Upload app.py
Browse files
app.py
ADDED
@@ -0,0 +1,257 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import streamlit as st
|
2 |
+
import uuid
|
3 |
+
import sys
|
4 |
+
import requests
|
5 |
+
from peft import *
|
6 |
+
import bitsandbytes as bnb
|
7 |
+
import pandas as pd
|
8 |
+
import torch
|
9 |
+
import torch.nn as nn
|
10 |
+
import transformers
|
11 |
+
from datasets import load_dataset
|
12 |
+
from huggingface_hub import notebook_login
|
13 |
+
from peft import (
|
14 |
+
LoraConfig,
|
15 |
+
PeftConfig,
|
16 |
+
get_peft_model,
|
17 |
+
prepare_model_for_kbit_training,
|
18 |
+
)
|
19 |
+
from transformers import (
|
20 |
+
AutoConfig,
|
21 |
+
AutoModelForCausalLM,
|
22 |
+
AutoTokenizer,
|
23 |
+
BitsAndBytesConfig,
|
24 |
+
)
|
25 |
+
|
26 |
+
|
27 |
+
USER_ICON = "images/user-icon.png"
|
28 |
+
AI_ICON = "images/ai-icon.png"
|
29 |
+
MAX_HISTORY_LENGTH = 5
|
30 |
+
|
31 |
+
if 'user_id' in st.session_state:
|
32 |
+
user_id = st.session_state['user_id']
|
33 |
+
else:
|
34 |
+
user_id = str(uuid.uuid4())
|
35 |
+
st.session_state['user_id'] = user_id
|
36 |
+
|
37 |
+
if 'chat_history' not in st.session_state:
|
38 |
+
st.session_state['chat_history'] = []
|
39 |
+
|
40 |
+
if "chats" not in st.session_state:
|
41 |
+
st.session_state.chats = [
|
42 |
+
{
|
43 |
+
'id': 0,
|
44 |
+
'question': '',
|
45 |
+
'answer': ''
|
46 |
+
}
|
47 |
+
]
|
48 |
+
|
49 |
+
if "questions" not in st.session_state:
|
50 |
+
st.session_state.questions = []
|
51 |
+
|
52 |
+
if "answers" not in st.session_state:
|
53 |
+
st.session_state.answers = []
|
54 |
+
|
55 |
+
if "input" not in st.session_state:
|
56 |
+
st.session_state.input = ""
|
57 |
+
|
58 |
+
st.markdown("""
|
59 |
+
<style>
|
60 |
+
.block-container {
|
61 |
+
padding-top: 32px;
|
62 |
+
padding-bottom: 32px;
|
63 |
+
padding-left: 0;
|
64 |
+
padding-right: 0;
|
65 |
+
}
|
66 |
+
.element-container img {
|
67 |
+
background-color: #000000;
|
68 |
+
}
|
69 |
+
|
70 |
+
.main-header {
|
71 |
+
font-size: 24px;
|
72 |
+
}
|
73 |
+
</style>
|
74 |
+
""", unsafe_allow_html=True)
|
75 |
+
|
76 |
+
def write_top_bar():
|
77 |
+
col1, col2, col3 = st.columns([1,10,2])
|
78 |
+
with col1:
|
79 |
+
st.image(AI_ICON, use_column_width='always')
|
80 |
+
with col2:
|
81 |
+
header = "Cogwise Intelligent Assistant"
|
82 |
+
st.write(f"<h3 class='main-header'>{header}</h3>", unsafe_allow_html=True)
|
83 |
+
with col3:
|
84 |
+
clear = st.button("Clear Chat")
|
85 |
+
return clear
|
86 |
+
|
87 |
+
clear = write_top_bar()
|
88 |
+
|
89 |
+
if clear:
|
90 |
+
st.session_state.questions = []
|
91 |
+
st.session_state.answers = []
|
92 |
+
st.session_state.input = ""
|
93 |
+
st.session_state["chat_history"] = []
|
94 |
+
|
95 |
+
def handle_input():
|
96 |
+
input = st.session_state.input
|
97 |
+
question_with_id = {
|
98 |
+
'question': input,
|
99 |
+
'id': len(st.session_state.questions)
|
100 |
+
}
|
101 |
+
st.session_state.questions.append(question_with_id)
