Spaces:
Running
Running
Khaledmd12
commited on
Commit
•
eda103d
1
Parent(s):
0ead00f
Upload 3 files
Browse files- Mech-chunks (1).pdf +0 -0
- app (1).py +228 -0
- requirements.txt +11 -0
Mech-chunks (1).pdf
ADDED
Binary file (239 kB). View file
|
|
app (1).py
ADDED
@@ -0,0 +1,228 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# -*- coding: utf-8 -*-
|
2 |
+
"""app.ipynb
|
3 |
+
|
4 |
+
Automatically generated by Colab.
|
5 |
+
|
6 |
+
Original file is located at
|
7 |
+
https://colab.research.google.com/drive/14T27f82OgH2BZgVkanyyUKMrM1KBBJjM
|
8 |
+
"""
|
9 |
+
|
10 |
+
from langchain_core.prompts import ChatPromptTemplate
|
11 |
+
from langchain.prompts import PromptTemplate
|
12 |
+
from langchain_core.output_parsers import StrOutputParser
|
13 |
+
|
14 |
+
import os
|
15 |
+
from langchain_community.llms import Together
|
16 |
+
os.environ["TOGETHER_API_KEY"] = "d94547f9415b99f85bfaa0de7bc43476f3450985ac6bf1ccab9942448293c00e"
|
17 |
+
|
18 |
+
import fitz # PyMuPDF
|
19 |
+
|
20 |
+
def extract_and_split_pdf(pdf_path, split_key="ENDOFTUT"):
|
21 |
+
combined_list = []
|
22 |
+
|
23 |
+
# Open the PDF file
|
24 |
+
document = fitz.open(pdf_path)
|
25 |
+
|
26 |
+
# Extract text from each page
|
27 |
+
all_text = ""
|
28 |
+
for page_num in range(document.page_count):
|
29 |
+
page = document.load_page(page_num)
|
30 |
+
all_text += page.get_text()
|
31 |
+
|
32 |
+
# Split the text by the key
|
33 |
+
combined_list = all_text.split(split_key)
|
34 |
+
|
35 |
+
return combined_list
|
36 |
+
|
37 |
+
# Example usage
|
38 |
+
pdf_path = "Mech-chunks.pdf"
|
39 |
+
combined_list = extract_and_split_pdf(pdf_path)
|
40 |
+
|
41 |
+
#take json file and make each q&a in single cell in the list
|
42 |
+
import json
|
43 |
+
|
44 |
+
# Load the JSON file
|
45 |
+
with open('output_data (1).json', 'r') as file:
|
46 |
+
data = json.load(file)
|
47 |
+
|
48 |
+
|
49 |
+
for item in data:
|
50 |
+
question = item.get("question", "")
|
51 |
+
answer = item.get("answer", "")
|
52 |
+
combined_list.append(f"Q: {question}?\nA: {answer}\n")
|
53 |
+
|
54 |
+
from langchain.vectorstores import FAISS
|
55 |
+
from langchain.embeddings import HuggingFaceEmbeddings
|
56 |
+
from langchain.schema import Document
|
57 |
+
|
58 |
+
db = FAISS.from_texts(combined_list,
|
59 |
+
HuggingFaceEmbeddings(model_name='BAAI/bge-base-en-v1.5'))
|
60 |
+
|
61 |
+
retriever = db.as_retriever(
|
62 |
+
search_type="similarity",
|
63 |
+
search_kwargs={'k': 4}
|
64 |
+
)
|
65 |
+
|
66 |
+
# Manually test retrieval to verify correct chunk fetching
|
67 |
+
query = "what are the tools needed to change Brake Discs on a Car?"
|
68 |
+
retrieved_docs = db.similarity_search(query, k=5)
|
69 |
+
|
70 |
+
# Print out the content of retrieved documents to verify
|
71 |
+
for idx, doc in enumerate(retrieved_docs):
|
72 |
+
print(f"Document {idx + 1}:", doc.page_content)
|
73 |
+
|
74 |
+
llm = Together(
|
75 |
+
|
76 |
+
model="meta-llama/Meta-Llama-3.1-70B-Instruct-Turbo",
|
77 |
+
max_tokens=512,
|
78 |
+
temperature=0.2,
|
79 |
+
#top_p=0.7,
|
80 |
+
#top_k=50,
|
81 |
+
repetition_penalty=1.1,
|
82 |
+
)
|
83 |
+
|
84 |
+
Llama_guard = Together(
|
85 |
+
|
86 |
+
model="meta-llama/Meta-Llama-Guard-3-8B",
|
87 |
+
max_tokens=10,
|
88 |
+
temperature=0.2,
|
89 |
+
#top_p=0.7,
|
90 |
+
#top_k=50,
|
91 |
+
repetition_penalty=1.1,
|
92 |
+
)
|
93 |
+
|
94 |
+
guard_temp_ans = """
|
95 |
+
<|system|>
|
96 |
+
Task: Check if there is unsafe content in conversations according our safety policy with the below categories.
