Compare commits
12 Commits
| Author | SHA1 | Date | |
|---|---|---|---|
|
d231707572 | ||
|
|
6a88c92b9b | ||
|
|
e391a86cb5 | ||
|
|
4c90088308 | ||
|
|
f2c7dcbeb6 | ||
|
|
6b27149460 | ||
|
|
5bf005cd81 | ||
|
|
e05e05f829 | ||
|
|
07485f9a23 | ||
|
|
4d4b23177b | ||
|
|
26d2d19329 | ||
|
|
fb7a03b16a |
1
.gitignore
vendored
1
.gitignore
vendored
@@ -2,3 +2,4 @@
|
||||
__pycache__/
|
||||
outputs/*
|
||||
venv/
|
||||
|
||||
|
||||
118
README.md
118
README.md
@@ -1,59 +1,59 @@
|
||||
# FWAL WebUI (Faster Whisper And LLM WebUI) by swrneko
|
||||
|
||||
<div align="center">
|
||||
<img src="https://count.getloli.com/get/@swrneko-faster-whisper-llm?theme=rule34"/>
|
||||
</div>
|
||||
|
||||
## Screenshots
|
||||
<div align="center">
|
||||
<div>
|
||||
<img src="src/1.png" style="object-fit: cover;"/>
|
||||
</div>
|
||||
<div style="align-items: center;">
|
||||
<img src="src/2.png" style="width: 49.7%; object-fit: cover;"/>
|
||||
<img src="src/3.png" style="width: 49.7%; object-fit: cover;"/>
|
||||
</div>
|
||||
</div>
|
||||
|
||||
|
||||
## Requirements
|
||||
- python-conda or miniconda;
|
||||
- python 3.10 or above;
|
||||
- linux (windows not tested but probably working);
|
||||
|
||||
Python requirements are listed in `requirements.txt`.
|
||||
|
||||
## Installation
|
||||
1. Clone repository:
|
||||
```
|
||||
git clone https://github.com/swrneko/faster-whisper-n-ionet-llm.git
|
||||
cd faster-whisper-n-ionet-llm
|
||||
```
|
||||
also (if not insatlled)
|
||||
- Insatll conda:
|
||||
```
|
||||
wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh -O miniconda.sh && bash miniconda.sh
|
||||
```
|
||||
|
||||
2. Create virtual env:
|
||||
```
|
||||
conda create -n faster-whisper-n-ionet-llm python=3.10
|
||||
conda activate faster-whisper-n-ionet-llm
|
||||
conda install nvidia::cudnn cuda-version=12
|
||||
```
|
||||
|
||||
3. Install requirements:
|
||||
```
|
||||
pip install -r requirements.txt
|
||||
```
|
||||
|
||||
4. Get api key from [io.net](https://ai.io.net/ai/api-keys) and insert into `.env` file (need to create it in root of repository directory).
|
||||
It should looks like this:
|
||||
```
|
||||
API_KEY='your_api_key_without_qoutes'
|
||||
```
|
||||
|
||||
5. Done! Now you can just run it like that:
|
||||
```shell
|
||||
python app.py
|
||||
```
|
||||
# FWAL WebUI (Faster Whisper And LLM WebUI) by swrneko
|
||||
|
||||
<div align="center">
|
||||
<img src="https://count.getloli.com/get/@swrneko-faster-whisper-llm?theme=rule34"/>
|
||||
</div>
|
||||
|
||||
## Screenshots
|
||||
<div align="center">
|
||||
<div>
|
||||
<img src="src/1.png" style="object-fit: cover;"/>
|
||||
</div>
|
||||
<div style="align-items: center;">
|
||||
<img src="src/2.png" style="width: 49.7%; object-fit: cover;"/>
|
||||
<img src="src/3.png" style="width: 49.7%; object-fit: cover;"/>
|
||||
</div>
|
||||
</div>
|
||||
|
||||
|
||||
## Requirements
|
||||
- python-conda or miniconda;
|
||||
- python 3.10 or above;
|
||||
- linux (windows not tested but probably working);
|
||||
|
||||
Python requirements are listed in `requirements.txt`.
|
||||
|
||||
## Installation
|
||||
1. Clone repository:
|
||||
```
|
||||
git clone https://github.com/swrneko/faster-whisper-n-ionet-llm.git
|
||||
cd faster-whisper-n-ionet-llm
|
||||
```
|
||||
also (if not insatlled)
|
||||
- Insatll conda:
|
||||
```
|
||||
wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh -O miniconda.sh && bash miniconda.sh
|
||||
```
|
||||
|
||||
2. Create virtual env:
|
||||
```
|
||||
conda create -n faster-whisper-n-ionet-llm python=3.10
|
||||
conda activate faster-whisper-n-ionet-llm
|
||||
conda install nvidia::cudnn cuda-version=12
|
||||
```
|
||||
|
||||
3. Install requirements:
|
||||
```
|
||||
pip install -r requirements.txt
|
||||
```
|
||||
|
||||
4. Get api key from [io.net](https://ai.io.net/ai/api-keys) and insert into `.env` file (need to create it in root of repository directory).
|
||||
It should looks like this:
|
||||
```
|
||||
API_KEY='your_api_key_without_qoutes'
|
||||
```
|
||||
|
||||
5. Done! Now you can just run it like that:
|
||||
```shell
|
||||
python app.py
|
||||
```
|
||||
|
||||
340
app.py
340
app.py
@@ -1,195 +1,145 @@
|
||||
import gradio as gr
|
||||
from pathlib import Path
|
||||
|
||||
# Подгрузка сервисов
|
||||
from services.llm import Llm
|
||||
from services.fasterWhisper import FasterWhisper
|
||||
from services.convertMdToPdf import ConvertMdToPdf
|
||||
|
||||
# Загрузка параметров конфигурации
|
||||
from config import *
|
||||
|
||||
# Функция транскрибации
|
||||
def generateByCondition(api_key, llm_model, system_prompt, recognized_text, llm_temperature, is_pipeline_enabled, trigger, isSaveFile, filename, filenamePdf):
|
||||
llm = Llm(api_key)
|
||||
|
||||
# если чекбокс включен и событие было change → обрабатываем
|
||||
if is_pipeline_enabled and trigger == "change":
|
||||
result, md = llm.generate(llm_model, system_prompt, recognized_text, llm_temperature)
|
||||
|
||||
# Конвертируем текст с латексом в юникод
|
||||
pdf, unicodeText = ConvertMdToPdf().convertLatexToText(md)
|
||||
|
||||
if isSaveFile:
|
||||
savePdf(filenamePdf, pdf)
|
||||
saveFile(filename, result)
|
||||
|
||||
return result, unicodeText
|
||||
|
||||
# если чекбокс выключен и событие было click → обрабатываем
|
||||
if not is_pipeline_enabled and trigger == "click":
|
||||
result, md = llm.generate(llm_model, system_prompt, recognized_text, llm_temperature)
|
||||
|
||||
# Конвертируем текст с латексом в юникод
|
||||
pdf, unicodeText = ConvertMdToPdf().convertLatexToText(md)
|
||||
|
||||
if isSaveFile:
|
||||
savePdf(filenamePdf, pdf)
|
||||
saveFile(filename, result)
|
||||
|
||||
return result, unicodeText
|
||||
|
||||
# если нет чекбокса и было событие change
|
||||
return gr.skip(), gr.skip()
|
||||
|
||||
|
||||
def savePdf(filename, pdf):
|
||||
directory = Path(OUTPUT_PATH)
|
||||
filePath = directory / filename
|
||||
filePath.parent.mkdir(parents=True, exist_ok=True)
|
||||
pdf.save(filePath)
|
||||
|
||||
# Функция сохранеhния файла
|
||||
def saveFile(filename, text):
|
||||
directory = Path(OUTPUT_PATH)
|
||||
filePath = directory / filename
|
||||
filePath.parent.mkdir(parents=True, exist_ok=True)
|
||||
filePath.write_text(text, encoding='utf-8')
|
||||
|
||||
# ConvertMdToPdf().convert(text)
|
||||
|
||||
# Функция для динамического обновления кнопки в зависимости от состояния checkbox
|
||||
def updateButton(isChecked):
|
||||
if not isChecked:
|
||||
variant = 'primary'
|
||||
else:
|
||||
variant = 'secondary'
|
||||
|
||||
return gr.update(interactive=not isChecked, variant=variant)
|
||||
|
||||
|
||||
def updateTextbox(isChecked):
|
||||
return gr.update(visible=isChecked)
|
||||
|
||||
###################################################
|
||||