|
102 |
+
|
103 |
+
chat_history = st.session_state["chat_history"]
|
104 |
+
if len(chat_history) == MAX_HISTORY_LENGTH:
|
105 |
+
chat_history = chat_history[:-1]
|
106 |
+
|
107 |
+
# api_url = "https://9pl792yjf9.execute-api.us-east-1.amazonaws.com/beta/chatcogwise"
|
108 |
+
# api_request_data = {"question": input, "session": user_id}
|
109 |
+
# api_response = requests.post(api_url, json=api_request_data)
|
110 |
+
# result = api_response.json()
|
111 |
+
|
112 |
+
# answer = result['answer']
|
113 |
+
# !pip install -Uqqq pip --progress-bar off
|
114 |
+
# !pip install -qqq bitsandbytes == 0.39.0
|
115 |
+
# !pip install -qqqtorch --2.0.1 --progress-bar off
|
116 |
+
# !pip install -qqq -U git + https://github.com/huggingface/transformers.git@e03a9cc --progress-bar off
|
117 |
+
# !pip install -qqq -U git + https://github.com/huggingface/peft.git@42a184f --progress-bar off
|
118 |
+
# !pip install -qqq -U git + https://github.com/huggingface/accelerate.git@c9fbb71 --progress-bar off
|
119 |
+
# !pip install -qqq datasets == 2.12.0 --progress-bar off
|
120 |
+
# !pip install -qqq loralib == 0.1.1 --progress-bar off
|
121 |
+
# !pip install einops
|
122 |
+
|
123 |
+
import os
|
124 |
+
# from pprint import pprint
|
125 |
+
# import json
|
126 |
+
|
127 |
+
|
128 |
+
|
129 |
+
os.environ["CUDA_VISIBLE_DEVICES"] = "0"
|
130 |
+
|
131 |
+
# notebook_login()
|
132 |
+
# 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
|
133 |
+
|
134 |
+
# """### Load dataset"""
|
135 |
+
|
136 |
+
from datasets import load_dataset
|
137 |
+
|
138 |
+
dataset_name = "nisaar/Lawyer_GPT_India"
|
139 |
+
# dataset_name = "patrick11434/TEST_LLM_DATASET"
|
140 |
+
dataset = load_dataset(dataset_name, split="train")
|
141 |
+
|
142 |
+
# """## Load adapters from the Hub
|
143 |
+
|
144 |
+
# You can also directly load adapters from the Hub using the commands below:
|
145 |
+
# """
|
146 |
+
|
147 |
+
|
148 |
+
# change peft_model_id
|
149 |
+
bnb_config = BitsAndBytesConfig(
|
150 |
+
load_in_4bit=True,
|
151 |
+
load_4bit_use_double_quant=True,
|
152 |
+
bnb_4bit_quant_type="nf4",
|
153 |
+
bnb_4bit_compute_dtype=torch.bfloat16,
|
154 |
+
)
|
155 |
+
|
156 |
+
peft_model_id = "nisaar/falcon7b-Indian_Law_150Prompts"
|
157 |
+
config = PeftConfig.from_pretrained(peft_model_id)
|
158 |
+
model = AutoModelForCausalLM.from_pretrained(
|
159 |
+
config.base_model_name_or_path,
|
160 |
+
return_dict=True,
|
161 |
+
quantization_config=bnb_config,
|
162 |
+
device_map="auto",
|
163 |
+
trust_remote_code=True,
|
164 |
+
)
|
165 |
+
tokenizer = AutoTokenizer.from_pretrained(config.base_model_name_or_path)
|
166 |
+
tokenizer.pad_token = tokenizer.eos_token
|
167 |
+
|
168 |
+
model = PeftModel.from_pretrained(model, peft_model_id)
|
169 |
+
|
170 |
+
"""## Inference
|
171 |
+
|
172 |
+
You can then directly use the trained model or the model that you have loaded from the 🤗 Hub for inference as you would do it usually in `transformers`.
