Spaces:
Runtime error
Runtime error
KonradSzafer
commited on
Commit
•
bfdf8df
1
Parent(s):
c6dce39
channel id added to config
Browse files- data/hugging_face_videos_dataset.py +151 -0
- data/indexer.ipynb +35 -25
- data/requirements-audio.txt +5 -0
- discord_bot/__main__.py +1 -0
- discord_bot/client/client.py +32 -34
- qa_engine/config.py +1 -0
data/hugging_face_videos_dataset.py
ADDED
@@ -0,0 +1,151 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import re
|
3 |
+
import time
|
4 |
+
|
5 |
+
import torch
|
6 |
+
import scrapetube
|
7 |
+
from pytube import YouTube
|
8 |
+
from faster_whisper import WhisperModel
|
9 |
+
from tqdm import tqdm
|
10 |
+
|
11 |
+
|
12 |
+
# Available models:
|
13 |
+
# tiny.en, tiny, base.en, base, small.en, small, medium.en, medium
|
14 |
+
# large-v1, large-v2, large-v3, large
|
15 |
+
MODEL_NAME = "large-v3"
|
16 |
+
AUDIO_SAVE_PATH = 'datasets/huggingface_audio/'
|
17 |
+
TRANSCRIPTS_SAVE_PATH = 'datasets/huggingface_audio_transcribed/'
|
18 |
+
|
19 |
+
if torch.cuda.is_available():
|
20 |
+
# requires: conda install -c anaconda cudnn
|
21 |
+
print(f"Using {MODEL_NAME} on GPU and float16")
|
22 |
+
model = WhisperModel(MODEL_NAME, device="cuda", compute_type="float16", device_index=[5])
|
23 |
+
else:
|
24 |
+
print(f"Using {MODEL_NAME} on CPU and int8")
|
25 |
+
model = WhisperModel(MODEL_NAME, device="cpu", compute_type="int8")
|
26 |
+
|
27 |
+
|
28 |
+
def replace_unallowed_chars(filename: str) -> str:
|
29 |
+
unallowed_chars = [' ', '/', '\\', ':', '*', '?', '"', '<', '>', '|']
|
30 |
+
for char in unallowed_chars:
|
31 |
+
filename = filename.replace(char, '_')
|
32 |
+
return filename
|
33 |
+
|
34 |
+
|
35 |
+
def get_videos_urls(channel_url: str) -> list[str]:
|
36 |
+
videos = scrapetube.get_channel(channel_url=channel_url)
|
37 |
+
return [
|
38 |
+
f"https://www.youtube.com/watch?v={video['videoId']}"
|
39 |
+
for video in videos
|
40 |
+
]
|
41 |
+
|
42 |
+
|
43 |
+
def get_audio_from_video(video_url: str, save_path: str) -> tuple[str, int, str, int]:
|
44 |
+
yt = YouTube(video_url)
|
45 |
+
if check_if_file_exists(yt.title, save_path):
|
46 |
+
print(f'Audio already exists for: {yt.title}')
|
47 |
+
return (video_url, yt.title.replace(" ", "_")+".mp3", yt.title, yt.length)
|
48 |
+
else:
|
49 |
+
print(f'Downloading audio for: {yt.title}')
|
50 |
+
video = yt.streams.filter(only_audio=True).first()
|
51 |
+
out_file = video.download(output_path=save_path)
|
52 |
+
base, ext = os.path.splitext(out_file)
|
53 |
+
new_filename = save_path + replace_unallowed_chars(yt.title) + '.mp3'
|
54 |
+
print(f'Saving audio to: {new_filename}')
|
55 |
+
os.rename(out_file, new_filename)
|
56 |
+
print(f'Video length: {yt.length} seconds')
|
57 |
+
return (video_url, new_filename, yt.title, yt.length)
|
58 |
+
|
59 |
+
|
60 |
+
def check_if_file_exists(filename: str, save_path: str) -> bool:
|
61 |
+
title = filename.replace(' ', '_')
|
62 |
+
return any([
|
63 |
+