# ____ ___.___ ___. .__ #
|
||||
#| | \ | \_ |__ ____ | | ______ _ __#
|
||||
#| | / | | __ \_/ __ \| | / _ \ \/ \/ /#
|
||||
#| | /| | | \_\ \ ___/| |_( <_> ) / #
|
||||
#|______/ |___| |___ /\___ >____/\____/ \/\_/ #
|
||||
# \/ \/ #
|
||||
###################################################
|
||||
|
||||
with gr.Blocks() as demo:
|
||||
gr.HTML('''
|
||||
<div align=center>
|
||||
<h1>
|
||||
Faster Whisper WebUI
|
||||
</h1>
|
||||
</div>
|
||||
''')
|
||||
|
||||
with gr.Row():
|
||||
# Вкладка с основным взаимодействием
|
||||
with gr.Tab('Actions'):
|
||||
isPipelineEnabledCheckbox = gr.Checkbox(label='is pipeline enabled', value=True, interactive=True)
|
||||
|
||||
with gr.Row():
|
||||
with gr.Accordion(label='Recognization and integration'):
|
||||
with gr.Column():
|
||||
audioFile = gr.Audio(label='Load audio for transcribe', type="filepath")
|
||||
images = gr.Files(label='Upload images', file_types=['image'])
|
||||
recognizeBtn = gr.Button('recognize and integrate', variant='primary')
|
||||
|
||||
with gr.Accordion(label='Recognized text'):
|
||||
recognizedText = gr.TextArea(label='')
|
||||
|
||||
with gr.Accordion(label='LLM'):
|
||||
with gr.Column():
|
||||
refineTextBtn = gr.Button('refine text', variant='secondary', interactive=False)
|
||||
|
||||
with gr.Accordion(label='Refined text raw'):
|
||||
refinedText = gr.Textbox(label='', show_copy_button=True)
|
||||
|
||||
with gr.Accordion(label='Refined text md formated'):
|
||||
refinedTextMD = gr.Markdown(label='')
|
||||
|
||||
# Вкладка с настройками
|
||||
with gr.Tab('Settings'):
|
||||
with gr.Column():
|
||||
# Первое поле на всю ширину в акордионе настроек
|
||||
with gr.Accordion('File settings'):
|
||||
saveFileCheckbox = gr.Checkbox(label='save file', value=True, interactive=True)
|
||||
filename = gr.Textbox(label='Output filename', value='output.txt', interactive=True)
|
||||
filenamePdf = gr.Textbox(label='Output filename for pdf', value='output.pdf', interactive=True)
|
||||
|
||||
# Акордион настроек faster whisper
|
||||
with gr.Accordion(label='Faster whisper settings'):
|
||||
with gr.Row():
|
||||
# Левая колонка в акордионе
|
||||
with gr.Column():
|
||||
device = gr.Dropdown(label='Device', choices=["cpu", "cuda"], value="cuda", interactive=True)
|
||||
compute_type = gr.Dropdown(label='compute_type', choices=["auto", "int8", "float16", "float32"], value="auto", interactive=True)
|
||||
fastWhisperModel = gr.Dropdown(label='Model', choices=FAST_WHISPER_MODELS, value=FAST_WHISPER_MODELS[11], interactive=True)
|
||||
|
||||
beamSize = gr.Number(label='beam_size', value=8, interactive=True)
|
||||
noSpeechThreshold = gr.Number(label='no_speech_threshold', value=0.5, interactive=True)
|
||||
vadFilter = gr.Checkbox(label='vad_filter', value=True, interactive=True)
|
||||
wordTimestamps = gr.Checkbox(label='word_timestamps', value=True, interactive=True)
|
||||
conditionOnPreviousText = gr.Checkbox(label='condition_on_previous_text', value=False, interactive=True)
|
||||
|
||||
|
||||
|
||||
# Правая колонка в акордионе
|
||||
with gr.Column():
|
||||
with gr.Accordion(label='Vad parameters'):
|
||||
minSilenceDurationMs = gr.Number(label='min_silence_duration_ms', value=300, interactive=True)
|
||||
speechPadMs = gr.Number(label='speech_pad_ms', value=200, interactive=True)
|
||||
|
||||
with gr.Accordion(label='Temperature'):
|
||||
temp0 = gr.Number(label='temp_0', value=0.0, interactive=True)
|
||||
temp1 = gr.Number(label='temp_1', value=0.2, interactive=True)
|
||||
temp2 = gr.Number(label='temp_2', value=0.4, interactive=True)
|
||||
|
||||
# Нижний акордион настроек для api ключа llm
|
||||
with gr.Accordion(label='ai.io.net api settings'):
|
||||
apiKey = gr.Textbox(label='API key', value=DEFAULT_API_KEY, interactive=True)
|
||||
|
||||
with gr.Accordion(label='System prompt'):
|
||||
systemPrompt = gr.Textbox(label='', value=DEFAULT_SYSTEM_PROMPT, interactive=True)
|
||||
|
||||
with gr.Row():
|
||||
llmModel = gr.Dropdown(label='models', choices=LLM_MODELS, value=LLM_MODELS[1], interactive=True)
|
||||
llmTemperature = gr.Number(label='Temperature', value=0.8, interactive=True )
|
||||
|
||||
######################################################################
|
||||
#.____ .__ ___. .__ #
|
||||
#| | ____ ____ |__| ____ \_ |__ ____ | | ______ _ __#
|
||||
#| | / _ \ / ___\| |/ ___\ | __ \_/ __ \| | / _ \ \/ \/ /#
|
||||
#| |__( <_> ) /_/ > \ \___ | \_\ \ ___/| |_( <_> ) / #
|
||||
#|_______ \____/\___ /|__|\___ > |___ /\___ >____/\____/ \/\_/ #
|
||||
# \/ /_____/ \/ \/ \/ #
|
||||
######################################################################
|
||||
|
||||
isPipelineEnabledCheckbox.change(updateButton, inputs=[isPipelineEnabledCheckbox], outputs=refineTextBtn)
|
||||
saveFileCheckbox.change(updateTextbox, inputs=saveFileCheckbox, outputs=filename)
|
||||
saveFileCheckbox.change(updateTextbox, inputs=saveFileCheckbox, outputs=filenamePdf)
|
||||
|
||||
recognizeBtn.click(FasterWhisper().recognize, outputs=[recognizedText], inputs=[fastWhisperModel, device, compute_type, audioFile, beamSize, vadFilter, minSilenceDurationMs, speechPadMs, temp0, temp1, temp2, wordTimestamps, noSpeechThreshold, conditionOnPreviousText])
|
||||
|
||||
# Если пайплайн включен то тогда делаем автоматически
|
||||
# автоматический пайплайн
|
||||
recognizedText.change(
|
||||
generateByCondition,
|
||||
inputs=[apiKey, llmModel, systemPrompt, recognizedText, llmTemperature, isPipelineEnabledCheckbox, gr.State("change"), saveFileCheckbox, filename, filenamePdf],
|
||||
outputs=[refinedText, refinedTextMD]
|
||||
)
|
||||
|
||||
# ручной запуск по кнопке
|
||||
refineTextBtn.click(
|
||||
generateByCondition,
|
||||
inputs=[apiKey, llmModel, systemPrompt, recognizedText, llmTemperature, isPipelineEnabledCheckbox, gr.State("click"), saveFileCheckbox, filename, filenamePdf],
|
||||
outputs=[refinedText, refinedTextMD]
|
||||
)
|
||||
|
||||
demo.launch()
|
||||
import gradio as gr
|
||||
|
||||
# Загрузка параметров конфигурации
|
||||
from config import *
|
||||
|
||||
# Подгрузка сервисов
|
||||
from services.llm_factory import get_llm_provider
|
||||
|
||||
|
||||
|
||||
# Загрузка доп. модулей
|
||||
from handlers.gradioHandler import GradioHandlers
|
||||
from handlers.fileHandlers import FileHandlers
|
||||
from services.fasterWhisper import FasterWhisper
|
||||
from handlers.convertMdToPdf import ConvertMdToPdf
|
||||
from handlers.glueAudio import GlueAudio
|
||||
|
||||
gh = GradioHandlers(get_llm_provider, ConvertMdToPdf, FileHandlers, FasterWhisper, GlueAudio)
|
||||
|
||||
def main():
|
||||
with gr.Blocks() as demo:
|
||||
gr.HTML('''
|
||||
<div align=center>
|
||||
<h1>
|
||||
Faster Whisper WebUI
|
||||
</h1>
|
||||
</div>
|
||||
''')
|
||||
|
||||
with gr.Row():
|
||||
# Вкладка с основным взаимодействием
|
