import os
import multiprocessing
import concurrent.futures
from langchain.document_loaders import TextLoader, DirectoryLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.vectorstores import FAISS
from sentence_transformers import SentenceTransformer
import faiss
import torch
import numpy as np
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline, BitsAndBytesConfig
from datetime import datetime
import json
import gradio as gr
import re
# from unsloth import FastLanguageModel
import transformers
from transformers import BloomForCausalLM
from transformers import BloomForTokenClassification
from transformers import BloomForTokenClassification
from transformers import BloomTokenizerFast
import torch
class DocumentRetrievalAndGeneration:
def __init__(self, embedding_model_name, lm_model_id, data_folder):
# hf_token = os.getenv('HF_TOKEN')
hf="hf_VuNNBwnFqlcKzV"
token="vCfLXEBxyAOftxvlWpwf"
self.hf_token=hf+token
# print(HF_TOKEN,hf_token)
self.all_splits = self.load_documents(data_folder)
self.embeddings = SentenceTransformer(embedding_model_name)
self.cpu_index = self.create_faiss_index()
self.llm = self.initialize_llm2(lm_model_id)
def load_documents(self, folder_path):
loader = DirectoryLoader(folder_path, loader_cls=TextLoader)
documents = loader.load()
text_splitter = RecursiveCharacterTextSplitter(chunk_size=5000, chunk_overlap=250)
all_splits = text_splitter.split_documents(documents)
print('Length of documents:', len(documents))
print("LEN of all_splits", len(all_splits))
return all_splits
def create_faiss_index(self):
all_texts = [split.page_content for split in self.all_splits]
embeddings = self.embeddings.encode(all_texts, convert_to_tensor=True).cpu().numpy()
index = faiss.IndexFlatL2(embeddings.shape[1])
index.add(embeddings)
return index
def initialize_llm(self, model_id):
bnb_config = BitsAndBytesConfig(
load_in_4bit=True,
bnb_4bit_use_double_quant=True,
bnb_4bit_quant_type="nf4",
bnb_4bit_compute_dtype=torch.bfloat16
)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model = AutoModelForCausalLM.from_pretrained(model_id, quantization_config=bnb_config,token=self.hf_token)
tokenizer = AutoTokenizer.from_pretrained(model_id)
generate_text = pipeline(
model=model,
tokenizer=tokenizer,
return_full_text=True,
task='text-generation',
temperature=0.6,
max_new_tokens=256,
)
return generate_text
def initialize_llm2(self,model_id):
tokenizer = BloomTokenizerFast.from_pretrained("bigscience/bloom-1b3", local_files_only=True)
model = BloomForCausalLM.from_pretrained("bigscience/bloom-1b3", local_files_only=True)
result_length = 200
inputs = tokenizer(prompt, return_tensors="pt")
# return generate_text
def generate_response_with_timeout(self, model_inputs):
try:
with concurrent.futures.ThreadPoolExecutor() as executor:
future = executor.submit(self.llm.model.generate, model_inputs, max_new_tokens=1000, do_sample=True)
generated_ids = future.result(timeout=80) # Timeout set to 60 seconds
return generated_ids
except concurrent.futures.TimeoutError:
return "Text generation process timed out"
raise TimeoutError("Text generation process timed out")
def query_and_generate_response(self, query):
query_embedding = self.embeddings.encode(query, convert_to_tensor=True).cpu().numpy()
distances, indices = self.cpu_index.search(np.array([query_embedding]), k=5)
content = ""
for idx in indices[0]:
content += "-" * 50 + "\n"
content += self.all_splits[idx].page_content + "\n"
distance=distances[0][i]
print("CHUNK", idx)
print("Distance :",distance)
print(self.all_splits[idx].page_content)
print("############################")
prompt = f"""<s>
You are a knowledgeable assistant with access to a comprehensive database.
I need you to answer my question and provide related information in a specific format.