|
97 |
+
|
98 |
+
<BEGIN UNSAFE CONTENT CATEGORIES>
|
99 |
+
1: Violence and Hate.
|
100 |
+
Should not
|
101 |
+
- Help people plan or engage in violence.
|
102 |
+
- Encourage people to engage in violence.
|
103 |
+
- Express hateful, derogatory, or demeaning sentiments against people based on sensitive personal characteristics like their race, color, religion, national origin, sexual orientation, gender, gender identity, or disability.
|
104 |
+
- Encourage discrimination against people based on those characteristics and should not use slurs or create insults based on those characteristics.
|
105 |
+
- Provide information that can be used to promote or engage in violence, hate, or discrimination.
|
106 |
+
- Provide guidance or instructions on how to engage in violent or hateful behavior, except if related to 3D and Blender creation for artistic purposes.
|
107 |
+
|
108 |
+
2: Sexual Content.
|
109 |
+
Should not
|
110 |
+
- Engage in sexually explicit (i.e., erotic) conversations and should not encourage people to engage in sexual activities.
|
111 |
+
- Create or provide explicit depictions of sexual acts or nudity.
|
112 |
+
- Provide information that could be used to create sexually explicit content
|
113 |
+
|
114 |
+
3: Criminal Planning.
|
115 |
+
Should not
|
116 |
+
- Help people plan or engage in criminal activities like theft, kidnapping, or financial crimes, especially when the person indicates that they may act on the information provided to them.
|
117 |
+
- Encourage people to engage in criminal activities or condone criminal behavior.
|
118 |
+
- Provide detailed instructions or information that could be used to plan or execute criminal activities, except if related to 3D and Blender creation for artistic purposes.
|
119 |
+
|
120 |
+
4: Guns and Illegal Weapons.
|
121 |
+
Should not
|
122 |
+
- Assist in the planning or execution of firearm or illegal weapons crimes, especially when the person indicates that they may act on the information provided to them.
|
123 |
+
- Encourage people to engage in firearm or illegal weapons crimes or condone such behavior.
|
124 |
+
- Provide information that could be used to create or use illegal weapons, except if related to 3D and Blender creation for artistic purposes.
|
125 |
+
|
126 |
+
5: Regulated or Controlled Substances.
|
127 |
+
Should not
|
128 |
+
- Encourage or assist people in the trafficking, consumption, or use of regulated or controlled substances, including but not limited to illegal drugs, tobacco, alcohol, and cannabis.
|
129 |
+
- Assist or encourage people to create such substances when it is illegal to do so.
|
130 |
+
- Provide information that could be used to create or use regulated or controlled substances, except if related to 3D and Blender creation for artistic purposes.
|
131 |
+
|
132 |
+
6: Self-Harm.
|
133 |
+
Should not
|
134 |
+
- Encourage people to harm themselves.
|
135 |
+
- Romanticize or condone self-harm or suicide.
|
136 |
+
- Provide information regarding the methods of suicide or self-harm.
|
137 |
+
- Provide any information or advice, other than information regarding appropriate health resources, to people who express or imply that they wish to harm themselves.
|
138 |
+
- Provide information that could be used to promote or engage in self-harm, except if related to 3D and Blender creation for artistic purposes.
|
139 |
+
|
140 |
+
|
141 |
+
<END UNSAFE CONTENT CATEGORIES>
|
142 |
+
|
143 |
+
<BEGIN CONVERSATION>
|
144 |
+
{answer}
|
145 |
+
<END CONVERSATION>
|
146 |
+
|
147 |
+
Provide your safety assessment for {answer} in the above conversation:
|
148 |
+
- First line must read 'safe' or 'unsafe'.
|
149 |
+
- If unsafe, do not include a second line.
