File size: 6,243 Bytes
fc74eab 6fc434e fc74eab 6fc434e fc74eab 6fc434e fc74eab 6fc434e fc74eab 763ebc4 fc74eab 763ebc4 fc74eab 763ebc4 fc74eab 6fc434e fc74eab 763ebc4 fc74eab |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 |
from datetime import datetime
from llama_index.core import VectorStoreIndex, SimpleDirectoryReader
from llama_index.embeddings.huggingface import HuggingFaceEmbedding
from llama_parse import LlamaParse
from llama_index.llms.huggingface_api import HuggingFaceInferenceAPI
import os
from dotenv import load_dotenv
import gradio as gr
import markdowm as md
import base64
# Load environment variables
load_dotenv()
llm_models = [
"mistralai/Mixtral-8x7B-Instruct-v0.1",
"meta-llama/Meta-Llama-3-8B-Instruct",
"mistralai/Mistral-7B-Instruct-v0.2",
"tiiuae/falcon-7b-instruct",
]
embed_models = [
"BAAI/bge-small-en-v1.5", # 33.4M
"NeuML/pubmedbert-base-embeddings",
"BAAI/llm-embedder", # 109M
"BAAI/bge-large-en" # 335M
]
# Global variable for selected model
selected_llm_model_name = llm_models[0] # Default to the first model in the list
selected_embed_model_name = embed_models[0] # Default to the first model in the list
vector_index = None
# Initialize the parser
parser = LlamaParse(api_key=os.getenv("LLAMA_INDEX_API"), result_type='markdown')
# Define file extractor with various common extensions
file_extractor = {
'.pdf': parser, # PDF documents
'.docx': parser, # Microsoft Word documents
'.doc': parser, # Older Microsoft Word documents
'.txt': parser, # Plain text files
'.csv': parser, # Comma-separated values files
'.xlsx': parser, # Microsoft Excel files (requires additional processing for tables)
'.pptx': parser, # Microsoft PowerPoint files (for slides)
'.html': parser, # HTML files (web pages)
# Image files for OCR processing
'.jpg': parser, # JPEG images
'.jpeg': parser, # JPEG images
'.png': parser, # PNG images
# Scanned documents in image formats
'.webp': parser, # WebP images
'.svg': parser, # SVG files (vector format, may contain embedded text)
}
# File processing function
def load_files(file_path: str, embed_model_name: str):
try:
global vector_index
document = SimpleDirectoryReader(input_files=[file_path], file_extractor=file_extractor).load_data()
embed_model = HuggingFaceEmbedding(model_name=embed_model_name)
vector_index = VectorStoreIndex.from_documents(document, embed_model=embed_model)
print(f"Parsing done for {file_path}")
filename = os.path.basename(file_path)
return f"Ready to give response on {filename}"
except Exception as e:
return f"An error occurred: {e}"
# Function to handle the selected model from dropdown
def set_llm_model(selected_model):
global selected_llm_model_name
selected_llm_model_name = selected_model # Update the global variable
# print(f"Model selected: {selected_model_name}")
# return f"Model set to: {selected_model_name}"
# Respond function that uses the globally set selected model
def respond(message, history):
try:
# Initialize the LLM with the selected model
llm = HuggingFaceInferenceAPI(
model_name=selected_llm_model_name,
contextWindow=8192, # Context window size (typically max length of the model)
maxTokens=1024, # Tokens per response generation (512-1024 works well for detailed answers)
temperature=0.3, # Lower temperature for more focused answers (0.2-0.4 for factual info)
topP=0.9, # Top-p sampling to control diversity while retaining quality
frequencyPenalty=0.5, # Slight penalty to avoid repetition
presencePenalty=0.5, # Encourages exploration without digressing too much
token=os.getenv("TOKEN")
)
# Set up the query engine with the selected LLM
query_engine = vector_index.as_query_engine(llm=llm)
bot_message = query_engine.query(message)
print(f"\n{datetime.now()}:{selected_llm_model_name}:: {message} --> {str(bot_message)}\n")
return f"{selected_llm_model_name}:\n{str(bot_message)}"
except Exception as e:
if str(e) == "'NoneType' object has no attribute 'as_query_engine'":
return "Please upload a file."
return f"An error occurred: {e}"
def encode_image(image_path):
with open(image_path, "rb") as image_file:
return base64.b64encode(image_file.read()).decode('utf-8')
# UI Setup
with gr.Blocks(theme=gr.themes.Soft(font=[gr.themes.GoogleFont("Roboto Mono")]), css='footer {visibility: hidden}') as demo:
gr.Markdown("# HundAI QA📄")
with gr.Tabs():
with gr.TabItem("Introduction"):
gr.Markdown(md.description)
with gr.TabItem("Chatbot"):
with gr.Accordion("IMPORTANT: READ ME FIRST", open=False):
guid = gr.Markdown(md.guide)
with gr.Row():
with gr.Column(scale=1):
file_input = gr.File(file_count="single", type='filepath', label="Upload document")
# gr.Markdown("Dont know what to select check out in Intro tab")
embed_model_dropdown = gr.Dropdown(embed_models, label="Select Embedding", interactive=True)
with gr.Row():
btn = gr.Button("Submit", variant='primary')
clear = gr.ClearButton()
output = gr.Text(label='Vector Index')
llm_model_dropdown = gr.Dropdown(llm_models, label="Select LLM", interactive=True)
with gr.Column(scale=3):
gr.ChatInterface(
fn=respond,
chatbot=gr.Chatbot(height=500),
theme = "soft",
show_progress='full',
# cache_mode='lazy',
textbox=gr.Textbox(placeholder="Ask me any questions on the uploaded document!", container=False)
)
# Set up Gradio interactions
llm_model_dropdown.change(fn=set_llm_model, inputs=llm_model_dropdown)
btn.click(fn=load_files, inputs=[file_input, embed_model_dropdown], outputs=output)
clear.click(lambda: [None] * 3, outputs=[file_input, embed_model_dropdown, output])
# Launch the demo with a public link option
if __name__ == "__main__":
demo.launch() |