I have provided five relatable json files {content}, choose the most suitable chunks for answering the query
Here's what I need:
Include a final answer without additional comments, sign-offs, or extra phrases. Be direct and to the point.
content
Here's my question:
Query:{query}
Solution==>
RETURN ONLY SOLUTION . IF THEIR IS NO ANSWER RELATABLE IN RETRIEVED CHUNKS , RETURN " NO SOLUTION AVAILABLE"
IF THE QUERY AND THE RETRIEVED CHUNKS DO NOT CORRELATE MEANINGFULLY, OR IF THE QUERY IS NOT RELEVANT TO TDA2 OR RELATED TOPICS, THEN "NO SOLUTION AVAILABLE."
Example1
Query: "How to use IPU1_0 instead of A15_0 to process NDK in TDA2x-EVM",
Solution: "To use IPU1_0 instead of A15_0 to process NDK in TDA2x-EVM, you need to modify the configuration file of the NDK application. Specifically, change the processor reference from 'A15_0' to 'IPU1_0'.",
Example2
Query: "Can BQ25896 support I2C interface?",
Solution: "Yes, the BQ25896 charger supports the I2C interface for communication."
Example3
Query: "Who is the fastest runner in the world",
Solution:"NO SOLUTION AVAILABLE"
Example4
Query:"What is the price of latest apple MACBOOK "
Solution:"NO SOLUTION AVAILABLE"
</s>
"""
# messages = [{"role": "user", "content": prompt}]
# encodeds = self.llm.tokenizer.apply_chat_template(messages, return_tensors="pt")
# model_inputs = encodeds.to(self.llm.device)
# start_time = datetime.now()
# generated_ids = self.generate_response_with_timeout(model_inputs)
# elapsed_time = datetime.now() - start_time
# decoded = self.llm.tokenizer.batch_decode(generated_ids)
# generated_response = decoded[0]
generated_response=tokenizer.decode(model.generate(inputs["input_ids"], max_length=result_length,no_repeat_ngram_size=2)[0])
print(generated_response)
match1 = re.search(r'\[/INST\](.*?)</s>', generated_response, re.DOTALL)
match2 = re.search(r'Solution:(.*?)</s>', generated_response, re.DOTALL | re.IGNORECASE)
if match1:
solution_text = match1.group(1).strip()
if "Solution:" in solution_text:
solution_text = solution_text.split("Solution:", 1)[1].strip()
elif match2:
solution_text = match2.group(1).strip()
else:
solution_text=generated_response
print("Generated response:", generated_response)
print("Time elapsed:", elapsed_time)
print("Device in use:", self.llm.device)
return solution_text, content
def qa_infer_gradio(self, query):
response = self.query_and_generate_response(query)
return response
if __name__ == "__main__":
embedding_model_name = 'flax-sentence-embeddings/all_datasets_v3_MiniLM-L12'
# lm_model_id = "mistralai/Mistral-7B-Instruct-v0.2"
lm_model_id= "unsloth/Phi-3-mini-4k-instruct-bnb-4bit"
data_folder = 'text_files'
doc_retrieval_gen = DocumentRetrievalAndGeneration(embedding_model_name, lm_model_id, data_folder)
def launch_interface():
css_code = """
.gradio-container {
background-color: #daccdb;
}
/* Button styling for all buttons */
button {
background-color: #927fc7; /* Default color for all other buttons */
color: black;
border: 1px solid black;
padding: 10px;
margin-right: 10px;
font-size: 16px; /* Increase font size */
font-weight: bold; /* Make text bold */
}
"""
EXAMPLES = ["On which devices can the VIP and CSI2 modules operate simultaneously? ",
"I'm using Code Composer Studio 5.4.0.00091 and enabled FPv4SPD16 floating point support for CortexM4 in TDA2. However, after building the project, the .asm file shows --float_support=vfplib instead of FPv4SPD16. Why is this happening?",
"Could you clarify the maximum number of cameras that can be connected simultaneously to the video input ports on the TDA2x SoC, considering it supports up to 10 multiplexed input ports and includes 3 dedicated video input modules?"]
interface = gr.Interface(
fn=doc_retrieval_gen.qa_infer_gradio,
inputs=[gr.Textbox(label="QUERY", placeholder="Enter your query here")],
allow_flagging='never',
examples=EXAMPLES,
cache_examples=False,
outputs=[gr.Textbox(label="SOLUTION"), gr.Textbox(label="RELATED QUERIES")],
css=css_code
)
interface.launch(debug=True)
launch_interface()