Skip to content

Commit 81ce118

Browse files
stellarnodestellarnode
committed
Various optimizations
1 parent cf57542 commit 81ce118

6 files changed

Lines changed: 582 additions & 160 deletions

File tree

‎youtubevideotranscriptbot/model_params.py‎

Lines changed: 104 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,104 @@
1+
# model_params.py
2+
import openai
3+
import deepseek
4+
import logging
5+
from config import OPENAI_API_KEY
6+
from config import DEEPSEEK_API_KEY
7+
from config import XAI_API_KEY
8+
from config import ANTHROPIC_API_KEY
9+
from config import MODEL_TO_USE # Import the model selection
10+
import tiktoken
11+
import math
12+
import asyncio
13+
14+
# Set up OpenAI
15+
# openai.api_key = OPENAI_API_KEY
16+
17+
logger = logging.getLogger(__name__)
18+
19+
def get_model_params(model=MODEL_TO_USE):
20+
"""
21+
Returns the model parameters based on the selected model.
22+
23+
Args:
24+
model (str): The model to use, either 'gpt' for OpenAI or 'deepseek' for DeepSeek.
25+
26+
Returns:
27+
dict: A dictionary containing the model parameters.
28+
"""
29+
if "gpt" in model.lower():
30+
logger.info(f"Using OpenAI {model} model.")
31+
return {
32+
"tokens_per_chunk": 100000,
33+
"max_chunks_allowed": 4,
34+
"max_tokens": 1024,
35+
"model": model or "gpt-4o-mini",
36+
"client": openai.OpenAI(api_key=OPENAI_API_KEY, base_url="https://api.openai.com/v1")
37+
}
38+
elif "deepseek" in model.lower():
39+
logger.info(f"Using DeepSeek {model}.")
40+
return {
41+
"tokens_per_chunk": 64000,
42+
"max_chunks_allowed": 5,
43+
"max_tokens": 1024,
44+
"model": model or "deepseek-chat",
45+
"client": openai.OpenAI(api_key=DEEPSEEK_API_KEY, base_url="https://api.deepseek.com")
46+
}
47+
elif "grok" in model.lower():
48+
logger.info(f"Using xAI Grok {model}.")
49+
return {
50+
"tokens_per_chunk": 100000,
51+
"max_chunks_allowed": 5,
52+
"max_tokens": 1024,
53+
"model": model or "grok-3-mini",
54+
"client": openai.OpenAI(api_key=XAI_API_KEY, base_url="https://api.x.ai/v1")
55+
}
56+
elif "claude" in model.lower():
57+
logger.info(f"Using Anthropic's Claude {model}.")
58+
return {
59+
"tokens_per_chunk": 180000,
60+
"max_chunks_allowed": 3,
61+
"max_tokens": 1024,
62+
"model": model or "claude-3-haiku-20240307",
63+
"client": openai.OpenAI(api_key=ANTHROPIC_API_KEY, base_url="https://api.anthropic.com/v1/")
64+
}
65+
else:
66+
logger.error("Invalid model selection in config. Please select 'gpt' for OpenAI or 'deepseek' for DeepSeek.")
67+
logger.warning("Falling back to DeepSeek model as default.")
68+
return {
69+
"tokens_per_chunk": 64000,
70+
"max_chunks_allowed": 5,
71+
"max_tokens": 1024,
72+
"model": model or "deepseek-chat",
73+
"client": openai.OpenAI(api_key=DEEPSEEK_API_KEY, base_url="https://api.deepseek.com")
74+
}
75+
76+
77+
# if MODEL_TO_USE:
78+
# if "gpt" in MODEL_TO_USE.lower():
79+
# model_to_use = 1 # OpenAI model
80+
# elif "deepseek" in MODEL_TO_USE.lower():
81+
# model_to_use = 2
82+
# else:
83+
# logger.error("Invalid model selection in config. Please select 'gpt' for OpenAI or 'deepseek' for DeepSeek.")
84+
# logger.warning("Falling back to DeepSeek model as default.")
85+
# model_to_use = 2 # 1 for OpenAI, 2 for DeepSeek
86+
# raise ValueError("Invalid model selection in config. Please select 'gpt' for OpenAI or 'deepseek' for DeepSeek.")
87+
88+
# if model_to_use == 1:
89+
# tokens_per_chunk = 100000
90+
# max_chunks_allowed = 4
91+
# max_tokens = 1024
92+
# model = "gpt-4o-mini"
93+
# client = openai.OpenAI(api_key=OPENAI_API_KEY, base_url="https://api.openai.com/v1")
94+
# logger.info(f"Using OpenAI {model} model for summarization.")
95+
# elif model_to_use == 2:
96+
# tokens_per_chunk = 64000
97+
# max_chunks_allowed = 5
98+
# max_tokens = 1024
99+
# model = "deepseek-chat"
100+
# # for DeepSeek backward compatibility, you can still use `https://api.deepseek.com/v1` as `base_url`.
101+
# client = openai.OpenAI(api_key=DEEPSEEK_API_KEY, base_url="https://api.deepseek.com")
102+
# logger.info(f"Using DeepSeek {model} for summarization.")
103+
# else:
104+
# logger.error("Invalid model selection. Please select 1 for OpenAI or 2 for DeepSeek.")

