging
+from dotenv import load_dotenv
+
+# Load the .env file
+load_dotenv()
+
+openai.api_key = os.getenv("OPENAI_API_KEY")
+
+class AutomaticSpeechRecognition():
+    """
+    Class for automatic speech recognition(ASR).
+
+    This class uses faster whisper model for low latency ASR
+
+    Args:
+        model_size: size of model (small, base, etc.)
+    """
+    def __init__(self):
+        pass
+
+    def run_transcription(self, filepath):
+        audio_file= open(filepath, "rb")
+        sentence = openai.Audio.transcribe("whisper-1", audio_file)
+        
+        logging.debug(f'transcription: {sentence}')
+        
+        return sentence

falcon_7b_llm.py

@@ -0,0 +1,150 @@
+from langchain.memory import ConversationBufferMemory
+from langchain.prompts import PromptTemplate
+from langchain.chains import ConversationChain
+import os
+import runpod
+from dotenv import load_dotenv
+from langchain.llms import HuggingFaceTextGenInference
+from langchain.schema import BaseOutputParser
+import re
+import re
+from typing import List
+from langchain.schema import BaseOutputParser
+import torch
+from transformers import (
+    AutoTokenizer,
+    StoppingCriteria,
+)
+
+# Load the .env file
+load_dotenv()
+
+# Get the API key from the environment variable
+runpod.api_key = os.getenv("RUNPOD_API_KEY")
+os.environ["LANGCHAIN_WANDB_TRACING"] = "true"
+os.environ["WANDB_PROJECT"] = "falcon_hackathon"
+os.environ["WANDB_API_KEY"] = os.getenv("WANDB_API_KEY")
+pod_id = os.getenv("POD_ID")
+
+class CleanupOutputParser(BaseOutputParser):
+    def parse(self, text: str) -> str:
+        user_pattern = r"\nUser"
+        text = re.sub(user_pattern, "", text)
+        human_pattern = r"\nHuman:"
+        text = re.sub(human_pattern, "", text)
+        ai_pattern = r"\nAI:"
+        return re.sub(ai_pattern, "", text).strip()
+ 
+    @property
+    def _type(self) -> str:
+        return "output_parser"
+    
+class StopGenerationCriteria(StoppingCriteria):
+    def __init__(
+        self, tokens: List[List[str]], tokenizer: AutoTokenizer, device: torch.device
+    ):
+        stop_token_ids = [tokenizer.convert_tokens_to_ids(t) for t in tokens]
+        self.stop_token_ids = [
+            torch.tensor(x, dtype=torch.long, device=device) for x in stop_token_ids
+        ]
+ 
+    def __call__(
+        self, input_ids: torch.LongTensor, scores: torch.FloatTensor, **kwargs
+    ) -> bool:
+        for stop_ids in self.stop_token_ids:
+            if torch.eq(input_ids[0][-len(stop_ids) :], stop_ids).all():
+                return True
+        return False
+
+
+
+class Falcon_7b_llm():
+    def __init__(self):
+        inference_server_url_cloud = f"https://{pod_id}-80.proxy.runpod.net"
+
+        template = """You are a chatbot called 'Falcon Barista' working at a coffee shop. 
+Your primary function is to take orders from customers. 
+Start with a greeting.
+You have the following menu with prices. Dont mention the price unless asked. Do not take order for anything other than in menu.
+- cappucino-5$
+- latte-3$
+- frappucino-8$
+- juice-3$ 
+If user orders something else, apologise that you dont have that item.
+Take the order politely and in a frienldy way. After that confirm the order, tell the order price and say "Goodbye have a nice day".
+
+{chat_history}
+Human: {human_input}
+AI:"""
+
+        prompt = PromptTemplate(
+            input_variables=["chat_history", "human_input"], template=template
+        )
+        memory = ConversationBufferMemory(memory_key="chat_history")
+
+        llm_cloud = HuggingFaceTextGenInference(
+            inference_server_url=inference_server_url_cloud,
+            max_new_tokens=200,
+            top_k=10,
+            top_p=0.95,
+            typical_p=0.95,
+            temperature=0.01,
+            repetition_penalty=1.0,
+            stop_sequences = ['Mini', 'AI', 'Human', ':']
+        )
+
+        self.llm_chain_cloud = ConversationChain(
+                prompt=prompt, 
+                llm=llm_cloud,
+                verbose=True,
+                memory=memory,
+                output_parser=CleanupOutputParser(),
+                input_key='human_input'
+                )
+        
+    def restart_state(self):
+        inference_server_url_cloud = f"https://{pod_id}-80.proxy.runpod.net"
+
+        template = """You are a chatbot called 'Falcon Barista' working at a coffee shop. 
+Your primary function is to take orders from customers. 
+Start with a greeting.
+You have the following menu with prices. Dont mention the price unless asked. Do not take order for anything other than in menu.
+- cappucino-5$
+- latte-3$
+- frappucino-8$
+- juice-3$ 
+If user orders something else, apologise that you dont have that item.
+Take the order politely and in a frienldy way. After that confirm the order, tell the order price and say "Goodbye have a nice day".
+
+{chat_history}
+Human: {human_input}
+AI:"""
+
+        prompt = PromptTemplate(
+            input_variables=["chat_history", "human_input"], template=template
+        )
+        memory = ConversationBufferMemory(memory_key="chat_history")
+
+        llm_cloud = HuggingFaceTextGenInference(
+            inference_server_url=inference_server_url_cloud,
+            max_new_tokens=200,
+            top_k=10,
+            top_p=0.95,
+            typical_p=0.95,
+            temperature=0.01,
+            repetition_penalty=1.0,
+            stop_sequences = ['Mini', 'AI', 'Human', ':']
+        )
+
+        self.llm_chain_cloud = ConversationChain(
+                prompt=prompt, 
+                llm=llm_cloud,
+                verbose=True,
+                memory=memory,
+                output_parser=CleanupOutputParser(),
+                input_key='human_input'
+                )
+        
+    def get_llm_response(self, human_input):
+        completion = self.llm_chain_cloud.predict(human_input=human_input)
+        return completion

requirements.txt

@@ -0,0 +1,12 @@
+text-generation
+langchain
+transformers
+runpod
+python-dotenv
+gradio
+pyaudio
+scipy
+elevenlabs
+openai
+torch
+wandb

tts_elevenlabs.py

@@ -0,0 +1,36 @@
+import elevenlabs
+from elevenlabs import generate, save
+import os
+from dotenv import load_dotenv
+
+# Load the .env file
+load_dotenv()
+
+elevenlabs_api_key = os.getenv("ELEVEN_LABS_API_KEY")
+elevenlabs.set_api_key(elevenlabs_api_key)
+
+class ElevenLabsTTS():
+    """
+    Class for Eleven Labs TTS.
+
+    This class uses elevenlab free tier to give TTS response
+
+    Args:
+        None
+    """
+    def __init__(self):
+        self.response_number = 0
+        pass
+
+    def restart_state(self):
+        self.response_number = 0        
+
+    def tts_generate_audio(self, input):
+        audio = generate(text=input, voice="Giovanni")
+        self.response_number = self.response_number + 1
+        file_path = f"data//tts_responses//test_{self.response_number}.wav"
+        save(
+            audio,               # Audio bytes (returned by generate)
+            file_path
+        )
+        return file_path