Spaces:
Running
Running
Commit
•
6996634
1
Parent(s):
b52e154
Load utils + javascript files
Browse files- my_functions.js +24 -0
- script.js +27 -0
- utils.py +244 -0
my_functions.js
ADDED
@@ -0,0 +1,24 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
var tag = document.createElement('script');
|
2 |
+
tag.src = 'https://www.youtube.com/iframe_api';
|
3 |
+
var firstScriptTag = document.getElementsByTagName('script')[0];
|
4 |
+
firstScriptTag.parentNode.insertBefore(tag, firstScriptTag);
|
5 |
+
var player, seconds = 0;
|
6 |
+
|
7 |
+
|
8 |
+
function onPlayerReady(event) {
|
9 |
+
event.target.playVideo();
|
10 |
+
}
|
11 |
+
function seek(sec){
|
12 |
+
var documentContainer = document;
|
13 |
+
var iframe = documentContainer.getElementById('player');
|
14 |
+
console.log("iframe");
|
15 |
+
console.log(iframe);
|
16 |
+
player = new YT.Player(iframe, {
|
17 |
+
events: {
|
18 |
+
'onReady': onPlayerReady
|
19 |
+
}
|
20 |
+
});
|
21 |
+
if(player){
|
22 |
+
player.seekTo(sec, true);
|
23 |
+
}
|
24 |
+
}
|
script.js
ADDED
@@ -0,0 +1,27 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
var tag = document.createElement('script');
|
2 |
+
tag.src = 'https://www.youtube.com/iframe_api';
|
3 |
+
var firstScriptTag = document.getElementsByTagName('script')[0];
|
4 |
+
firstScriptTag.parentNode.insertBefore(tag, firstScriptTag);
|
5 |
+
|
6 |
+
function onYouTubeIframeAPIReady() {
|
7 |
+
var iframe = document.getElementById('player');
|
8 |
+
player = new YT.Player(iframe, {
|
9 |
+
events: {
|
10 |
+
'onReady': onPlayerReady
|
11 |
+
}
|
12 |
+
});
|
13 |
+
}
|
14 |
+
function onPlayerReady(event) {
|
15 |
+
event.target.playVideo();
|
16 |
+
}
|
17 |
+
var iframe = document.getElementById('player');
|
18 |
+
player = new YT.Player(iframe, {
|
19 |
+
events: {
|
20 |
+
'onReady': onPlayerReady
|
21 |
+
}
|
22 |
+
});
|
23 |
+
function seek(sec){
|
24 |
+
if(player){
|
25 |
+
player.seekTo(sec, true);
|
26 |
+
}
|
27 |
+
}
|
utils.py
ADDED
@@ -0,0 +1,244 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# -- Utils .py file
|
2 |
+
# -- Libraries
|
3 |
+
from typing import Any, Dict, List, Mapping, Optional
|
4 |
+
from pydantic import Extra, Field, root_validator
|
5 |
+
from langchain.llms.base import LLM
|
6 |
+
from langchain.utils import get_from_dict_or_env
|
7 |
+
from langchain.vectorstores import Chroma
|
8 |
+
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
9 |
+
from langchain.chains import RetrievalQA
|
10 |
+
from langchain.document_loaders import TextLoader
|
11 |
+
from langchain.embeddings import HuggingFaceEmbeddings
|
12 |
+
from googletrans import Translator
|
13 |
+
import streamlit as st
|
14 |
+
import together
|
15 |
+
import textwrap
|
16 |
+
import spacy
|
17 |
+
import os
|
18 |
+
import re
|
19 |
+
|
20 |
+
os.environ["TOGETHER_API_KEY"] = "6101599d6e33e3bda336b8d007ca22e35a64c72cfd52c2d8197f663389fc50c5"
|
21 |
+
|
22 |
+
# -- LLM class
|
23 |
+
class TogetherLLM(LLM):
|
24 |
+
"""Together large language models."""
