@@ -8,6 +8,7 @@ from ml_model.services.djourney import Djourney from ml_model.services.epicphotogasm import Epicphotogasm from ml_model.services.flux import Flux from ml_model.services.granite import Granite +from ml_model.services.gemini import Gemini from ml_model.services.iconic import Iconic from ml_model.services.kandinsky import Kandinsky from ml_model.services.lightning import Lightning @@ -0,0 +1,153 @@ +import base64 +import time +import filetype + +from decimal import Decimal +from datetime import timedelta +from io import BytesIO +from typing import Iterator, Any +from PIL import Image + +from ml_model.models import ModelCategory, ModelVersion, ModelInput +from ml_model.services.base import SimpleService +from ml_model.tasks import openrouter_run + +from tools.chats.models import Chat +from tools.copywrite.models import Copywrite +from tools.public_api.models import APIStore + +from messages.models import Message + + +class Gemini(SimpleService): + """ + Gemini Service + contains abstract method make, which makes a generation + """ + + title = 'Gemini' + description = 'Нейросеть, способная генерировать еще больше текста из вашего текста' + category = ModelCategory(title='Чат-боты', slug='chat-bots') + + versions = [ + ModelVersion(name='Gemini', default=True, slug='google/gemini-2.0-flash-001'), + ] + inputs = [ + ModelInput(type=ModelInput.TypeChoices.TEXT, required=True), + ModelInput(type=ModelInput.TypeChoices.IMAGE) + ] + + parameters = [] + + TOKENS_COST = { + 'gemini-2.0-flash-001': { + 'input': Decimal('20'), + 'output': Decimal('80'), + 'input_imgs': Decimal('5.16'), + }, + } + + def calculate_price( + self, version: str, input_tokens: int, output_tokens: int + ) -> Decimal: + price_map = self.TOKENS_COST[version.split('/')[1]] + price = ( + input_tokens * price_map['input'] / 1_000_000 + + output_tokens * price_map['output'] / 1_000_000 + + price_map['input_imgs'] / 1_000 + ) + return price.quantize(Decimal('0.1'), rounding='ROUND_UP') + + def save_results( + self, content: Iterator[Any], t: timedelta, save: bool = True + ) -> list[Message]: + msgs = [ + Message( + content=content, + content_object=self.store, + elapsed_time=t, + ) + ] + if save: + return Message.objects.bulk_create(msgs) + return msgs + + def make(self, input_message: Message, save: bool = True) -> list[Message]: + start_time = time.time() + info = input_message.info.copy() + version = info.pop('version') + callback_data = { + 'provider': { + 'order': ['Google AI Studio'] + }, + **input_message.info + } + messages = self.get_chat_history() + image = input_message.file + if image: + kind = filetype.guess(image.read(20)) + mime = kind.mime if kind else 'application/octet-stream' + normalized_image = Image.open(image) + format = 'jpeg' if kind.extension == 'jpg' else kind.extension + buf = BytesIO() + normalized_image.save(buf, format=format) + image_url = ( + f'data:{mime};base64,{base64.b64encode(buf.getvalue()).decode("utf-8")}' + ) + buf.close() + messages[-1]['content'] = [ + { + 'type': 'text', + 'text': input_message.content + }, + { + 'type': 'image_url', + 'image_url': { + 'url': image_url + } + } + ] + result = openrouter_run(version, messages, callback_data, self.title) + process_time = timedelta(seconds=(time.time() - start_time)) + self.handle_invoice( + input_message.content_object.model, + version=version, + input_tokens=result[1], + output_tokens=result[2] + ) + msgs = self.save_results(result[0], process_time) + return msgs + + def get_chat_history(self, message_limit: int = 10, max_character_limit: int = 1500): + if isinstance(self.store, Chat): + air_messages = list( + reversed( + Message.objects.filter( + chats_chats_messages=self.store, is_deleted=False, is_sent=True + ).order_by('-created_at')[:message_limit] + ) + ) + elif isinstance(self.store, APIStore): + air_messages = [] + elif isinstance(self.store, Copywrite): + air_messages = list( + reversed( + Message.objects.filter( + copywrite_copywrites_messages=self.store, + is_deleted=False, + is_sent=True, + ).order_by('-created_at')[:message_limit] + ) + ) + memory = [] + for msg in air_messages: + content = msg.content or '' + if msg.from_model: + memory.append({"role": "assistant", "content": content}) + else: + memory.append({"role": "user", "content": content}) + character_length = sum(len(content['content']) for content in memory) + while character_length > max_character_limit: + memory.pop(0) + character_length = sum(len(content['content']) for content in memory) + return memory @@ -1,5 +1,6 @@ import base64 import json +import httpx # import uuid from io import BytesIO @@ -111,6 +112,37 @@ def replicate_run(callback_url: str, payload: dict[str, Any]): input=payload, ) +@shared_task +def openrouter_run(version: str, messages: list, callback_data: dict, model_name: str): + for proxy in Proxy.objects.all(): + with httpx.Client( + base_url='https://openrouter.ai/api/v1', + headers={'Authorization': f'Bearer {settings.OPENROUTER_API_KEY}'}, + proxy=f'{proxy.protocol}://{proxy.address}', + ) as client: + resp = client.post( + 'chat/completions', + json={ + 'model': version, + 'messages': messages, + **callback_data + }, + ) + if ( + (data := resp.json()) + and data.get('choices') + and ( + content := ','.join( + [choice['message']['content'] for choice in data.get('choices')] + ) + ) + ): + return ( + content, + data['usage']['prompt_tokens'], + data['usage']['completion_tokens'] + ) + raise Exception(f'No answer from {model_name}, please retry later') @shared_task def upscale_run(payload: dict[str, tuple[str, IO]]) -> list[str]: @@ -13,6 +13,7 @@ MISTRAL_API_KEY=CYtZSCQXZFzHcpJvWOjWNx4EHjf5kWQc DEEPL_API_KEY=4bb58b98-ca95-5978-9be0-ed437df6c15c:fx SERPER_API_KEY=ed8e0dbcc26dacf3f7f99fbc8b3add9ada0c793e FLUX_API_KEY=dccaf377-aecf-4cf0-aff4-dde47cee340d +OPENROUTER_API_KEY=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 # EXTERNAL SERVICES OPENAI_PROXY_HOST=neuron-proxy:8080