Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,167 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from llama_index.llms import HuggingFaceInferenceAPI
|
2 |
+
from llama_index.llms import ChatMessage, MessageRole
|
3 |
+
from llama_index.prompts import ChatPromptTemplate
|
4 |
+
from llama_index import VectorStoreIndex, SimpleDirectoryReader, LLMPredictor, ServiceContext, StorageContext, load_index_from_storage
|
5 |
+
import gradio as gr
|
6 |
+
import sys
|
7 |
+
import logging
|
8 |
+
import torch
|
9 |
+
from huggingface_hub import InferenceClient
|
10 |
+
import tqdm as notebook_tqdm
|
11 |
+
import requests
|
12 |
+
|
13 |
+
def download_file(url, filename):
|
14 |
+
"""
|
15 |
+
Download a file from the specified URL and save it locally under the given filename.
|
16 |
+
"""
|
17 |
+
response = requests.get(url, stream=True)
|
18 |
+
|
19 |
+
# Check if the request was successful
|
20 |
+
if response.status_code == 200:
|
21 |
+
with open(filename, 'wb') as file:
|
22 |
+
for chunk in response.iter_content(chunk_size=1024):
|
23 |
+
if chunk: # filter out keep-alive new chunks
|
24 |
+
file.write(chunk)
|
25 |
+
print(f"Download complete: {filename}")
|
26 |
+
else:
|
27 |
+
print(f"Error: Unable to download file. HTTP status code: {response.status_code}")
|
28 |
+
|
29 |
+
def generate(prompt, history, file_link, temperature=0.9, max_new_tokens=256, top_p=0.95, repetition_penalty=1.0,):
|
30 |
+
mixtral = HuggingFaceInferenceAPI(
|
31 |
+
model_name="mistralai/Mixtral-8x7B-Instruct-v0.1"
|
32 |
+
)
|
33 |
+
|
34 |
+
service_context = ServiceContext.from_defaults(
|
35 |
+
llm=mixtral, embed_model="local:BAAI/bge-small-en-v1.5"
|
36 |
+
)
|
37 |
+
|
38 |
+
|
39 |
+
download = download_file(file_link,file_link.split("/")[-1])
|
40 |
+
|
41 |
+
documents = SimpleDirectoryReader("/content").load_data()
|
42 |
+
index = VectorStoreIndex.from_documents(documents,service_context=service_context)
|
43 |
+
|
44 |
+
# Text QA Prompt
|
45 |
+
chat_text_qa_msgs = [
|
46 |
+
ChatMessage(
|
47 |
+
role=MessageRole.SYSTEM,
|
48 |
+
content=(
|
49 |
+
"Always answer the question, even if the context isn't helpful."
|
50 |
+
),
|
51 |
+
),
|
52 |
+
ChatMessage(
|
53 |
+
role=MessageRole.USER,
|
54 |
+
content=(
|
55 |
+
"Context information is below.\n"
|
56 |
+
"---------------------\n"
|
57 |
+
"{context_str}\n"
|
58 |
+
"---------------------\n"
|
59 |
+
"Given the context information and not prior knowledge, "
|
60 |
+
"answer the question: {query_str}\n"
|
61 |
+
),
|
62 |
+
),
|
63 |
+
]
|
64 |
+
text_qa_template = ChatPromptTemplate(chat_text_qa_msgs)
|
65 |
+
|
66 |
+
# Refine Prompt
|
67 |
+
chat_refine_msgs = [
|
68 |
+
ChatMessage(
|
69 |
+
role=MessageRole.SYSTEM,
|
70 |
+
content=(
|
71 |
+
"Always answer the question, even if the context isn't helpful."