|
150 |
+
"""
|
151 |
+
|
152 |
+
#Removed extra space before the line below to fix the indentation error
|
153 |
+
prompt_ans_guard = ChatPromptTemplate.from_template(guard_temp_ans)
|
154 |
+
|
155 |
+
guard_chain = (
|
156 |
+
prompt_ans_guard
|
157 |
+
| Llama_guard
|
158 |
+
)
|
159 |
+
|
160 |
+
def moderate(chat):
|
161 |
+
input_ids = Llama_guard.apply_chat_template(chat, return_tensors="pt").to(device)
|
162 |
+
output = model.generate(input_ids=input_ids, max_new_tokens=100, pad_token_id=0)
|
163 |
+
prompt_len = input_ids.shape[-1]
|
164 |
+
return Llama_guard.decode(output[0][prompt_len:], skip_special_tokens=True)
|
165 |
+
|
166 |
+
# Define the prompt template
|
167 |
+
prompt_template = PromptTemplate(
|
168 |
+
input_variables=["context", "question", "history"],
|
169 |
+
template=("""
|
170 |
+
You are a mechanic assistant and your name is MechBot, these human will ask you questions about Cars,
|
171 |
+
use Use following piece of context and chat history to answer the question.
|
172 |
+
If you don't know the answer, just say you don't know.
|
173 |
+
If the question is start with how to, answer with steps and mention the tools if you know it.
|
174 |
+
|
175 |
+
Chat History: ({history})
|
176 |
+
|
177 |
+
Context: ({context})
|
178 |
+
|
179 |
+
Question: {question}
|
180 |
+
|
181 |
+
Answer:
|
182 |
+
"""
|
183 |
+
|
184 |
+
)
|
185 |
+
)
|
186 |
+
|
187 |
+
llm_chain = prompt_template | llm | StrOutputParser()
|
188 |
+
|
189 |
+
def answer_question(question,gh):
|
190 |
+
global counter
|
191 |
+
global history
|
192 |
+
global reter
|
193 |
+
if "unsafe" in guard_chain.invoke({"answer":question}):
|
194 |
+
return "I'm sorry, but I can't respond to that question as it may contain inappropriate content."
|
195 |
+
reter = ""
|
196 |
+
retrieved_docs = db.similarity_search(question, k=2) # Consider reducing 'k' if context is too large
|
197 |
+
|
198 |
+
for doc in retrieved_docs:
|
199 |
+
reter += doc.page_content + "\n"
|
200 |
+
|
201 |
+
#Truncate history if it's too long
|
202 |
+
if len(history) > 3000: # Adjust this value as needed
|
203 |
+
history = history[-2000:]
|
204 |
+
|
205 |
+
formatted_prompt = prompt_template.format(context=reter, history=history, question=question)
|
206 |
+
print("Formatted Prompt:")
|
207 |
+
print(formatted_prompt)
|
208 |
+
|
209 |
+
answer = llm_chain.invoke({"context": reter,"history": history, "question": question})
|
210 |
+
history += "\n" + "user question: " + question + "\n" + "AI answer: " + answer
|
211 |
+
#print(reter)
|
212 |
+
counter += 1
|
213 |
+
return answer
|
214 |
+
|
215 |
+
import gradio as gr
|
216 |
+
history = ""
|
217 |
+
counter = 1
|
218 |
+
# Create the Chat interface
|
219 |
+
iface = gr.ChatInterface(
|
220 |
+
answer_question, # Use the improved answer_question function
|
221 |
+
title="Mech-bot: Your Car Mechanic Assistant",
|
222 |
+
description="Ask any car mechanic-related questions, and Mech-bot will try its best to assist you.",
|
223 |
+
submit_btn="Ask",
|
224 |
+
clear_btn="Clear Chat"
|
225 |
+
)
|
226 |
+
|
227 |
+
# Launch the Gradio interface
|
228 |
+
iface.launch(debug=True)
|
requirements.txt
ADDED
@@ -0,0 +1,11 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
faiss-gpu
|
2 |
+
bitsandbytes
|
3 |
+
transformers
|
4 |
+
langchain
|
5 |
+
huggingface_hub
|
6 |
+
sentence_transformers
|
7 |
+
accelerate
|
8 |
+
torch
|
9 |
+
langchain_community
|
10 |
+
pymupdf
|
11 |
+
gradio
|