‎youtubevideotranscriptbot/requirements.txt‎

Lines changed: 2 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -6,4 +6,5 @@ mysql-connector-python==8.1.0
66
isodate==0.6.1
77
openai==1.65.1 # 0.28.0 # Add this line
88
tiktoken==0.9.0
9-
deepseek==1.0.0
9+
deepseek==1.0.0
10+
aiofiles>=0.4.0

‎youtubevideotranscriptbot/summarize.py‎

Lines changed: 58 additions & 43 deletions
Original file line numberDiff line numberDiff line change
@@ -3,11 +3,12 @@
33
import deepseek
44
import logging
55
from translate import translate_text # Import the translation function
6+
from translate import split_text
7+
from model_params import get_model_params
68
from config import OPENAI_API_KEY
79
from config import DEEPSEEK_API_KEY
810
from config import MODEL_TO_USE # Import the model selection
911
import tiktoken
10-
from translate import split_text
1112
import math
1213
import asyncio
1314

@@ -16,34 +17,35 @@
1617

1718
logger = logging.getLogger(__name__)
1819

19-
if MODEL_TO_USE:
20-
if "gpt" in MODEL_TO_USE.lower():
21-
model_to_use = 1 # OpenAI model
22-
elif "deepseek" in MODEL_TO_USE.lower():
23-
model_to_use = 2
24-
else:
25-
logger.error("Invalid model selection in config. Please select 'gpt' for OpenAI or 'deepseek' for DeepSeek.")
26-
logger.warning("Falling back to DeepSeek model as default.")
27-
model_to_use = 2 # 1 for OpenAI, 2 for DeepSeek
28-
raise ValueError("Invalid model selection in config. Please select 'gpt' for OpenAI or 'deepseek' for DeepSeek.")
29-
30-
if model_to_use == 1:
31-
tokens_per_chunk = 100000
32-
max_chunks_allowed = 4
33-
max_tokens = 1024
34-
model = "gpt-4o-mini"
35-
client = openai.OpenAI(api_key=OPENAI_API_KEY, base_url="https://api.openai.com/v1")
36-
logger.info(f"Using OpenAI {model} model for summarization.")
37-
elif model_to_use == 2:
38-
tokens_per_chunk = 64000
39-
max_chunks_allowed = 3
40-
max_tokens = 1024
41-
model = "deepseek-chat"
42-
# for DeepSeek backward compatibility, you can still use `https://api.deepseek.com/v1` as `base_url`.
43-
client = openai.OpenAI(api_key=DEEPSEEK_API_KEY, base_url="https://api.deepseek.com")
44-
logger.info(f"Using DeepSeek {model} for summarization.")
45-
else:
46-
logger.error("Invalid model selection. Please select 1 for OpenAI or 2 for DeepSeek.")
20+
21+
# if MODEL_TO_USE:
22+
# if "gpt" in MODEL_TO_USE.lower():
23+
# model_to_use = 1 # OpenAI model
24+
# elif "deepseek" in MODEL_TO_USE.lower():
25+
# model_to_use = 2
26+
# else:
27+
# logger.error("Invalid model selection in config. Please select 'gpt' for OpenAI or 'deepseek' for DeepSeek.")
28+
# logger.warning("Falling back to DeepSeek model as default.")
29+
# model_to_use = 2 # 1 for OpenAI, 2 for DeepSeek
30+
# raise ValueError("Invalid model selection in config. Please select 'gpt' for OpenAI or 'deepseek' for DeepSeek.")
31+