|
25 |
+
|
26 |
+
model: str = "togethercomputer/llama-2-70b-chat"
|
27 |
+
"""model endpoint to use"""
|
28 |
+
|
29 |
+
together_api_key: str = os.environ["TOGETHER_API_KEY"]
|
30 |
+
"""Together API key"""
|
31 |
+
|
32 |
+
temperature: float = 0.7
|
33 |
+
"""What sampling temperature to use."""
|
34 |
+
|
35 |
+
max_tokens: int = 512
|
36 |
+
"""The maximum number of tokens to generate in the completion."""
|
37 |
+
|
38 |
+
original_transcription: str = ""
|
39 |
+
"""Original transcription"""
|
40 |
+
|
41 |
+
class Config:
|
42 |
+
extra = Extra.forbid
|
43 |
+
|
44 |
+
#@root_validator(skip_on_failure=True)
|
45 |
+
def validate_environment(cls, values: Dict) -> Dict:
|
46 |
+
"""Validate that the API key is set."""
|
47 |
+
api_key = get_from_dict_or_env(
|
48 |
+
values, "together_api_key", "TOGETHER_API_KEY"
|
49 |
+
)
|
50 |
+
values["together_api_key"] = api_key
|
51 |
+
return values
|
52 |
+
|
53 |
+
@property
|
54 |
+
def _llm_type(self) -> str:
|
55 |
+
"""Return type of LLM."""
|
56 |
+
return "together"
|
57 |
+
|
58 |
+
def clean_duplicates(self, transcription: str) -> str:
|
59 |
+
transcription = transcription.strip().replace('/n/n ', """
|
60 |
+
""")
|
61 |
+
new_transcription_aux = []
|
62 |
+
for text in transcription.split('\n\n'):
|
63 |
+
if text not in new_transcription_aux:
|
64 |
+
new_transcription_aux.append(text)
|
65 |
+
return '\n\n'.join(new_transcription_aux)
|
66 |
+
|
67 |
+
def _call(
|
68 |
+
self,
|
69 |
+
prompt: str,
|
70 |
+
**kwargs: Any,
|
71 |
+
) -> str:
|
72 |
+
"""Call to Together endpoint."""
|
73 |
+
regex_transcription = r'CONTEXTO:(\n.*)+PREGUNTA'
|
74 |
+
regex_init_transcription = r"Desde el instante [0-9]+:[0-9]+:[0-9]+(?:\.[0-9]+)? hasta [0-9]+:[0-9]+:[0-9]+(?:\.[0-9]+)? [a-zA-Z ]+ dice: ?"
|
75 |
+
|
76 |
+
# -- Extract transcription
|
77 |
+
together.api_key = self.together_api_key
|
78 |
+
cleaned_prompt = self.clean_duplicates(prompt)
|
79 |
+
print(cleaned_prompt)
|
80 |
+
resultado = re.search(regex_transcription, cleaned_prompt, re.DOTALL)
|
81 |
+
|
82 |
+
resultado = re.sub(regex_init_transcription, "", resultado.group(1).strip()).replace('\"', '')
|
83 |
+
resultado_alpha_num = [re.sub(r'\W+', ' ', resultado_aux).strip().lower() for resultado_aux in resultado.split('\n\n')]
|
84 |
+
|
85 |
+
# -- Setup new transcription format, without duplicates and with its correspondent speaker
|
86 |
+
new_transcription = []
|
87 |
+
for transcription in self.original_transcription.split('\n\n'):
|
88 |
+
transcription_cleaned = re.sub(regex_init_transcription, "", transcription.strip()).replace('\"', '')
|
89 |
+
transcription_cleaned = re.sub(r'\W+', ' ', transcription_cleaned).strip().lower()
|
90 |
+
for resultado_aux in resultado_alpha_num:
|
91 |
+
if resultado_aux in transcription_cleaned or transcription_cleaned in resultado_aux:
|
92 |
+
init_transcription = re.findall(regex_init_transcription, transcription)[0]
|
93 |
+
new_transcription.append(init_transcription + '\"' + resultado_aux + '\"')
|
94 |
+
# -- Merge with original transcription
|
95 |
+
new_transcription = '\n\n'.join(list(set(new_transcription)))
|
96 |
+
new_cleaned_prompt = re.sub(regex_transcription, f"""CONTEXTO:
|
97 |
+
{new_transcription}
|
98 |
+
PREGUNTA:""", cleaned_prompt, re.DOTALL)
|
99 |
+
print(new_cleaned_prompt)
|
100 |
+
output = together.Complete.create(new_cleaned_prompt,
|
101 |
+
model=self.model,
|
102 |
+
max_tokens=self.max_tokens,
|
103 |
+
temperature=self.temperature,
|
104 |
+
)
|
105 |
+
text = output['output']['choices'][0]['text']
|
106 |
+
return text
|
107 |
+
|
108 |
+
# -- Python function to setup basic features: translator, SpaCy pipeline and LLM model
|
109 |
+
@st.cache_resource
|
110 |
+
def setup_app(transcription_path, emb_model, model, _logger):
|
111 |
+
# -- Setup enviroment and features
|
112 |
+
translator = Translator(service_urls=['translate.googleapis.com'])
|
113 |
+
nlp = spacy.load('es_core_news_lg')
|
114 |
+
|
115 |
+
_logger.info('Setup environment and features...')
