Upload app.py
Browse files
app.py
CHANGED
@@ -2,7 +2,6 @@
|
|
2 |
|
3 |
## Setup
|
4 |
# Import the necessary Libraries
|
5 |
-
|
6 |
import json
|
7 |
import gradio as gr
|
8 |
import uuid
|
@@ -20,8 +19,6 @@ from google.colab import userdata, drive
|
|
20 |
from huggingface_hub import CommitScheduler
|
21 |
|
22 |
|
23 |
-
|
24 |
-
|
25 |
# Create Client
|
26 |
load_dotenv()
|
27 |
|
@@ -34,9 +31,7 @@ client = OpenAI(
|
|
34 |
|
35 |
# Define the embedding model and the vectorstore
|
36 |
embedding_model = SentenceTransformerEmbeddings(model_name='thenlper/gte-large')
|
37 |
-
|
38 |
# Load the persisted vectorDB
|
39 |
-
|
40 |
reportdb = Chroma(
|
41 |
collection_name=collection_name,
|
42 |
persist_directory='./report_db1',
|
@@ -49,7 +44,7 @@ log_file = Path("logs/") / f"data_{uuid.uuid4()}.json"
|
|
49 |
log_folder = log_file.parent
|
50 |
|
51 |
scheduler = CommitScheduler(
|
52 |
-
repo_id="
|
53 |
repo_type="dataset",
|
54 |
folder_path=log_folder,
|
55 |
path_in_repo="data",
|
@@ -74,6 +69,7 @@ If the answer is not found in the context, respond "I don't know".
|
|
74 |
|
75 |
|
76 |
# Define the user message template
|
|
|
77 |
qna_user_message_template = """
|
78 |
###Context
|
79 |
Here are some documents that are relevant to the question mentioned below.
|
@@ -84,7 +80,6 @@ Here are some documents that are relevant to the question mentioned below.
|
|
84 |
"""
|
85 |
|
86 |
|
87 |
-
|
88 |
# Define the predict function that runs when 'Submit' is clicked or when a API request is made
|
89 |
def predict(user_input,company):
|
90 |
|
@@ -92,51 +87,47 @@ def predict(user_input,company):
|
|
92 |
relevant_document_chunks = vectorstore_persisted.similarity_search(user_input, k=5, filter={"source":filter})
|
93 |
|
94 |
# Create context_for_query
|
95 |
-
|
96 |
-
|
97 |
-
|
98 |
-
|
99 |
-
|
100 |
-
|
101 |
-
|
102 |
-
|
103 |
-
|
104 |
-
|
105 |
-
|
106 |
-
|
107 |
-
|
108 |
-
|
109 |
-
|
110 |
-
|
111 |
-
print(prompt)
|
112 |
-
|
113 |
# Create messages
|
114 |
-
response = client.chat.completions.create(
|
115 |
-
|
116 |
-
|
117 |
-
|
118 |
-
)
|
|
|
119 |
|
120 |
# Get response from the LLM
|
121 |
-
answer = response.choices[0].message.content.strip()
|
122 |
-
print (answer)
|
123 |
|
124 |
# While the prediction is made, log both the inputs and outputs to a local log file
|
125 |
# While writing to the log file, ensure that the commit scheduler is locked to avoid parallel
|
126 |
# access
|
127 |
|
128 |
-
with scheduler.lock:
|
129 |
-
|
130 |
-
|
131 |
-
|
132 |
-
|
133 |
-
|
134 |
-
|
135 |
-
|
136 |
-
|
137 |
-
|
138 |
|
139 |
-
|
140 |
|
141 |
# Set-up the Gradio UI
|
142 |
# Add text box and radio button to the interface
|
@@ -148,7 +139,6 @@ company = gr.Radio()
|
|
148 |
# Create the interface
|
149 |
# For the inputs parameter of Interface provide [textbox,company]
|
150 |
|
151 |
-
demo = gr.Interface(inputs=[textbox,company], fn = predict, output ='text')
|
152 |
|
153 |
demo.queue()
|
154 |