32+
# if model_to_use == 1:
33+
# tokens_per_chunk = 100000
34+
# max_chunks_allowed = 4
35+
# max_tokens = 1024
36+
# model = "gpt-4o-mini"
37+
# client = openai.OpenAI(api_key=OPENAI_API_KEY, base_url="https://api.openai.com/v1")
38+
# logger.info(f"Using OpenAI {model} model for summarization.")
39+
# elif model_to_use == 2:
40+
# tokens_per_chunk = 64000
41+
# max_chunks_allowed = 5
42+
# max_tokens = 1024
43+
# model = "deepseek-chat"
44+
# # for DeepSeek backward compatibility, you can still use `https://api.deepseek.com/v1` as `base_url`.
45+
# client = openai.OpenAI(api_key=DEEPSEEK_API_KEY, base_url="https://api.deepseek.com")
46+
# logger.info(f"Using DeepSeek {model} for summarization.")
47+
# else:
48+
# logger.error("Invalid model selection. Please select 1 for OpenAI or 2 for DeepSeek.")
4749

4850

4951
async def summarize_chunk(chunk, language):
@@ -64,21 +66,27 @@ def _summarize_sync(chunk, language):
6466
estimated_cost (float): The estimated cost of the OpenAI API call.
6567
word_count (int): The word count of the summary.
6668
"""
69+
model_params = get_model_params(MODEL_TO_USE)
70+
6771
try:
6872
# Define the prompt for summarization
6973
system_role = "You are a master of extracting pearls of knowledge from YouTube video transcripts. You grasp the very essence and distill it in a concise form for users. You always provide the response in the same language in which the transcript is provided. Your answers are always clear, concise and nicely formatted."
7074
prompt = (
7175
f"If I did not have time to read this YouTube video transcript, what are the most important things I absolutely must know. Enlighten me in no more than 200 words. "
7276
f"Always provide your response in the same language as the transcript. In this case it might be in '{language}' language. Here is the transcript itself:\n\n{chunk}"
7377
)
78+
# prompt = (
79+
# f"If I did not have time to read this YouTube video transcript, what are the most important things I absolutely must know. Enlighten me in no more than 200 words. "
80+
# f"Provide your response strictly in the following language: '{language}'. Here is the transcript itself:\n\n{chunk}"
81+
# )
7482

75-
response = client.chat.completions.create(
76-
model=model,
83+
response = model_params["client"].chat.completions.create(
84+
model=model_params["model"],
7785
messages=[
7886
{"role": "system", "content": system_role},
7987
{"role": "user", "content": prompt},
8088
],
81-
max_tokens=max_tokens,
89+
max_tokens=model_params["max_tokens"],
8290
temperature=0.5,
8391
stream=False
8492
)
@@ -113,6 +121,9 @@ async def summarize_text(text, language, num_key_points=3):
113121
word_count (int): The word count of the summary.
114122
"""
115123