|
116 |
+
|
117 |
+
# -- Setup LLM
|
118 |
+
together.api_key = os.environ["TOGETHER_API_KEY"]
|
119 |
+
# List available models and descriptons
|
120 |
+
models = together.Models.list()
|
121 |
+
# Set llama2 7b LLM
|
122 |
+
together.Models.start(model)
|
123 |
+
_logger.info('Setup environment and features - FINISHED!')
|
124 |
+
|
125 |
+
# -- Read translated transcription
|
126 |
+
_logger.info('Loading transcription...')
|
127 |
+
loader = TextLoader(transcription_path)
|
128 |
+
documents = loader.load()
|
129 |
+
# Splitting the text into chunks
|
130 |
+
text_splitter = RecursiveCharacterTextSplitter(chunk_size=1500, chunk_overlap=100)
|
131 |
+
texts = text_splitter.split_documents(documents)
|
132 |
+
_logger.info('Loading transcription - FINISHED!')
|
133 |
+
|
134 |
+
# -- Load embedding
|
135 |
+
_logger.info('Loading embedding...')
|
136 |
+
encode_kwargs = {'normalize_embeddings': True} # set True to compute cosine similarity
|
137 |
+
model_norm = HuggingFaceEmbeddings(
|
138 |
+
model_name=emb_model,
|
139 |
+
model_kwargs={'device': 'cpu'},
|
140 |
+
encode_kwargs=encode_kwargs
|
141 |
+
)
|
142 |
+
_logger.info('Loading embedding - FINISHED!')
|
143 |
+
|
144 |
+
# -- Create document database
|
145 |
+
_logger.info('Creating document database...')
|
146 |
+
# Embed and store the texts
|
147 |
+
# Supplying a persist_directory will store the embeddings on disk
|
148 |
+
persist_directory = 'db'
|
149 |
+
## Here is the nmew embeddings being used
|
150 |
+
embedding = model_norm
|
151 |
+
|
152 |
+
vectordb = Chroma.from_documents(documents=texts,
|
153 |
+
embedding=embedding,
|
154 |
+
persist_directory=persist_directory)
|
155 |
+
|
156 |
+
# -- Make a retreiver
|
157 |
+
retriever = vectordb.as_retriever(search_kwargs={"k": 5})
|
158 |
+
_logger.info('Creating document database - FINISHED!')
|
159 |
+
_logger.info('Setup finished!')
|
160 |
+
return translator, nlp, retriever
|
161 |
+
|
162 |
+
# -- Function to get prompt template
|
163 |
+
def get_prompt(instruction, system_prompt, b_sys, e_sys, b_inst, e_inst, _logger):
|
164 |
+
new_system_prompt = b_sys + system_prompt + e_sys
|
165 |
+
prompt_template = b_inst + new_system_prompt + instruction + e_inst
|
166 |
+
_logger.info('Prompt template created: {}'.format(instruction))
|
167 |
+
return prompt_template
|
168 |
+
|
169 |
+
# -- Function to create the chain to answer questions
|
170 |
+
@st.cache_resource
|
171 |
+
def create_llm_chain(model, _retriever, _chain_type_kwargs, _logger, transcription_path):
|
172 |
+
_logger.info('Creating LLM chain...')