||||
with gr.Tab('Actions'):
|
||||
isPipelineEnabledCheckbox = gr.Checkbox(label='is pipeline enabled', value=True, interactive=True)
|
||||
|
||||
with gr.Row():
|
||||
with gr.Accordion(label='Recognization and integration'):
|
||||
with gr.Column():
|
||||
audioFiles = gr.Files(label='Load audio for transcribe', type="filepath")
|
||||
images = gr.Files(label='Upload images', file_types=['image'])
|
||||
recognizeBtn = gr.Button('recognize and integrate', variant='primary')
|
||||
|
||||
with gr.Accordion(label='Recognized text'):
|
||||
recognizedText = gr.TextArea(label='')
|
||||
|
||||
with gr.Accordion(label='LLM'):
|
||||
with gr.Column():
|
||||
refineTextBtn = gr.Button('refine text', variant='secondary', interactive=False)
|
||||
|
||||
with gr.Accordion(label='Refined text raw'):
|
||||
refinedText = gr.Textbox(label='', show_copy_button=True)
|
||||
|
||||
with gr.Accordion(label='Refined text md formated'):
|
||||
refinedTextMD = gr.Markdown(label='')
|
||||
|
||||
# Вкладка с настройками
|
||||
with gr.Tab('Settings'):
|
||||
with gr.Column():
|
||||
# Первое поле на всю ширину в акордионе настроек
|
||||
with gr.Accordion('File settings'):
|
||||
saveFileCheckbox = gr.Checkbox(label='save file', value=True, interactive=True)
|
||||
filename = gr.Textbox(label='Output filename', value='output.txt', interactive=True)
|
||||
filenamePdf = gr.Textbox(label='Output filename for pdf', value='output.pdf', interactive=True)
|
||||
|
||||
# Акордион настроек faster whisper
|
||||
with gr.Accordion(label='Faster whisper settings'):
|
||||
with gr.Row():
|
||||
# Левая колонка в акордионе
|
||||
with gr.Column():
|
||||
device = gr.Dropdown(label='Device', choices=DEVICES, value=DEVICES[1], interactive=True)
|
||||
compute_type = gr.Dropdown(label='compute_type', choices=COMPUTE_TYPE, value=COMPUTE_TYPE[0], interactive=True)
|
||||
fastWhisperModel = gr.Dropdown(label='Model', choices=FAST_WHISPER_MODELS, value=FAST_WHISPER_MODELS[11], interactive=True)
|
||||
|
||||
beamSize = gr.Number(label='beam_size', value=8, interactive=True)
|
||||
noSpeechThreshold = gr.Number(label='no_speech_threshold', value=0.5, interactive=True)
|
||||
vadFilter = gr.Checkbox(label='vad_filter', value=True, interactive=True)
|
||||
wordTimestamps = gr.Checkbox(label='word_timestamps', value=True, interactive=True)
|
||||
conditionOnPreviousText = gr.Checkbox(label='condition_on_previous_text', value=False, interactive=True)
|
||||
|
||||
# Правая колонка в акордионе
|
||||
with gr.Column():
|
||||
with gr.Accordion(label='Vad parameters'):
|
||||
minSilenceDurationMs = gr.Number(label='min_silence_duration_ms', value=300, interactive=True)
|
||||
speechPadMs = gr.Number(label='speech_pad_ms', value=200, interactive=True)
|
||||
|
||||
with gr.Accordion(label='Temperature'):
|
||||
temp0 = gr.Number(label='temp_0', value=0.0, interactive=True)
|
||||
temp1 = gr.Number(label='temp_1', value=0.2, interactive=True)
|
||||
temp2 = gr.Number(label='temp_2', value=0.4, interactive=True)
|
||||
|
||||
# Нижний акордион настроек для api ключа llm
|
||||
with gr.Accordion(label='LLM settings'):
|
||||
apiKey = gr.Textbox(label='API key (required for io.net, Gemini)', value=DEFAULT_API_KEY, interactive=True)
|
||||
|
||||
with gr.Accordion(label='System prompt'):
|
||||
systemPrompt = gr.Textbox(label='', value=DEFAULT_SYSTEM_PROMPT, interactive=True)
|
||||
|
||||
with gr.Row():
|
||||
# ВЫБОР ПРОВАЙДЕРА
|
||||
llmProvider = gr.Dropdown(
|
||||
label='LLM Provider',
|
||||
choices=LLM_PROVIDERS,
|
||||
value=LLM_PROVIDERS[0],
|
||||
interactive=True
|
||||
)
|
||||
# СПИСОК МОДЕЛЕЙ (теперь зависит от провайдера)
|
||||
llmModel = gr.Dropdown(
|
||||
label='Models',
|
||||
choices=LLM_MODELS[LLM_PROVIDERS[0]], # Модели для провайдера по умолчанию
|
||||
value=LLM_MODELS[LLM_PROVIDERS[0]][1],
|
||||
interactive=True
|
||||
)
|
||||
llmTemperature = gr.Number(label='Temperature', value=0.8, interactive=True)
|
||||
|
||||
isPipelineEnabledCheckbox.change(gh.updateButton, inputs=[isPipelineEnabledCheckbox], outputs=refineTextBtn)
|
||||
saveFileCheckbox.change(gh.updateTextbox, inputs=saveFileCheckbox, outputs=filename)
|
||||
saveFileCheckbox.change(gh.updateTextbox, inputs=saveFileCheckbox, outputs=filenamePdf)
|
||||
|
||||
recognizeBtn.click(gh.handleRecognizeBtn, outputs=[recognizedText], inputs=[audioFiles, fastWhisperModel, device, compute_type, beamSize, vadFilter, minSilenceDurationMs, speechPadMs, temp0, temp1, temp2, wordTimestamps, noSpeechThreshold, conditionOnPreviousText, gr.State(GLUED_AUDIO_FILENAME), gr.State(OUTPUT_PATH)])
|
||||
|
||||
# Если пайплайн включен то тогда делаем автоматически
|
||||
# автоматический пайплайн
|
||||
recognizedText.change(
|
||||
gh.generateByCondition,
|
||||
inputs=[apiKey, llmProvider, llmModel, systemPrompt, recognizedText, llmTemperature, isPipelineEnabledCheckbox, gr.State("change"), saveFileCheckbox, filename, filenamePdf, gr.State(OUTPUT_PATH)],
|
||||
outputs=[refinedText, refinedTextMD]
|
||||
)
|
||||
|
||||
|
||||
|
||||
# ручной запуск по кнопке
|
||||
llmProvider.change(
|
||||
gh.update_model_dropdown,
|
||||
inputs=llmProvider,
|
||||
outputs=llmModel
|
||||
)
|
||||
refineTextBtn.click(
|
||||
gh.generateByCondition,
|
||||
inputs=[apiKey, llmProvider, llmModel, systemPrompt, recognizedText, llmTemperature, isPipelineEnabledCheckbox, gr.State("click"), saveFileCheckbox, filename, filenamePdf, gr.State(OUTPUT_PATH)],
|
||||
outputs=[refinedText, refinedTextMD]
|
||||
)
|
||||
|
||||
demo.launch()
|
||||
|
||||
if __name__ == '__main__':
|
||||
main()
|
||||
|
||||
135
config.py
135
config.py
@@ -1,48 +1,87 @@
|
||||
import os
|
||||
from dotenv import load_dotenv
|
||||
|
||||
load_dotenv()
|
||||
|
||||
FAST_WHISPER_MODELS = ['tiny', 'base', 'small', 'medium', 'large-v1', 'large-v2', 'large-v3', 'large', 'distil-large-v2', 'distil-large-v3', 'distil-large-v3.5', 'large-v3-turbo', 'turbo']
|
||||
LLM_MODELS = ['openai/gpt-oss-120b', 'Qwen/Qwen3-235B-A22B-Thinking-2507', 'deepseek-ai/DeepSeek-R1-0528', 'meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8', 'openai/gpt-oss-20b', 'Intel/Qwen3-Coder-480B-A35B-Instruct-int4-mixed-ar', 'meta-llama/Llama-3.2-90B-Vision-Instruct', 'mistralai/Mistral-Nemo-Instruct-2407', 'Qwen/Qwen2.5-VL-32B-Instruct', 'meta-llama/Llama-3.3-70B-Instruct', 'mistralai/Devstral-Small-2505', 'mistralai/Magistral-Small-2506', 'mistralai/Mistral-Large-Instruct-2411', 'CohereForAI/aya-expanse-32b']
|
||||
|
||||
|
||||
# Стандартный API ключ
|
||||
DEFAULT_API_KEY=os.getenv('API_KEY')
|
||||
# Задаем выходную директорию
|
||||
OUTPUT_PATH='outputs'
|
||||
|
||||
DEFAULT_SYSTEM_PROMPT='''You are a diligent university student who has recorded a lecture as an audio file and later transcribed it into raw text.