title in filename_
|
64 |
+
for filename_ in os.listdir(save_path)
|
65 |
+
])
|
66 |
+
|
67 |
+
|
68 |
+
def transcript_from_audio(audio_path: str) -> dict[str, list[str]]:
|
69 |
+
segments, info = model.transcribe(audio_path, beam_size=10)
|
70 |
+
return list(segments)
|
71 |
+
|
72 |
+
|
73 |
+
def process_text(text: str) -> str:
|
74 |
+
text = text.strip()
|
75 |
+
text = re.sub('\s+', ' ', text)
|
76 |
+
return text
|
77 |
+
|
78 |
+
|
79 |
+
def merge_transcripts_segements(
|
80 |
+
segments: list[str],
|
81 |
+
file_title: str,
|
82 |
+
num_segments_to_merge: int = 5,
|
83 |
+
) -> dict[str, list[str]]:
|
84 |
+
|
85 |
+
merged_segments = {}
|
86 |
+
temp_text = ''
|
87 |
+
start_time = None
|
88 |
+
end_time = None
|
89 |
+
|
90 |
+
for i, segment in enumerate(segments):
|
91 |
+
if i % num_segments_to_merge == 0:
|
92 |
+
start_time = segment.start
|
93 |
+
end_time = segment.end
|
94 |
+
temp_text += segment.text + ' '
|
95 |
+
|
96 |
+
if (i + 1) % num_segments_to_merge == 0 or i == len(segments) - 1:
|
97 |
+
key = f'{start_time:.2f}_{end_time:.2f}'
|
98 |
+
merged_segments[key] = process_text(temp_text)
|
99 |
+
temp_text = ''
|
100 |
+
|
101 |
+
return merged_segments
|
102 |
+
|
103 |
+
|
104 |
+
def main():
|
105 |
+
if not os.path.exists(AUDIO_SAVE_PATH):
|
106 |
+
os.makedirs(AUDIO_SAVE_PATH)
|
107 |
+
if not os.path.exists(TRANSCRIPTS_SAVE_PATH):
|
108 |
+
os.makedirs(TRANSCRIPTS_SAVE_PATH)
|
109 |
+
|
110 |
+
print('Getting videos urls')
|
111 |
+
videos_urls = get_videos_urls('https://www.youtube.com/@HuggingFace')
|
112 |
+
|
113 |
+
print('Downloading audio files')
|
114 |
+
audio_data = []
|
115 |
+
for video_url in tqdm(videos_urls):
|
116 |
+
try:
|
117 |
+
audio_data.append(
|
118 |
+
get_audio_from_video(video_url, save_path=AUDIO_SAVE_PATH)
|
119 |
+
)
|
120 |
+
except Exception as e:
|
121 |
+
print(f'Error downloading video: {video_url}')
|
122 |
+
print(e)
|
123 |
+
|
124 |
+
print('Transcribing audio files')
|
125 |
+
for video_url, filename, title, audio_length in tqdm(audio_data):
|
126 |
+
if check_if_file_exists(title, TRANSCRIPTS_SAVE_PATH):
|
127 |
+
print(f'Transcript already exists for: {title}')
|
128 |
+
continue
|
129 |
+
try:
|
130 |
+
print(f'Transcribing: {title}')
|
131 |
+
start_time = time.time()
|
132 |
+
segments = transcript_from_audio(filename)
|
133 |
+
print(f'Transcription took: {time.time() - start_time:.1f} seconds')
|
134 |
+
merged_segments = merge_transcripts_segements(
|
135 |
+
segments,
|
136 |
+
title,
|
137 |
+
num_segments_to_merge=10
|
138 |
+
)
|
139 |
+
# save transcripts to separate files
|
140 |
+
title = replace_unallowed_chars(title)
|
141 |
+
for segment, text in merged_segments.items():
|
142 |
+
with open(f'{TRANSCRIPTS_SAVE_PATH}{title}_{segment}.txt', 'w') as f:
|
143 |
+
video_url_with_time = f'{video_url}&t={float(segment.split("_")[0]):.0f}'
|
144 |
+
f.write(f'source: {video_url_with_time}\n\n' + text)
|
145 |
+
except Exception as e:
