Upload app.py
Browse files
app.py
CHANGED
@@ -3,18 +3,46 @@
|
|
3 |
## Setup
|
4 |
# Import the necessary Libraries
|
5 |
|
|
|
|
|
|
|
|
|
6 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
7 |
|
8 |
|
9 |
|
10 |
|
11 |
# Create Client
|
|
|
12 |
|
|
|
|
|
|
|
|
|
|
|
|
|
13 |
|
14 |
# Define the embedding model and the vectorstore
|
|
|
15 |
|
16 |
# Load the persisted vectorDB
|
17 |
|
|
|
|
|
|
|
|
|
|
|
18 |
|
19 |
# Prepare the logging functionality
|
20 |
|
@@ -22,7 +50,7 @@ log_file = Path("logs/") / f"data_{uuid.uuid4()}.json"
|
|
22 |
log_folder = log_file.parent
|
23 |
|
24 |
scheduler = CommitScheduler(
|
25 |
-
repo_id="
|
26 |
repo_type="dataset",
|
27 |
folder_path=log_folder,
|
28 |
path_in_repo="data",
|
@@ -31,12 +59,30 @@ scheduler = CommitScheduler(
|
|
31 |
|
32 |
# Define the Q&A system message
|
33 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
34 |
|
|
|
|
|
|
|
|
|
35 |
|
36 |
|
37 |
|
38 |
# Define the user message template
|
|
|
|
|
|
|
|
|
39 |
|
|
|
|
|
|
|
40 |
|
41 |
|
42 |
|
@@ -48,12 +94,33 @@ def predict(user_input,company):
|
|
48 |
|
49 |
# Create context_for_query
|
50 |
|
|
|
|
|
|
|
51 |
|
52 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
53 |
|
54 |
|
55 |
-
# Get response from the LLM
|
56 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
57 |
|
58 |
# While the prediction is made, log both the inputs and outputs to a local log file
|
59 |
# While writing to the log file, ensure that the commit scheduler is locked to avoid parallel
|
@@ -82,6 +149,7 @@ company = gr.Radio()
|
|
82 |
# Create the interface
|
83 |
# For the inputs parameter of Interface provide [textbox,company]
|
84 |
|
|
|
85 |
|
86 |
demo.queue()
|
87 |
demo.launch()
|
|
|
3 |
## Setup
|
4 |
# Import the necessary Libraries
|
5 |
|
6 |
+
import json
|
7 |
+
import tiktoken
|
8 |
+
import gradio as gr
|
9 |
+
import uuid
|
10 |
|
11 |
+
import pandas as pd
|
12 |
+
|
13 |
+
from openai import OpenAI
|
14 |
+
|
15 |
+
from langchain_community.embeddings.sentence_transformer import (
|
16 |
+
SentenceTransformerEmbeddings
|
17 |
+
)
|
18 |
+
from langchain_community.vectorstores import Chroma
|
19 |
+
|
20 |
+
from google.colab import userdata, drive
|
21 |
+
from huggingface_hub import CommitScheduler
|
22 |
|
23 |
|
24 |
|
25 |
|
26 |
# Create Client
|
27 |
+
load_dotenv()
|
28 |
|
29 |
+
os.environ["ANYSCALE_API_KEY"]=os.getenv("ANYSCALE_API_KEY")
|
30 |
+
|
31 |
+
client = OpenAI(
|
32 |
+
base_url="https://api.endpoints.anyscale.com/v1",
|
33 |
+
api_key=os.environ['ANYSCALE_API_KEY']
|
34 |
+
)
|
35 |
|
36 |
# Define the embedding model and the vectorstore
|
37 |
+
embedding_model = SentenceTransformerEmbeddings(model_name='thenlper/gte-large')
|
38 |
|
39 |
# Load the persisted vectorDB
|
40 |
|
41 |
+
reportdb = Chroma(
|
42 |
+
collection_name=collection_name,
|
43 |
+
persist_directory='./report_db1',
|
44 |
+
embedding_function=embedding_model
|
45 |
+
)
|
46 |
|
47 |
# Prepare the logging functionality
|
48 |
|
|
|
50 |
log_folder = log_file.parent
|
51 |
|
52 |
scheduler = CommitScheduler(
|
53 |
+
repo_id="report-logs",
|
54 |
repo_type="dataset",
|
55 |
folder_path=log_folder,
|
56 |
path_in_repo="data",
|
|
|
59 |
|
60 |
# Define the Q&A system message
|
61 |
|
62 |
+
qna_system_message = """
|
63 |
+
You are an assistant to a financial services firm who answers user queries on annual reports.
|
64 |
+
User input will have the context required by you to answer user questions.
|
65 |
+
This context will begin with the token: ###Context.
|
66 |
+
The context contains references to specific portions of a document relevant to the user query.
|
67 |
+
|
68 |
+
User questions will begin with the token: ###Question.
|
69 |
|
70 |
+
Please answer only using the context provided in the input. Do not mention anything about the context in your final answer.
|
71 |
+
|
72 |
+
If the answer is not found in the context, respond "I don't know".
|
73 |
+
"""
|
74 |
|
75 |
|
76 |
|
77 |
# Define the user message template
|
78 |
+
qna_user_message_template = """
|
79 |
+
###Context
|
80 |
+
Here are some documents that are relevant to the question mentioned below.
|
81 |
+
{context}
|
82 |
|
83 |
+
###Question
|
84 |
+
{question}
|
85 |
+
"""
|
86 |
|
87 |
|
88 |
|
|
|
94 |
|
95 |
# Create context_for_query
|
96 |
|
97 |
+
relevant_document_chunks = retriever.get_relevant_documents(user_question)
|
98 |
+
context_list = [d.page_content for d in relevant_document_chunks]
|
99 |
+
context_for_query = ". ".join(context_list)
|
100 |
|
101 |
+
prompt = [
|
102 |
+
{'role':'system', 'content': qna_system_message},
|
103 |
+
{'role': 'user', 'content': qna_user_message_template.format(
|
104 |
+
context=context_for_query,
|
105 |
+
question=user_question
|
106 |
+
)
|
107 |
+
}
|
108 |
+
]
|
109 |
|
110 |
|
|
|
111 |
|
112 |
+
print(prompt)
|
113 |
+
|
114 |
+
# Create messages
|
115 |
+
response = client.chat.completions.create(
|
116 |
+
model=model_name,
|
117 |
+
messages=prompt,
|
118 |
+
temperature=0
|
119 |
+
)
|
120 |
+
|
121 |
+
# Get response from the LLM
|
122 |
+
answer = response.choices[0].message.content.strip()
|
123 |
+
print (answer)
|
124 |
|
125 |
# While the prediction is made, log both the inputs and outputs to a local log file
|
126 |
# While writing to the log file, ensure that the commit scheduler is locked to avoid parallel
|
|
|
149 |
# Create the interface
|
150 |
# For the inputs parameter of Interface provide [textbox,company]
|
151 |
|
152 |
+
demo = gr.Interface(inputs=[textbox,company], fn = predict, output ='text')
|
153 |
|
154 |
demo.queue()
|
155 |
demo.launch()
|