|
||||
Your task is to rewrite this unstructured transcript into a clear, logically organized, and detailed lecture summary (lecture notes).
|
||||
|
||||
Guidelines:
|
||||
1. Structure:
|
||||
- Organize the text into a hierarchy of sections and subsections.
|
||||
- Use headings, bullet points, or numbering where appropriate.
|
||||
- Present the material in a logical flow (from introduction → main points → details → examples → conclusion).
|
||||
|
||||
2. Clarity & Cohesion:
|
||||
- Remove filler words, repetitions, and irrelevant fragments.
|
||||
- Rewrite incomplete sentences into full, grammatically correct sentences.
|
||||
- Ensure smooth transitions between topics, making the summary feel continuous and well-connected.
|
||||
|
||||
3. Depth & Detail:
|
||||
- Capture all important concepts, definitions, examples, and explanations from the lecture.
|
||||
- Expand shorthand or fragmented thoughts into full, precise explanations.
|
||||
- Where appropriate, rephrase or clarify confusing passages for better understanding.
|
||||
|
||||
4. Accuracy:
|
||||
- Preserve the lecturer’s original meaning, intent, and terminology.
|
||||
- Avoid adding personal opinions or new information that was not in the lecture.
|
||||
|
||||
5. Style:
|
||||
- Write in a formal, academic tone suitable for study notes.
|
||||
- Aim for readability: concise sentences, but thorough coverage of concepts.
|
||||
- Use emphasis (e.g., bold or italic text) only when it improves comprehension.
|
||||
|
||||
Final Output: A cohesive, detailed, and well-structured lecture summary, suitable for later studying and revision.
|
||||
Use only russian language!
|
||||
USE LATEX IN DOLLAR SIGN ($)!
|
||||
EXTRA BIG LENTH OF CONSPECT!
|
||||
'''
|
||||
|
||||
import os
|
||||
from dotenv import load_dotenv
|
||||
|
||||
load_dotenv()
|
||||
|
||||
FAST_WHISPER_MODELS = ['tiny', 'base', 'small', 'medium', 'large-v1', 'large-v2', 'large-v3', 'large', 'distil-large-v2', 'distil-large-v3', 'distil-large-v3.5', 'large-v3-turbo', 'turbo']
|
||||
DEVICES = ['cpu', 'cuda']
|
||||
COMPUTE_TYPE = ['auto', 'int8', 'float16', 'float32']
|
||||
|
||||
# Стандартный API ключ
|
||||
DEFAULT_API_KEY=os.getenv('API_KEY')
|
||||
|
||||
# Словарь провайдеров и их моделей
|
||||
LLM_PROVIDERS = ['io.net', 'Gemini', 'gpt4free']
|
||||
LLM_MODELS = {
|
||||
'io.net': [
|
||||
'openai/gpt-oss-120b', 'Qwen/Qwen3-235B-A22B-Thinking-2507',
|
||||
'deepseek-ai/DeepSeek-R1-0528', 'meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8',
|
||||
'openai/gpt-oss-20b', 'Intel/Qwen3-Coder-480B-A35B-Instruct-int4-mixed-ar',
|
||||
'meta-llama/Llama-3.2-90B-Vision-Instruct', 'mistralai/Mistral-Nemo-Instruct-2407',
|
||||
'Qwen/Qwen2.5-VL-32B-Instruct', 'meta-llama/Llama-3.3-70B-Instruct',
|
||||
'mistralai/Devstral-Small-2505', 'mistralai/Magistral-Small-2506',
|
||||
'mistralai/Mistral-Large-Instruct-2411', 'CohereForAI/aya-expanse-32b'
|
||||
],
|
||||
'Gemini': [
|
||||
'gemini-2.5-pro',
|
||||
'gemini-2.5-flash',
|
||||
'gemini-2.5-flash-lite'
|
||||
],
|
||||
'gpt4free': [ # Модели могут меняться, проверьте документацию g4f
|
||||
'default',
|
||||
'gpt-4',
|
||||
'sonar-reasoning',
|
||||
'command-r-plus',
|
||||
'llama-3.3-70b',
|
||||
'hermes-3-llama-3.1-405b'
|
||||
'qwen-3-235b',
|
||||
'gpt-4o-mini',
|
||||
'deepseek-r1',
|
||||
'PollinationsAI:gpt-5-nano'
|
||||
]
|
||||
}
|
||||
|
||||
# Задаем выходную директорию
|
||||
OUTPUT_PATH='outputs'
|
||||
|
||||
GLUED_AUDIO_FILENAME='glued.mp3'
|
||||
|
||||
DEFAULT_SYSTEM_PROMPT='''
|
||||
You are a smart university student creating easy-to-understand study notes summary of lesson for a classmate who is a beginner. Your source is a raw text/audio transcript.
|
||||
|
||||
GOAL: rewrite the information into a clear, structured summary in RUSSIAN.
|
||||
|
||||
KEY RULES FOR CONTENT:
|
||||
1. **Logical Structure:** Use Markdown headers (#, ##), bullet points, and short paragraphs.
|
||||
2. **No "Water":** Remove filler words. Keep only practical information.
|
||||
3. **Student Tone:** Write naturally, as if sharing notes with a friend. Avoid robotic phrases like "It is important to note".
|
||||
|
||||
KEY RULES FOR LATEX (CRITICAL FOR PYLATEXENC):
|
||||
1. **Math Mode:** ANY variable (like t, L, C), number in a formula, or equation MUST be wrapped in dollar signs `$`.
|
||||
* BAD: i(t) = i_pr + i_sv
|
||||
* GOOD: $i(t) = i_{pr} + i_{sv}$
|
||||
2. **Subscripts:** Always use curly braces `{}` for subscripts longer than one character.