|
146 |
+
print(f'Error transcribing: {title}')
|
147 |
+
print(e)
|
148 |
+
|
149 |
+
|
150 |
+
if __name__ == '__main__':
|
151 |
+
main()
|
data/indexer.ipynb
CHANGED
@@ -7,16 +7,18 @@
|
|
7 |
"outputs": [],
|
8 |
"source": [
|
9 |
"import math\n",
|
10 |
-
"import numpy as np\n",
|
11 |
"from pathlib import Path\n",
|
|
|
|
|
|
|
12 |
"from tqdm import tqdm\n",
|
13 |
-
"from typing import List, Any\n",
|
14 |
"from langchain.chains import RetrievalQA\n",
|
15 |
"from langchain.embeddings import HuggingFaceEmbeddings, HuggingFaceInstructEmbeddings\n",
|
16 |
"from langchain.document_loaders import TextLoader\n",
|
17 |
"from langchain.indexes import VectorstoreIndexCreator\n",
|
18 |
"from langchain.text_splitter import CharacterTextSplitter\n",
|
19 |
-
"from langchain.vectorstores import FAISS"
|
|
|
20 |
]
|
21 |
},
|
22 |
{
|
@@ -25,16 +27,32 @@
|
|
25 |
"metadata": {},
|
26 |
"outputs": [],
|
27 |
"source": [
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
28 |
"docs = []\n",
|
29 |
"metadata = []\n",
|
30 |
-
"for
|
31 |
-
"
|
32 |
-
" with open(p) as f:\n",
|
33 |
-
" # the first line is the source of the text\n",
|
34 |
-
" source = f.readline().strip().replace('source: ', '')\n",
|
35 |
-
" docs.append(f.read())\n",
|
36 |
-
" metadata.append({\"source\": source})\n",
|
37 |
-
" # break\n",
|
38 |
"\n",
|
39 |
"print(f'number of documents: {len(docs)}')"
|
40 |
]
|
@@ -88,7 +106,7 @@
|
|
88 |
" if self.max_length < 0:\n",
|
89 |
" print('max_length is not specified, using model default max_seq_length')\n",
|
90 |
"\n",
|
91 |
-
" def embed_documents(self, texts:
|
92 |
" all_embeddings = []\n",
|
93 |
" for text in tqdm(texts, desc=\"Embedding documents\"):\n",
|
94 |
" if len(text) > self.max_length and self.max_length > -1:\n",
|
@@ -109,7 +127,8 @@
|
|
109 |
" return all_embeddings\n",
|
110 |
"\n",
|
111 |
"\n",
|
112 |
-
"# max length fed to the
|
|
|
113 |
"max_length = 512\n",
|
114 |
"embedding_model = AverageInstructEmbeddings( \n",
|
115 |
" model_name=model_name,\n",
|
@@ -143,8 +162,8 @@
|
|
143 |
"metadata": {},
|
144 |
"outputs": [],
|
145 |
"source": [
|
146 |
-
"index_name = f'index-{model_name}-{chunk_size}-m{max_length}-
|
147 |
-
"index_name"
|
148 |
]
|
149 |
},
|
150 |
{
|
@@ -189,8 +208,6 @@
|
|
189 |
"metadata": {},
|
190 |
"outputs": [],
|
191 |
"source": [
|
192 |
-
"from huggingface_hub import HfApi\n",
|
193 |
-
"\n",
|
194 |
"api = HfApi()\n",
|
195 |
"api.create_repo(\n",
|
196 |
" repo_id=f'KonradSzafer/{index_name}',\n",
|
@@ -204,13 +221,6 @@
|
|
204 |
" repo_type='dataset',\n",
|
205 |
")"
|
206 |
]
|
207 |
-
},
|
208 |
-
{
|
209 |
-
"cell_type": "code",
|
210 |
-
"execution_count": null,
|
211 |
-
"metadata": {},
|
212 |
-
"outputs": [],
|
213 |
-
"source": []
|
214 |
}
|
215 |
],
|
216 |
"metadata": {
|
@@ -229,7 +239,7 @@
|
|
229 |
"name": "python",
|
230 |
"nbconvert_exporter": "python",
|
231 |
"pygments_lexer": "ipython3",
|
232 |
-
"version": "3.