124+
tokens_per_chunk, max_chunks_allowed, max_tokens, model, client = get_model_params(MODEL_TO_USE).values()
125+
logger.info(f"Tokens per chunk: {tokens_per_chunk} of type {type(tokens_per_chunk)}")
126+
116127
try:
117128
# Load the encoding for the model you're using (e.g., gpt-3.5-turbo)
118129
encoding = tiktoken.encoding_for_model("gpt-3.5-turbo")
@@ -125,7 +136,7 @@ async def summarize_text(text, language, num_key_points=3):
125136
raise
126137

127138
if token_count:
128-
number_of_chunks = token_count / tokens_per_chunk # 15,000 tokens per chunk for GPT-3.5-turbo
139+
number_of_chunks = token_count / int(tokens_per_chunk) # 15,000 tokens per chunk for GPT-3.5-turbo
129140
logger.info(f"Number of chunks for summarization of the transcript: {number_of_chunks}")
130141

131142
if number_of_chunks > max_chunks_allowed:
@@ -161,17 +172,17 @@ async def summarize_text(text, language, num_key_points=3):
161172
logger.error(f"Summarization failed: {e}")
162173
raise
163174

164-
logger.info(f"Summarization completed for all chunks.\nTotal word count: {total_word_count}.\nTotal tokens used: {total_tokens_used}.\nEstimated cost: ${total_estimated_cost:.2f}")
175+
logger.info(f"Summarization completed for all chunks. Total word count: {total_word_count}. Total tokens used: {total_tokens_used}. Estimated cost: ${total_estimated_cost:.2f}")
165176

166177
return combined_summary.strip(), total_tokens_used, total_estimated_cost, total_word_count
167178

168179

169-
async def translate_summary_async(summary, src_lang, dest_lang):
170-
loop = asyncio.get_event_loop()
171-
return await loop.run_in_executor(None, translate_summary, summary, src_lang, dest_lang)
180+
# async def translate_summary_async(summary, src_lang, dest_lang):
181+
# loop = asyncio.get_event_loop()
182+
# return await loop.run_in_executor(None, translate_summary, summary, src_lang, dest_lang)
172183

173184
# Translate the summary into the target language
174-
def translate_summary(summary, src_lang, dest_lang):
185+
async def translate_summary(summary, src_lang, dest_lang):
175186
"""
176187
Translates the summary into the target language.
177188
@@ -183,8 +194,10 @@ def translate_summary(summary, src_lang, dest_lang):
183194
Returns:
184195
translated_summary (str): The translated summary.
185196
"""
197+
model_params = get_model_params(MODEL_TO_USE)
198+
186199
try:
187-
translated_summary = translate_text(summary, src_lang=src_lang, dest_lang=dest_lang)
200+
translated_summary, total_tokens_used, total_estimated_cost, total_word_count = await translate_text(summary, src_lang=src_lang, dest_lang=dest_lang, model_params=model_params)
188201
if not translated_summary:
189202
raise Exception("Translation failed.")
190203
return translated_summary
@@ -212,19 +225,21 @@ async def handle_summarization_request(text, original_language, target_language,
212225

213226
logger.info(f"Summarization is handled for original language '{original_language}' and target '{target_language}'")
214227

228+
model_params = get_model_params(MODEL_TO_USE)
229+
215230
try:
216-
# Summarize the original transcript
231+
217232
summary, tokens_used, estimated_cost, word_count = await summarize_text(text, original_language, num_key_points)
218-
219233
if target_language == 'orig':
220234
target_language = original_language
235+
221236
logger.info(f"Summary generated in {target_language} language. Translation skipped.")
222237

223238
# Translate the summary if the target language is not the original language
224239
if target_language != original_language:
225-
summary = await translate_summary_async(summary, src_lang=original_language, dest_lang=target_language)
240+
summary = await translate_summary(summary, src_lang=original_language, dest_lang=target_language)
226241

227-
return summary, tokens_used, estimated_cost, word_count, model
242+
return summary, tokens_used, estimated_cost, word_count, model_params['model']
228243
except Exception as e:
229244
logger.error(f"Failed to handle summarization request: {e}")
230245
raise

0 commit comments

Comments
 (0)