|
||||
* BAD: $i_pr$
|
||||
* GOOD: $i_{pr}$ (or $i_{пр}$ if using cyrillic)
|
||||
3. **Symbols:** Use standard LaTeX commands for symbols.
|
||||
* Arrow: use `\to` (e.g., $t \to \infty$).
|
||||
* Infinity: use `\infty`.
|
||||
* Multiplication: use `\cdot` or just space.
|
||||
4. **Consistency:** Never leave a mathematical symbol as plain text. If you mention "current i", write "ток $i$".
|
||||
|
||||
EXAMPLE OF DESIRED OUTPUT FORMAT:
|
||||
# Тема лекции
|
||||
## Основные понятия
|
||||
* **Переходный процесс** — это когда цепь перестраивается с одного режима на другой (например, щелкнули выключателем).
|
||||
* Математически это описывается дифференциальными уравнениями. Порядок уравнения = количеству реактивных элементов ($L$ и $C$).
|
||||
|
||||
## Классический метод
|
||||
Решение ищется в виде суммы двух частей:
|
||||
$$i(t) = i_{pr} + i_{sv}$$
|
||||
|
||||
1. **Принужденная составляющая** ($i_{pr}$) — это режим, который установится в будущем, когда все успокоится ($t \to \infty$).
|
||||
2. **Свободная составляющая** ($i_{sv}$) — это то, что происходит "само по себе" из-за энергии, запасенной в $L$ и $C$.
|
||||
|
||||
***
|
||||
STRICTLY FOLLOW THESE FORMATTING RULES. OUTPUT IN RUSSIAN.
|
||||
'''
|
||||
|
||||
0
handlers/__init__.py
Normal file
0
handlers/__init__.py
Normal file
@@ -1,29 +1,36 @@
|
||||
import re
|
||||
from pylatexenc.latex2text import LatexNodes2Text
|
||||
from markdown_pdf import MarkdownPdf
|
||||
from markdown_pdf import Section
|
||||
|
||||
class ConvertMdToPdf:
|
||||
# Конвертирует md в pdf
|
||||
def convertLatexToText(self, text:str):
|
||||
# Обрабатываем только математические выражения
|
||||
text = re.sub(
|
||||
r'\$\$(.*?)\$\$|\$(.*?)\$',
|
||||
self.replace_math,
|
||||
text,
|
||||
flags=re.DOTALL
|
||||
)
|
||||
pdf = MarkdownPdf(toc_level=0, optimize=True)
|
||||
pdf.add_section(Section(text))
|
||||
return pdf, text
|
||||
|
||||
def replace_math(self, match):
|
||||
math_content = match.group(1) or match.group(2) # $$...$$ или $...$
|
||||
try:
|
||||
# Преобразуем только математическое выражение
|
||||
converted = LatexNodes2Text().latex_to_text(math_content)
|
||||
|
||||
return converted
|
||||
except:
|
||||
return math_content # В случае ошибки оставляем как есть
|
||||
|
||||
import re
|
||||
from pylatexenc.latex2text import LatexNodes2Text
|
||||
from markdown_pdf import MarkdownPdf
|
||||
from markdown_pdf import Section
|
||||
|
||||
class ConvertMdToPdf:
|
||||
# Конвертирует md в pdf
|
||||
def convertLatexToText(self, text:str):
|
||||
'''
|
||||
Функция для конвертации LaTeX в текст;
|
||||
|
||||
Args:
|
||||
:param text: текст содержащий LaTeX.
|
||||
'''
|
||||
|
||||
# Обрабатываем только математические выражения
|
||||
text = re.sub(
|
||||
r'\$\$(.*?)\$\$|\$(.*?)\$',
|
||||
self.replace_math,
|
||||
text,
|
||||
flags=re.DOTALL
|
||||
)
|
||||
pdf = MarkdownPdf(toc_level=0, optimize=True)
|
||||
pdf.add_section(Section(text))
|
||||
return pdf, text
|
||||
|
||||
def replace_math(self, match):
|
||||
math_content = match.group(1) or match.group(2) # $$...$$ или $...$
|
||||
try:
|
||||
# Преобразуем только математическое выражение
|
||||
converted = LatexNodes2Text().latex_to_text(math_content)
|
||||
|
||||
return converted
|
||||
except:
|
||||
return math_content # В случае ошибки оставляем как есть
|
||||
|
||||
31
handlers/fileHandlers.py
Normal file
31
handlers/fileHandlers.py
Normal file
@@ -0,0 +1,31 @@
|
||||
from pathlib import Path
|
||||
|
||||
# Для аннотации типов
|
||||
from markdown_pdf import MarkdownPdf
|
||||
from pydub import AudioSegment
|
||||
|
||||
class FileHandlers:
|
||||
# Функция сохранения файла
|
||||
def saveFile(self, filename, content, output_path, format='mp3'):
|
||||
'''
|
||||
Сохраняет текст, pdf из markdown_pdf или склеенный аудиофайл в файл с указанным названием и директорией.
|
||||
|
||||
Args:
|
||||
:param filename: название файла;
|
||||
:param content: содержание файла;
|
||||
:param output_path: выходная диретория файла.
|
||||
'''
|
||||
# Создание объекта директории
|
||||
directory = Path(output_path)
|
||||
filePath = directory / filename # Добавление пути директории
|
||||
filePath.parent.mkdir(parents=True, exist_ok=True) # Создание директории если не существует
|
||||
|
||||
# Сохранение для разных типов
|
||||
if type(content) == MarkdownPdf:
|
||||
return content.save(filePath)
|
||||
|
||||
elif type(content) == str:
|
||||
return filePath.write_text(content, encoding='utf-8')
|
||||
|
||||
elif type(content) == AudioSegment:
|
||||
return content.export(filePath, format=format)
|
||||
11
handlers/glueAudio.py
Normal file
11
handlers/glueAudio.py
Normal file
@@ -0,0 +1,11 @@
|
||||
from pydub import AudioSegment
|
||||
|
||||
class GlueAudio():
|
||||
def glue(self, audioFiles):
|
||||
glued = AudioSegment.empty()
|
||||
|
||||
for audioFile in audioFiles:
|
||||
audio = AudioSegment.from_file(audioFile)
|
||||
glued += audio
|
||||
|
||||
return glued
|
||||
67
handlers/gradioHandler.py
Normal file
67
handlers/gradioHandler.py
Normal file
@@ -0,0 +1,67 @@
|
||||
from config import LLM_MODELS # Импортируем словарь моделей
|
||||
import gradio as gr
|
||||
|
||||
|
||||
class GradioHandlers:
|
||||
def __init__(self, llm_factory, ConvertMdToPdf, FileHandlers, FasterWhisper, GlueAudio):
|
||||
# Объект для работы с файлами
|
||||
self.fh = FileHandlers()
|
||||
self.ga = GlueAudio()
|
||||
self.ConvertMdToPdf = ConvertMdToPdf()
|
||||
self.FasterWhisper = FasterWhisper()
|
||||
self.llm_factory = llm_factory # Сохраняем фабрику
|
||||
|
||||
def handleRecognizeBtn(self, audioFiles, model, device, compute_type, beamSize, vadFilter, minSilenceDurationMs, speechPadMs, temp0, temp1, temp2, wordTimestamps, noSpeechThreshold, conditionOnPreviousText, filename, outPath):
|
||||
audioFile = self.ga.glue(audioFiles)
|
||||
file = self.fh.saveFile(filename, audioFile, outPath)
|
||||
|
||||
return self.FasterWhisper.recognize(model, device, compute_type, file, beamSize, vadFilter, minSilenceDurationMs, speechPadMs, temp0, temp1, temp2, wordTimestamps, noSpeechThreshold, conditionOnPreviousText)
|
||||
|
||||
# Функция улучшения текста
|
||||
def generateByCondition(self, api_key, llm_provider, llm_model, system_prompt, recognized_text, llm_temperature, is_pipeline_enabled, trigger, isSaveFile, filename, filenamePdf, output_path):
|
||||
try:
|
||||
# Получаем нужный провайдер через фабрику
|
||||
provider = self.llm_factory(llm_provider, api_key)
|
||||
except ValueError as e:
|
||||
# Если API ключ не предоставлен для нужного провайдера, выводим ошибку
|
||||
self.gr.Warning(str(e))
|
||||
return self.gr.skip(), self.gr.skip()
|
||||
|
||||
def process():
|
||||
result, md = provider.generate(llm_model, system_prompt, recognized_text, llm_temperature)
|
||||
pdf, unicodeText = self.ConvertMdToPdf.convertLatexToText(md)
|
||||
if isSaveFile:
|
||||
self.fh.saveFile(filenamePdf, pdf, output_path)
|
||||
self.fh.saveFile(filename, result, output_path)
|
||||
return result, unicodeText
|
||||
|
||||
if (is_pipeline_enabled and trigger == "change") or (not is_pipeline_enabled and trigger == "click"):
|
||||
return process()
|
||||
|
||||
return gr.skip(), gr.skip()
|
||||
|
||||
# НОВАЯ ФУНКЦИЯ для обновления списка моделей
|
||||
def update_model_dropdown(self, provider):
|
||||
"""
|
||||
Вызывается при изменении llmProvider.