|
233 |
},
|
234 |
"orig_nbformat": 4
|
235 |
},
|
|
|
7 |
"outputs": [],
|
8 |
"source": [
|
9 |
"import math\n",
|
|
|
10 |
"from pathlib import Path\n",
|
11 |
+
"from typing import Any\n",
|
12 |
+
"\n",
|
13 |
+
"import numpy as np\n",
|
14 |
"from tqdm import tqdm\n",
|
|
|
15 |
"from langchain.chains import RetrievalQA\n",
|
16 |
"from langchain.embeddings import HuggingFaceEmbeddings, HuggingFaceInstructEmbeddings\n",
|
17 |
"from langchain.document_loaders import TextLoader\n",
|
18 |
"from langchain.indexes import VectorstoreIndexCreator\n",
|
19 |
"from langchain.text_splitter import CharacterTextSplitter\n",
|
20 |
+
"from langchain.vectorstores import FAISS\n",
|
21 |
+
"from huggingface_hub import HfApi"
|
22 |
]
|
23 |
},
|
24 |
{
|
|
|
27 |
"metadata": {},
|
28 |
"outputs": [],
|
29 |
"source": [
|
30 |
+
"def collect_docs(directory: str, docs: list[str], metadata: list[Any]):\n",
|
31 |
+
" for p in Path(directory).iterdir():\n",
|
32 |
+
" if not p.is_dir():\n",
|
33 |
+
" with open(p) as f:\n",
|
34 |
+
" # the first line is the source of the text\n",
|
35 |
+
" source = f.readline().strip().replace('source: ', '')\n",
|
36 |
+
" docs.append(f.read())\n",
|
37 |
+
" metadata.append({\"source\": source})\n",
|
38 |
+
" # break"
|
39 |
+
]
|
40 |
+
},
|
41 |
+
{
|
42 |
+
"cell_type": "code",
|
43 |
+
"execution_count": null,
|
44 |
+
"metadata": {},
|
45 |
+
"outputs": [],
|
46 |
+
"source": [
|
47 |
+
"DIRECTORIES = [\n",
|
48 |
+
" \"./datasets/huggingface_docs/\",\n",
|
49 |
+
" \"./datasets/huggingface_audio_transcribed/\"\n",
|
50 |
+
"]\n",
|
51 |
+
"\n",
|
52 |
"docs = []\n",
|
53 |
"metadata = []\n",
|
54 |
+
"for directory in DIRECTORIES:\n",
|
55 |
+
" collect_docs(directory, docs, metadata)\n",
|
|
|
|
|
|
|
|
|
|
|
|
|
56 |
"\n",
|
57 |
"print(f'number of documents: {len(docs)}')"
|
58 |
]
|
|
|
106 |
" if self.max_length < 0:\n",
|
107 |
" print('max_length is not specified, using model default max_seq_length')\n",
|
108 |
"\n",
|
109 |
+
" def embed_documents(self, texts: list[str]) -> list[list[float]]:\n",
|
110 |
" all_embeddings = []\n",
|
111 |
" for text in tqdm(texts, desc=\"Embedding documents\"):\n",
|
112 |
" if len(text) > self.max_length and self.max_length > -1:\n",
|
|
|
127 |
" return all_embeddings\n",
|
128 |
"\n",
|
129 |
"\n",
|
130 |
+
"# max length fed to the mode\n",
|
131 |
+
"# if longer than CHUNK_SIZE in previous steps: then N chunks + averaging of embeddings\n",
|
132 |
"max_length = 512\n",
|
133 |
"embedding_model = AverageInstructEmbeddings( \n",
|
134 |
" model_name=model_name,\n",
|
|
|
162 |
"metadata": {},
|
163 |
"outputs": [],
|
164 |
"source": [
|
165 |
+