|
173 |
+
"""
|
174 |
+
|
175 |
+
generation_config = model.generation_config
|
176 |
+
generation_config.max_new_tokens = 200
|
177 |
+
generation_config_temperature = 1
|
178 |
+
generation_config.top_p = 0.7
|
179 |
+
generation_config.num_return_sequences = 1
|
180 |
+
generation_config.pad_token_id = tokenizer.eos_token_id
|
181 |
+
generation_config_eod_token_id = tokenizer.eos_token_id
|
182 |
+
|
183 |
+
DEVICE = "cuda:0"
|
184 |
+
|
185 |
+
# Commented out IPython magic to ensure Python compatibility.
|
186 |
+
# %%time
|
187 |
+
# prompt = f"""
|
188 |
+
# <human>: Who appoints the Chief Justice of India?
|
189 |
+
# <assistant>:
|
190 |
+
# """.strip()
|
191 |
+
#
|
192 |
+
# encoding = tokenizer(prompt, return_tensors="pt").to(DEVICE)
|
193 |
+
# with torch.inference_mode():
|
194 |
+
# outputs = model.generate(
|
195 |
+
# input_ids=encoding.attention_mask,
|
196 |
+
# generation_config=generation_config,
|
197 |
+
# )
|
198 |
+
# print(tokenizer.decode(outputs[0],skip_special_tokens=True))
|
199 |
+
|
200 |
+
def generate_response(question: str) -> str:
|
201 |
+
prompt = f"""
|
202 |
+
<human>: {question}
|
203 |
+
<assistant>:
|
204 |
+
""".strip()
|
205 |
+
encoding = tokenizer(prompt, return_tensors="pt").to(DEVICE)
|
206 |
+
with torch.inference_mode():
|
207 |
+
outputs = model.generate(
|
208 |
+
input_ids=encoding.input_ids,
|
209 |
+
attention_mask=encoding.attention_mask,
|
210 |
+
generation_config=generation_config,
|
211 |
+
)
|
212 |
+
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
213 |
+
|
214 |
+
assistant_start = '<assistant>:'
|
215 |
+
response_start = response.find(assistant_start)
|
216 |
+
return response[response_start + len(assistant_start):].strip()
|
217 |
+
|
218 |
+
# prompt = "Debate the merits and demerits of introducing simultaneous elections in India?"
|
219 |
+
prompt=input
|
220 |
+
answer=print(generate_response(prompt))
|
221 |
+
|
222 |
+
# answer='Yes'
|
223 |
+
chat_history.append((input, answer))
|
224 |
+
|
225 |
+
st.session_state.answers.append({
|
226 |
+
'answer': answer,
|
227 |
+
'id': len(st.session_state.questions)
|
228 |
+
})
|
229 |
+
st.session_state.input = ""
|
230 |
+
|
231 |
+
def write_user_message(md):
|
232 |
+
col1, col2 = st.columns([1,12])
|
233 |
+
|
234 |
+
with col1:
|
235 |
+
st.image(USER_ICON, use_column_width='always')
|
236 |
+
with col2:
|
237 |
+
st.warning(md['question'])
|
238 |
+
|
239 |
+
def render_answer(answer):
|
240 |
+
col1, col2 = st.columns([1,12])
|
241 |
+
with col1:
|
242 |
+
st.image(AI_ICON, use_column_width='always')
|
243 |
+
with col2:
|
244 |
+
st.info(answer)
|
245 |
+
|
246 |
+
def write_chat_message(md, q):
|
247 |
+
chat = st.container()
|
248 |
+
with chat:
|
249 |
+
render_answer(md['answer'])
|
250 |
+
|
251 |
+
with st.container():
|
252 |
+
for (q, a) in zip(st.session_state.questions, st.session_state.answers):
|
253 |
+
write_user_message(q)
|
254 |
+
write_chat_message(a, q)
|
255 |
+
|
256 |
+
st.markdown('---')
|
257 |
+
input = st.text_input("You are talking to an AI, ask any question.", key="input", on_change=handle_input)
|