|
||||
Возвращает обновленный компонент Dropdown для моделей.
|
||||
"""
|
||||
# Получаем список моделей для выбранного провайдера
|
||||
models = LLM_MODELS.get(provider, [])
|
||||
|
||||
# Выбираем первое значение по умолчанию, если список не пуст
|
||||
default_value = models[0] if models else None
|
||||
|
||||
# Возвращаем обновленный компонент. Используем 'gr' напрямую.
|
||||
return gr.update(choices=models, value=default_value)
|
||||
|
||||
# Функция для динамического обновления кнопки
|
||||
def updateButton(self, isChecked):
|
||||
if not isChecked:
|
||||
variant = 'primary'
|
||||
else:
|
||||
variant = 'secondary'
|
||||
return gr.update(interactive=not isChecked, variant=variant)
|
||||
|
||||
def updateTextbox(self, isChecked):
|
||||
return gr.update(visible=isChecked)
|
||||
204
requirements.txt
204
requirements.txt
@@ -1,102 +1,102 @@
|
||||
aiofiles==24.1.0
|
||||
annotated-types==0.7.0
|
||||
anyio==4.10.0
|
||||
av==15.1.0
|
||||
beautifulsoup4==4.13.5
|
||||
Brotli==1.1.0
|
||||
bs4==0.0.2
|
||||
certifi==2025.8.3
|
||||
cffi==2.0.0
|
||||
charset-normalizer==3.4.3
|
||||
click==8.2.1
|
||||
coloredlogs==15.0.1
|
||||
colour==0.1.5
|
||||
cssselect2==0.8.0
|
||||
ctranslate2==4.6.0
|
||||
distro==1.9.0
|
||||
dotenv==0.9.9
|
||||
exceptiongroup==1.3.0
|
||||
fastapi==0.116.1
|
||||
faster-whisper==1.2.0
|
||||
ffmpeg-python==0.2.0
|
||||
ffmpy==0.6.1
|
||||
filelock==3.19.1
|
||||
flatbuffers==25.2.10
|
||||
flatlatex==0.15
|
||||
fonttools==4.59.2
|
||||
fsspec==2025.9.0
|
||||
future==1.0.0
|
||||
gradio==5.44.1
|
||||
gradio_client==1.12.1
|
||||
groovy==0.1.2
|
||||
h11==0.16.0
|
||||
hf-xet==1.1.9
|
||||
httpcore==1.0.9
|
||||
httpx==0.28.1
|
||||
huggingface-hub==0.34.4
|
||||
humanfriendly==10.0
|
||||
idna==3.10
|
||||
iso639-lang==2.6.3
|
||||
Jinja2==3.1.6
|
||||
jiter==0.10.0
|
||||
joblib==1.5.2
|
||||
langdetect==1.0.9
|
||||
littleutils==0.2.4
|
||||
markdown-it-py==3.0.0
|
||||
markdown_pdf==1.9
|
||||
MarkupSafe==3.0.2
|
||||
mdurl==0.1.2
|
||||
mpmath==1.3.0
|
||||
nltk==3.9.1
|
||||
numpy==2.2.6
|
||||
onnxruntime==1.22.1
|
||||
openai==1.106.1
|
||||
orjson==3.11.3
|
||||
outdated==0.2.2
|
||||
packaging==25.0
|
||||
pandas==2.3.2
|
||||
pillow==11.3.0
|
||||
protobuf==6.32.0
|
||||
pycparser==2.22
|
||||
pydantic==2.11.7
|
||||
pydantic_core==2.33.2
|
||||
pydub==0.25.1
|
||||
pydyf==0.11.0
|
||||
Pygments==2.19.2
|
||||
pylatexenc==2.10
|
||||
pymultidictionary==1.3.2
|
||||
PyMuPDF==1.26.4
|
||||
pyperclip==1.9.0
|
||||
pyphen==0.17.2
|
||||
python-dateutil==2.9.0.post0
|
||||
python-dotenv==1.1.1
|
||||
python-multipart==0.0.20
|
||||
pytz==2025.2
|
||||
PyYAML==6.0.2
|
||||
regex==2025.9.1
|
||||
requests==2.32.5
|
||||
rich==14.1.0
|
||||
ruff==0.12.12
|
||||
safehttpx==0.1.6
|
||||
semantic-version==2.10.0
|
||||
shellingham==1.5.4
|
||||
six==1.17.0
|
||||
sniffio==1.3.1
|
||||
soupsieve==2.8
|
||||
starlette==0.47.3
|
||||
sympy==1.14.0
|
||||
tinycss2==1.4.0
|
||||
tinyhtml5==2.0.0
|
||||
tkmacosx==1.0.5
|
||||
tokenizers==0.22.0
|
||||
tomlkit==0.13.3
|
||||
tqdm==4.67.1
|
||||
typer==0.17.4
|
||||
typing-inspection==0.4.1
|
||||
typing_extensions==4.15.0
|
||||
tzdata==2025.2
|
||||
urllib3==2.5.0
|
||||
uvicorn==0.35.0
|
||||
webencodings==0.5.1
|
||||
websockets==15.0.1
|
||||
zopfli==0.2.3.post1
|
||||
aiofiles==24.1.0
|
||||
annotated-types==0.7.0
|
||||
anyio==4.10.0
|
||||
av==15.1.0
|
||||
beautifulsoup4==4.13.5
|
||||
Brotli==1.1.0
|
||||
bs4==0.0.2
|
||||
certifi==2025.8.3
|
||||
cffi==2.0.0
|
||||
charset-normalizer==3.4.3
|
||||
click==8.2.1
|
||||
coloredlogs==15.0.1
|
||||
colour==0.1.5
|
||||
cssselect2==0.8.0
|
||||
ctranslate2==4.6.0
|
||||
distro==1.9.0
|
||||
dotenv==0.9.9
|
||||
exceptiongroup==1.3.0
|
||||
fastapi==0.116.1
|
||||
faster-whisper==1.2.0
|
||||
ffmpeg-python==0.2.0
|
||||
ffmpy==0.6.1
|
||||
filelock==3.19.1
|
||||
flatbuffers==25.2.10
|
||||
flatlatex==0.15
|
||||
fonttools==4.59.2
|
||||
fsspec==2025.9.0
|
||||
future==1.0.0
|
||||
gradio==5.44.1
|
||||
gradio_client==1.12.1
|
||||
groovy==0.1.2
|
||||
h11==0.16.0
|
||||
hf-xet==1.1.9
|
||||
httpcore==1.0.9
|
||||
httpx==0.28.1
|
||||
huggingface-hub==0.34.4
|
||||
humanfriendly==10.0
|
||||
idna==3.10
|
||||
iso639-lang==2.6.3
|
||||
Jinja2==3.1.6
|
||||
jiter==0.10.0
|
||||
joblib==1.5.2
|
||||
langdetect==1.0.9
|
||||
littleutils==0.2.4
|
||||
markdown-it-py==3.0.0
|
||||
markdown_pdf==1.9
|
||||
MarkupSafe==3.0.2
|
||||
mdurl==0.1.2
|
||||
mpmath==1.3.0
|
||||
nltk==3.9.1
|
||||
numpy==2.2.6
|
||||
onnxruntime==1.22.1
|
||||
openai==1.106.1
|
||||
orjson==3.11.3
|
||||
outdated==0.2.2
|
||||
packaging==25.0
|
||||
pandas==2.3.2
|
||||
pillow==11.3.0
|
||||
protobuf==6.32.0
|
||||
pycparser==2.22
|
||||
pydantic==2.11.7
|
||||
pydantic_core==2.33.2
|
||||
pydub==0.25.1
|
||||
pydyf==0.11.0
|
||||
Pygments==2.19.2
|
||||
pylatexenc==2.10
|
||||
pymultidictionary==1.3.2
|
||||
PyMuPDF==1.26.4
|
||||
pyperclip==1.9.0
|
||||
pyphen==0.17.2
|
||||
python-dateutil==2.9.0.post0
|
||||