"index_name = f'index-{model_name}-{chunk_size}-m{max_length}-11_Jan_2024'\n",
|
166 |
+
"index_name = index_name.replace('/', '_')"
|
167 |
]
|
168 |
},
|
169 |
{
|
|
|
208 |
"metadata": {},
|
209 |
"outputs": [],
|
210 |
"source": [
|
|
|
|
|
211 |
"api = HfApi()\n",
|
212 |
"api.create_repo(\n",
|
213 |
" repo_id=f'KonradSzafer/{index_name}',\n",
|
|
|
221 |
" repo_type='dataset',\n",
|
222 |
")"
|
223 |
]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
224 |
}
|
225 |
],
|
226 |
"metadata": {
|
|
|
239 |
"name": "python",
|
240 |
"nbconvert_exporter": "python",
|
241 |
"pygments_lexer": "ipython3",
|
242 |
+
"version": "3.11.5"
|
243 |
},
|
244 |
"orig_nbformat": 4
|
245 |
},
|
data/requirements-audio.txt
ADDED
@@ -0,0 +1,5 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
scrapetube>=2.5.1
|
2 |
+
pytube>=15.0.0
|
3 |
+
faster-whisper>=0.10.0
|
4 |
+
torch>=2.0.1
|
5 |
+
tqdm>=4.66.1
|
discord_bot/__main__.py
CHANGED
@@ -16,6 +16,7 @@ qa_engine = QAEngine(
|
|
16 |
)
|
17 |
client = DiscordClient(
|
18 |
qa_engine=qa_engine,
|
|
|
19 |
num_last_messages=config.num_last_messages,
|
20 |
use_names_in_context=config.use_names_in_context,
|
21 |
enable_commands=config.enable_commands,
|
|
|
16 |
)
|
17 |
client = DiscordClient(
|
18 |
qa_engine=qa_engine,
|
19 |
+
channel_ids=config.discotd_channel_ids,
|
20 |
num_last_messages=config.num_last_messages,
|
21 |
use_names_in_context=config.use_names_in_context,
|
22 |
enable_commands=config.enable_commands,
|
discord_bot/client/client.py
CHANGED
@@ -31,6 +31,7 @@ class DiscordClient(discord.Client):
|
|
31 |
def __init__(
|
32 |
self,
|
33 |
qa_engine: QAEngine,
|
|
|
34 |
num_last_messages: int = 5,
|
35 |
use_names_in_context: bool = True,
|
36 |
enable_commands: bool = True,
|
@@ -45,6 +46,7 @@ class DiscordClient(discord.Client):
|
|
45 |
'The number of last messages in context should be at least 1'
|
46 |
|
47 |
self.qa_engine: QAEngine = qa_engine
|
|
|
48 |
self.num_last_messages: int = num_last_messages
|
49 |
self.use_names_in_context: bool = use_names_in_context
|
50 |
self.enable_commands: bool = enable_commands
|
@@ -98,38 +100,34 @@ class DiscordClient(discord.Client):
|
|
98 |
|
99 |
|
100 |
async def on_message(self, message):
|
101 |
-
|
102 |
-
|
103 |
-
|
104 |
-
|
105 |
-
|
106 |
-
|
107 |
-
|
108 |
-
|
109 |
-
|
110 |
-
|
111 |
-
|
112 |
-
|
113 |
-
|
114 |
-
|
115 |
-
|
116 |
-
|
117 |
-
|
118 |
-
|
119 |
-
|
120 |
-
|
121 |
-
|
122 |
-
|
123 |
-
|
124 |
-
|
125 |
-
|
|
|
|
|
|
|
126 |
)
|
127 |
-
|
128 |
-
|
129 |
-
await self.send_message(
|
130 |
-
message,
|
131 |
-
response.get_answer(),
|
132 |
-
response.get_sources_as_text()
|
133 |
-
)
|
134 |
-
except Exception as e:
|