|
173 |
+
# -- Keep original transcription
|
174 |
+
with open(transcription_path, 'r') as f:
|
175 |
+
formatted_transcription = f.read()
|
176 |
+
|
177 |
+
llm = TogetherLLM(
|
178 |
+
model= model,
|
179 |
+
temperature = 0.0,
|
180 |
+
max_tokens = 1024,
|
181 |
+
original_transcription = formatted_transcription
|
182 |
+
)
|
183 |
+
qa_chain = RetrievalQA.from_chain_type(llm=llm,
|
184 |
+
chain_type="stuff",
|
185 |
+
retriever=_retriever,
|
186 |
+
chain_type_kwargs=_chain_type_kwargs,
|
187 |
+
return_source_documents=True)
|
188 |
+
_logger.info('Creating LLM chain - FINISHED!')
|
189 |
+
return qa_chain
|
190 |
+
|
191 |
+
# -------------------------------------------
|
192 |
+
# -- Auxiliar functions
|
193 |
+
def wrap_text_preserve_newlines(text, width=110):
|
194 |
+
# Split the input text into lines based on newline characters
|
195 |
+
lines = text.split('\n')
|
196 |
+
|
197 |
+
# Wrap each line individually
|
198 |
+
wrapped_lines = [textwrap.fill(line, width=width) for line in lines]
|
199 |
+
|
200 |
+
# Join the wrapped lines back together using newline characters
|
201 |
+
wrapped_text = '\n'.join(wrapped_lines)
|
202 |
+
|
203 |
+
return wrapped_text
|
204 |
+
|
205 |
+
def process_llm_response(llm_response, nlp):
|
206 |
+
response = llm_response['result']
|
207 |
+
return wrap_text_preserve_newlines(response)
|
208 |
+
|
209 |
+
|
210 |
+
def time_to_seconds(time_str):
|
211 |
+
parts = time_str.split(':')
|
212 |
+
hours, minutes, seconds = map(float, parts)
|
213 |
+
return int((hours * 3600) + (minutes * 60) + seconds)
|
214 |
+
|
215 |
+
# -- Extract seconds from transcription
|
216 |
+
def add_hyperlink_and_convert_to_seconds(text):
|
217 |
+
time_pattern = r'(\d{2}:\d{2}:\d{2}(?:.\d{6})?)'
|
218 |
+
|
219 |
+
def get_seconds(match):
|
220 |
+
start_time_str, end_time_str = match[0], match[1]
|
221 |
+
start_time_seconds = time_to_seconds(start_time_str)
|
222 |
+
end_time_seconds = time_to_seconds(end_time_str)
|
223 |
+
return start_time_str, start_time_seconds, end_time_str, end_time_seconds
|
224 |
+
start_time_str, start_time_seconds, end_time_str, end_time_seconds = get_seconds(re.findall(time_pattern, text))
|
225 |
+
return start_time_str, start_time_seconds, end_time_str, end_time_seconds
|
226 |
+
|
227 |
+
# -- Streamlit HTML template
|
228 |
+
def typewrite(youtube_video_url, i=0):
|
229 |
+
youtube_video_url = youtube_video_url.replace("?enablejsapi=1", "")
|
230 |
+
margin = "{margin: 0;}"
|
231 |
+
html = f"""
|
232 |
+
<html>
|
233 |
+
<style>
|
234 |
+
p {margin}
|
235 |
+
</style>
|
236 |
+
<body>
|
237 |
+
<script src="https://www.youtube.com/player_api"></script>
|
238 |
+
<p align="center">
|
239 |
+
<iframe id="player_{i}" src="{youtube_video_url}" width="600" height="450"></iframe>
|
240 |
+
</p>
|
241 |
+
</body>
|
242 |
+
</html>
|
243 |
+
"""
|
244 |
+
return html
|