python-dotenv==1.1.1
|
||||
python-multipart==0.0.20
|
||||
pytz==2025.2
|
||||
PyYAML==6.0.2
|
||||
regex==2025.9.1
|
||||
requests==2.32.5
|
||||
rich==14.1.0
|
||||
ruff==0.12.12
|
||||
safehttpx==0.1.6
|
||||
semantic-version==2.10.0
|
||||
shellingham==1.5.4
|
||||
six==1.17.0
|
||||
sniffio==1.3.1
|
||||
soupsieve==2.8
|
||||
starlette==0.47.3
|
||||
sympy==1.14.0
|
||||
tinycss2==1.4.0
|
||||
tinyhtml5==2.0.0
|
||||
tkmacosx==1.0.5
|
||||
tokenizers==0.22.0
|
||||
tomlkit==0.13.3
|
||||
tqdm==4.67.1
|
||||
typer==0.17.4
|
||||
typing-inspection==0.4.1
|
||||
typing_extensions==4.15.0
|
||||
tzdata==2025.2
|
||||
urllib3==2.5.0
|
||||
uvicorn==0.35.0
|
||||
webencodings==0.5.1
|
||||
websockets==15.0.1
|
||||
zopfli==0.2.3.post1
|
||||
|
||||
@@ -1,34 +1,34 @@
|
||||
from faster_whisper import WhisperModel
|
||||
|
||||
class FasterWhisper:
|
||||
def recognize(self, model, device, compute_type, audioFile, beamSize, vadFilter, minSilenceDurationMs, speechPadMs, temp0, temp1, temp2, wordTimestamps, noSpeechThreshold, conditionOnPreviousText):
|
||||
model = WhisperModel(model, device=device, compute_type=compute_type) # Задаем модель
|
||||
|
||||
segments, _ = model.transcribe( # Распознаем текст
|
||||
audioFile,
|
||||
beam_size=beamSize,
|
||||
vad_filter=vadFilter,
|
||||
vad_parameters={
|
||||
"min_silence_duration_ms": minSilenceDurationMs,
|
||||
"speech_pad_ms": speechPadMs
|
||||
},
|
||||
temperature= [temp0, temp1, temp2],
|
||||
word_timestamps=wordTimestamps,
|
||||
no_speech_threshold=noSpeechThreshold,
|
||||
condition_on_previous_text=conditionOnPreviousText
|
||||
)
|
||||
|
||||
text = ''
|
||||
|
||||
for seg in segments:
|
||||
text += f"[{self.format_timestamp(seg.start)} -> {self.format_timestamp(seg.end)}] {seg.text}" + '\n'
|
||||
|
||||
return(text)
|
||||
|
||||
def format_timestamp(self, seconds: float) -> str:
|
||||
millis = int(seconds * 1000)
|
||||
hours = millis // (3600 * 1000)
|
||||
minutes = (millis % (3600 * 1000)) // (60 * 1000)
|
||||
seconds_int = (millis % (60 * 1000)) // 1000
|
||||
millis = millis % 1000
|
||||
return f"{hours:02d}:{minutes:02d}:{seconds_int:02d},{millis:03d}"
|
||||
from faster_whisper import WhisperModel
|
||||
|
||||
class FasterWhisper:
|
||||
def recognize(self, model, device, compute_type, audioFile, beamSize, vadFilter, minSilenceDurationMs, speechPadMs, temp0, temp1, temp2, wordTimestamps, noSpeechThreshold, conditionOnPreviousText):
|
||||
model = WhisperModel(model, device=device, compute_type=compute_type) # Задаем модель
|
||||
|
||||
segments, _ = model.transcribe( # Распознаем текст
|
||||
audioFile,
|
||||
beam_size=beamSize,
|
||||
vad_filter=vadFilter,
|
||||
vad_parameters={
|
||||
"min_silence_duration_ms": minSilenceDurationMs,
|
||||
"speech_pad_ms": speechPadMs
|
||||
},
|
||||
temperature= [temp0, temp1, temp2],
|
||||
word_timestamps=wordTimestamps,
|
||||
no_speech_threshold=noSpeechThreshold,
|
||||
condition_on_previous_text=conditionOnPreviousText
|
||||
)
|
||||
|
||||
text = ''
|
||||
|
||||
for seg in segments:
|
||||
text += f"[{self.format_timestamp(seg.start)} -> {self.format_timestamp(seg.end)}] {seg.text}" + '\n'
|
||||
|
||||
return text
|
||||
|
||||
def format_timestamp(self, seconds: float) -> str:
|
||||
millis = int(seconds * 1000)
|
||||
hours = millis // (3600 * 1000)
|
||||
minutes = (millis % (3600 * 1000)) // (60 * 1000)
|
||||
seconds_int = (millis % (60 * 1000)) // 1000
|
||||
millis = millis % 1000
|
||||
return f"{hours:02d}:{minutes:02d}:{seconds_int:02d},{millis:03d}"
|
||||
|
||||
@@ -1,31 +0,0 @@
|
||||
import openai
|
||||
|
||||
class Llm:
|
||||
def __init__(self, apiKey:str):
|
||||
self.client = openai.OpenAI(
|
||||
api_key=apiKey,
|
||||
base_url='https://api.intelligence.io.solutions/api/v1/'
|
||||
)
|
||||
|
||||
def generate(self, model:str, systemPrompt:str, userPrompt:str, temp:float):
|
||||
'''
|
||||
prompt[user_promtp, system_ptompt]
|
||||
Function generate text by prompt
|
||||
'''
|
||||
|
||||
# Получаем ответ от нейросети
|
||||
response = self.client.chat.completions.create(
|
||||
model=model,
|
||||
messages=[
|
||||
{'role': 'system', 'content': systemPrompt},
|
||||
{'role': 'user', 'content': userPrompt},
|
||||
],
|
||||
temperature=temp,
|
||||
stream=False
|
||||
)
|
||||
|
||||
# Достаем текст
|
||||
text = str(response.choices[0].message.content)
|
||||
|
||||
return text, text
|
||||
|
||||
22
services/llm_factory.py
Normal file
22
services/llm_factory.py
Normal file
@@ -0,0 +1,22 @@
|
||||
# services/llm_factory.py
|
||||
from services.llm_providers.ionet_provider import IoNetProvider
|
||||
from services.llm_providers.gemini_provider import GeminiProvider
|
||||
from services.llm_providers.gpt4free_provider import Gpt4FreeProvider
|
||||
from services.llm_providers.base_provider import BaseLLMProvider
|
||||
|
||||
def get_llm_provider(provider_name: str, api_key: str | None) -> BaseLLMProvider:
|
||||
"""
|
||||
Фабричная функция для получения экземпляра провайдера LLM.