135 |
-
logger.error('Failed to send response: {0}'.format(e))
|
|
|
31 |
def __init__(
|
32 |
self,
|
33 |
qa_engine: QAEngine,
|
34 |
+
channel_ids: list[int] = [],
|
35 |
num_last_messages: int = 5,
|
36 |
use_names_in_context: bool = True,
|
37 |
enable_commands: bool = True,
|
|
|
46 |
'The number of last messages in context should be at least 1'
|
47 |
|
48 |
self.qa_engine: QAEngine = qa_engine
|
49 |
+
self.channel_ids: list[int] = channel_ids
|
50 |
self.num_last_messages: int = num_last_messages
|
51 |
self.use_names_in_context: bool = use_names_in_context
|
52 |
self.enable_commands: bool = enable_commands
|
|
|
100 |
|
101 |
|
102 |
async def on_message(self, message):
|
103 |
+
|
104 |
+
if self.channel_ids and message.channel.id not in self.channel_ids:
|
105 |
+
return
|
106 |
+
|
107 |
+
if message.author == self.user:
|
108 |
+
return
|
109 |
+
|
110 |
+
"""
|
111 |
+
if self.enable_commands and message.content.startswith('!'):
|
112 |
+
if message.content == '!clear':
|
113 |
+
await message.channel.purge()
|
114 |
+
return
|
115 |
+
"""
|
116 |
+
|
117 |
+
last_messages = await self.get_last_messages(message)
|
118 |
+
context = '\n'.join(last_messages)
|
119 |
+
|
120 |
+
logger.info('Received message: {0.content}'.format(message))
|
121 |
+
response = self.qa_engine.get_response(
|
122 |
+
question=message.content,
|
123 |
+
messages_context=context
|
124 |
+
)
|
125 |
+
logger.info('Sending response: {0}'.format(response))
|
126 |
+
try:
|
127 |
+
await self.send_message(
|
128 |
+
message,
|
129 |
+
response.get_answer(),
|
130 |
+
response.get_sources_as_text()
|
131 |
)
|
132 |
+
except Exception as e:
|
133 |
+
logger.error('Failed to send response: {0}'.format(e))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
qa_engine/config.py
CHANGED
@@ -36,6 +36,7 @@ class Config:
|
|
36 |
|
37 |
# Discord bot config - optional
|
38 |
discord_token: str = get_env('DISCORD_TOKEN', '-', warn=False)
|
|
|
39 |
num_last_messages: int = int(get_env('NUM_LAST_MESSAGES', 2, warn=False))
|
40 |
use_names_in_context: bool = eval(get_env('USE_NAMES_IN_CONTEXT', 'False', warn=False))
|
41 |
enable_commands: bool = eval(get_env('ENABLE_COMMANDS', 'True', warn=False))
|
|
|
36 |
|
37 |
# Discord bot config - optional
|
38 |
discord_token: str = get_env('DISCORD_TOKEN', '-', warn=False)
|
39 |
+
discotd_channel_ids: list[int] = eval(get_env('DISCORD_CHANNEL_IDS', [], warn=False))
|
40 |
num_last_messages: int = int(get_env('NUM_LAST_MESSAGES', 2, warn=False))
|
41 |
use_names_in_context: bool = eval(get_env('USE_NAMES_IN_CONTEXT', 'False', warn=False))
|
42 |
enable_commands: bool = eval(get_env('ENABLE_COMMANDS', 'True', warn=False))
|