demo.launch()
|
|
|
2 |
|
3 |
## Setup
|
4 |
# Import the necessary Libraries
|
|
|
5 |
import json
|
6 |
import gradio as gr
|
7 |
import uuid
|
|
|
19 |
from huggingface_hub import CommitScheduler
|
20 |
|
21 |
|
|
|
|
|
22 |
# Create Client
|
23 |
load_dotenv()
|
24 |
|
|
|
31 |
|
32 |
# Define the embedding model and the vectorstore
|
33 |
embedding_model = SentenceTransformerEmbeddings(model_name='thenlper/gte-large')
|
|
|
34 |
# Load the persisted vectorDB
|
|
|
35 |
reportdb = Chroma(
|
36 |
collection_name=collection_name,
|
37 |
persist_directory='./report_db1',
|
|
|
44 |
log_folder = log_file.parent
|
45 |
|
46 |
scheduler = CommitScheduler(
|
47 |
+
repo_id="---------",
|
48 |
repo_type="dataset",
|
49 |
folder_path=log_folder,
|
50 |
path_in_repo="data",
|
|
|
69 |
|
70 |
|
71 |
# Define the user message template
|
72 |
+
|
73 |
qna_user_message_template = """
|
74 |
###Context
|
75 |
Here are some documents that are relevant to the question mentioned below.
|
|
|
80 |
"""
|
81 |
|
82 |
|
|
|
83 |
# Define the predict function that runs when 'Submit' is clicked or when a API request is made
|
84 |
def predict(user_input,company):
|
85 |
|
|
|
87 |
relevant_document_chunks = vectorstore_persisted.similarity_search(user_input, k=5, filter={"source":filter})
|
88 |
|
89 |
# Create context_for_query
|
90 |
+
relevant_document_chunks = retriever.get_relevant_documents(user_question)
|
91 |
+
context_list = [d.page_content for d in relevant_document_chunks]
|
92 |
+
context_for_query = ". ".join(context_list)
|
93 |
+
|
94 |
+
prompt = [
|
95 |
+
{'role':'system', 'content': qna_system_message},
|
96 |
+
{'role': 'user', 'content': qna_user_message_template.format(
|
97 |
+
context=context_for_query,
|
98 |
+
question=user_question
|
99 |
+
)
|
100 |
+
}
|
101 |
+
]
|
102 |
+
print(prompt)
|
|
|
|
|
|
|
|
|
|
|
103 |
# Create messages
|
104 |
+
response = client.chat.completions.create(
|
105 |
+
model=model_name,
|
106 |
+
messages=prompt,
|
107 |
+
temperature=0
|
108 |
+
)
|
109 |
+
|
110 |
|
111 |
# Get response from the LLM
|
112 |
+
answer = response.choices[0].message.content.strip()
|
113 |
+
print (answer)
|
114 |
|
115 |
# While the prediction is made, log both the inputs and outputs to a local log file
|
116 |
# While writing to the log file, ensure that the commit scheduler is locked to avoid parallel
|
117 |
# access
|
118 |
|
119 |
+
with scheduler.lock:
|
120 |
+
with log_file.open("a") as f:
|
121 |
+
f.write(json.dumps(
|
122 |
+
{
|
123 |
+
'user_input': user_input,
|
124 |
+
'retrieved_context': context_for_query,
|
125 |
+
'model_response': prediction
|
126 |
+
}
|
127 |
+
))
|
128 |
+
f.write("\n")
|
129 |
|
130 |
+
return prediction
|
131 |
|
132 |
# Set-up the Gradio UI
|
133 |
# Add text box and radio button to the interface
|
|
|
139 |
# Create the interface
|
140 |
# For the inputs parameter of Interface provide [textbox,company]
|
141 |
|
|
|
142 |
|
143 |
demo.queue()
|
144 |
demo.launch()
|