|
||||
"""
|
||||
if provider_name == 'io.net':
|
||||
if not api_key:
|
||||
raise ValueError("API ключ обязателен для io.net")
|
||||
return IoNetProvider(api_key)
|
||||
elif provider_name == 'Gemini':
|
||||
if not api_key:
|
||||
raise ValueError("API ключ обязателен для Gemini")
|
||||
return GeminiProvider(api_key)
|
||||
elif provider_name == 'gpt4free':
|
||||
return Gpt4FreeProvider()
|
||||
else:
|
||||
raise ValueError(f"Неизвестный провайдер: {provider_name}")
|
||||
18
services/llm_providers/base_provider.py
Normal file
18
services/llm_providers/base_provider.py
Normal file
@@ -0,0 +1,18 @@
|
||||
from abc import ABC, abstractmethod
|
||||
|
||||
class BaseLLMProvider(ABC):
|
||||
"""
|
||||
Абстрактный базовый класс для всех провайдеров LLM.
|
||||
Каждый провайдер должен реализовать метод generate.
|
||||
"""
|
||||
def __init__(self, api_key: str | None = None):
|
||||
self.api_key = api_key
|
||||
|
||||
@abstractmethod
|
||||
def generate(self, model: str, system_prompt: str, user_prompt: str, temp: float):
|
||||
"""
|
||||
Основной метод для генерации текста.
|
||||
|
||||
Должен возвращать кортеж из двух строк: (чистый_текст, markdown_текст)
|
||||
"""
|
||||
pass
|
||||
74
services/llm_providers/gemini_provider.py
Normal file
74
services/llm_providers/gemini_provider.py
Normal file
@@ -0,0 +1,74 @@
|
||||
# services/llm_providers/gemini_provider.py
|
||||
|
||||
import requests
|
||||
from .base_provider import BaseLLMProvider
|
||||
|
||||
class GeminiProvider(BaseLLMProvider):
|
||||
"""
|
||||
Провайдер для Google Gemini, использующий прямые REST API вызовы
|
||||
через библиотеку requests для надежной работы с SOCKS-прокси.
|
||||
"""
|
||||
def __init__(self, api_key: str):
|
||||
super().__init__(api_key)
|
||||
self.base_url = "https://generativelanguage.googleapis.com/v1beta/models/"
|
||||
|
||||
def generate(self, model: str, system_prompt: str, user_prompt: str, temp: float):
|
||||
"""
|
||||
Генерирует текст с помощью модели Gemini, отправляя запрос через прокси.
|
||||
"""
|
||||
# 1. Формируем URL для запроса
|
||||
api_url = f"{self.base_url}{model}:generateContent?key={self.api_key}"
|
||||
|
||||
# 2. Задаем настройки прокси из вашего примера
|
||||
# socks5h:// означает, что DNS-запросы также будут идти через прокси
|
||||
proxies = {
|
||||
'http': 'socks5://192.168.1.6:2080',
|
||||
'https': 'socks5h://192.168.1.6:2080'
|
||||
}
|
||||
|
||||
# 3. Собираем тело запроса (payload) в формате, который ожидает Gemini API
|
||||
data = {
|
||||
"system_instruction": {
|
||||
"parts": {"text": system_prompt}
|
||||
},
|
||||
"contents": [{
|
||||
"parts": [{"text": user_prompt}]
|
||||
}],
|
||||
"generationConfig": {
|
||||
"temperature": temp
|
||||
}
|
||||
}
|
||||
|
||||
try:
|
||||
# 4. Отправляем POST-запрос с данными и настройками прокси
|
||||
response = requests.post(api_url, json=data, proxies=proxies, timeout=90)
|
||||
|
||||
# Проверяем, не вернул ли сервер ошибку (например, 4xx или 5xx)
|
||||
response.raise_for_status()
|
||||
|
||||
# 5. Парсим JSON-ответ и извлекаем сгенерированный текст
|
||||
response_json = response.json()
|
||||
|
||||
# Добавим проверку на случай, если контент был заблокирован
|
||||
if "candidates" not in response_json or not response_json["candidates"]:
|
||||
block_reason = response_json.get("promptFeedback", {}).get("blockReason", "неизвестная причина")
|
||||
error_message = f"Контент заблокирован. Причина: {block_reason}"
|
||||
return error_message, error_message
|
||||
|
||||
text = response_json["candidates"][0]["content"]["parts"][0]["text"]
|
||||
return text, text
|
||||
|
||||
except requests.exceptions.ProxyError as e:
|
||||
error_message = f"Ошибка подключения к прокси. Убедитесь, что Nekobox запущен и слушает порт 2080. Ошибка: {e}"
|
||||
print(error_message)
|
||||
return error_message, error_message
|
||||
except requests.exceptions.RequestException as e:
|
||||
# Ловим все остальные ошибки requests (таймаут, проблемы с сетью и т.д.)
|
||||
error_message = f"Произошла ошибка при обращении к API Gemini: {e}"
|
||||
print(error_message)
|
||||
return error_message, error_message
|
||||
except (KeyError, IndexError) as e:
|
||||
# Ловим ошибки, если структура JSON-ответа неожиданная
|
||||
error_message = f"Не удалось разобрать ответ от API Gemini. Структура ответа изменилась. Ошибка: {e}"
|
||||
print(error_message)
|
||||
return error_message, error_message
|
||||
28
services/llm_providers/gpt4free_provider.py
Normal file
28
services/llm_providers/gpt4free_provider.py
Normal file
@@ -0,0 +1,28 @@
|
||||
# services/llm_providers/gpt4free_provider.py
|
||||
from g4f.client import Client
|
||||
from .base_provider import BaseLLMProvider
|
||||
|
||||
class Gpt4FreeProvider(BaseLLMProvider):
|
||||
# gpt4free не требует API ключа
|
||||
def __init__(self, api_key: str | None = None):
|
||||
super().__init__(api_key)
|
||||
self.client = Client()
|
||||
|
||||
|
||||
def generate(self, model: str, system_prompt: str, user_prompt: str, temp: float):
|
||||
# temp в g4f может работать не для всех внутренних провайдеров
|
||||
try:
|
||||
response = self.client.chat.completions.create(
|
||||
model=model, # Пример модели, может варьироваться в зависимости от доступности провайдеров
|
||||
messages=[
|
||||
{"role": "system", "content": system_prompt},
|
||||
{"role": "user", "content": user_prompt}
|
||||
],
|
||||
temperature=temp
|
||||
)
|
||||
text = response.choices[0].message.content
|
||||
return text, text
|
||||
except Exception as e:
|
||||
error_message = f"Ошибка при работе с gpt4free: {e}"
|
||||
print(error_message)
|
||||
return error_message, error_message
|
||||
23
services/llm_providers/ionet_provider.py
Normal file
23
services/llm_providers/ionet_provider.py
Normal file
@@ -0,0 +1,23 @@
|
||||
import openai
|
||||
from .base_provider import BaseLLMProvider
|
||||
|
||||
class IoNetProvider(BaseLLMProvider):
|
||||
def __init__(self, api_key: str):
|
||||
super().__init__(api_key)
|
||||
self.client = openai.OpenAI(
|
||||
api_key=self.api_key,
|
||||
base_url='https://api.intelligence.io.solutions/api/v1/'
|
||||
)
|
||||
|
||||
def generate(self, model: str, system_prompt: str, user_prompt: str, temp: float):
|
||||
response = self.client.chat.completions.create(
|
||||
model=model,
|
||||
messages=[
|
||||
{'role': 'system', 'content': system_prompt},
|
||||
{'role': 'user', 'content': user_prompt},
|
||||
],
|
||||
temperature=temp,
|
||||
stream=False
|
||||
)
|
||||
text = str(response.choices[0].message.content)
|
||||
return text, text # Возвращаем как чистый текст, так и Markdown
|
||||
Reference in New Issue
Block a user