|
72 |
+
),
|
73 |
+
),
|
74 |
+
ChatMessage(
|
75 |
+
role=MessageRole.USER,
|
76 |
+
content=(
|
77 |
+
"We have the opportunity to refine the original answer "
|
78 |
+
"(only if needed) with some more context below.\n"
|
79 |
+
"------------\n"
|
80 |
+
"{context_msg}\n"
|
81 |
+
"------------\n"
|
82 |
+
"Given the new context, refine the original answer to better "
|
83 |
+
"answer the question: {query_str}. "
|
84 |
+
"If the context isn't useful, output the original answer again.\n"
|
85 |
+
"Original Answer: {existing_answer}"
|
86 |
+
),
|
87 |
+
),
|
88 |
+
]
|
89 |
+
refine_template = ChatPromptTemplate(chat_refine_msgs)
|
90 |
+
|
91 |
+
stream= index.as_query_engine(
|
92 |
+
text_qa_template=text_qa_template, refine_template=refine_template, similarity_top_k=6
|
93 |
+
).query(prompt)
|
94 |
+
print(str(stream))
|
95 |
+
|
96 |
+
output=""
|
97 |
+
|
98 |
+
for response in str(stream):
|
99 |
+
output += response
|
100 |
+
yield output
|
101 |
+
return output
|
102 |
+
|
103 |
+
def upload_file(files):
|
104 |
+
file_paths = [file.name for file in files]
|
105 |
+
return file_paths
|
106 |
+
|
107 |
+
additional_inputs=[
|
108 |
+
gr.Textbox(
|
109 |
+
label="File Link",
|
110 |
+
max_lines=1,
|
111 |
+
interactive=True,
|
112 |
+
value="https://arxiv.org/pdf/2401.10020.pdf"
|
113 |
+
),
|
114 |
+
gr.Slider(
|
115 |
+
label="Temperature",
|
116 |
+
value=0.9,
|
117 |
+
minimum=0.0,
|
118 |
+
maximum=1.0,
|
119 |
+
step=0.05,
|
120 |
+
interactive=True,
|
121 |
+
info="Higher values produce more diverse outputs",
|
122 |
+
),
|
123 |
+
gr.Slider(
|
124 |
+
label="Max new tokens",
|
125 |
+
value=1024,
|
126 |
+
minimum=0,
|
127 |
+
maximum=2048,
|
128 |
+
step=64,
|
129 |
+
interactive=True,
|
130 |
+
info="The maximum numbers of new tokens",
|
131 |
+
),
|
132 |
+
gr.Slider(
|
133 |
+
label="Top-p (nucleus sampling)",
|
134 |
+
value=0.90,
|
135 |
+
minimum=0.0,
|
136 |
+
maximum=1,
|
137 |
+
step=0.05,
|
138 |
+
interactive=True,
|
139 |
+
info="Higher values sample more low-probability tokens",
|
140 |
+
),
|
141 |
+
gr.Slider(
|
142 |
+
label="Repetition penalty",
|
143 |
+
value=1.2,
|
144 |
+
minimum=1.0,
|
145 |
+
maximum=2.0,
|
146 |
+
step=0.05,
|
147 |
+
interactive=True,
|
148 |
+
info="Penalize repeated tokens",
|
149 |
+
)
|
150 |
+
]
|
151 |
+
|
152 |
+
examples=[["Explain the paper and describe its novelty", None, None, None, None, None, ],
|
153 |
+
["Can you write a short story about a time-traveling detective who solves historical mysteries?", None, None, None, None, None,],
|
154 |
+
["I'm trying to learn French. Can you provide some common phrases that would be useful for a beginner, along with their pronunciations?", None, None, None, None, None,],
|
155 |
+
["I have chicken, rice, and bell peppers in my kitchen. Can you suggest an easy recipe I can make with these ingredients?", None, None, None, None, None,],
|
156 |
+
["Can you explain how the QuickSort algorithm works and provide a Python implementation?", None, None, None, None, None,],
|
157 |
+
["What are some unique features of Rust that make it stand out compared to other systems programming languages like C++?", None, None, None, None, None,],
|
158 |
+
]
|
159 |
+
|
160 |
+
gr.ChatInterface(
|
161 |
+
fn=generate,
|
162 |
+
chatbot=gr.Chatbot(show_label=False, show_share_button=False, show_copy_button=True, likeable=True, layout="panel"),
|
163 |
+
additional_inputs=additional_inputs,
|
164 |
+
title="RAG Demo",
|
165 |
+
examples=examples,
|
166 |
+
concurrency_limit=20,
|
167 |
+
).launch(show_api=False,debug=True)
|