@@ -1,16 +1,19 @@ import base64 import time +import filetype + from datetime import timedelta from decimal import Decimal from io import BytesIO -from typing import Any, Iterator - -import filetype from PIL import Image +from django.db.models.fields.files import FieldFile + from messages.models import Message + 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 @@ -30,19 +33,18 @@ class Gemini(SimpleService): }, } - def calculate_price( - self, version: str, input_tokens: int, output_tokens: int - ) -> Decimal: + def calculate_price(self, version: str, input_tokens: int, output_tokens: int, image: FieldFile) -> 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 + input_tokens * price_map['input'] / 1_000_000 + + output_tokens * price_map['output'] / 1_000_000 ) + if image: + price += 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 + self, content: str, t: timedelta, save: bool = True ) -> list[Message]: msgs = [ Message( @@ -56,7 +58,6 @@ class Gemini(SimpleService): 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 = { @@ -77,9 +78,18 @@ class Gemini(SimpleService): ) buf.close() messages[-1]['content'] = [ - {'type': 'text', 'text': input_message.content}, - {'type': 'image_url', 'image_url': {'url': image_url}}, + { + 'type': 'text', + 'text': input_message.content + }, + { + 'type': 'image_url', + 'image_url': { + 'url': image_url + } + } ] + start_time = time.time() result = openrouter_run(version, messages, callback_data, self.title) process_time = timedelta(seconds=(time.time() - start_time)) self.handle_invoice( @@ -87,6 +97,7 @@ class Gemini(SimpleService): version=version, input_tokens=result[1], output_tokens=result[2], + image=image ) msgs = self.save_results(result[0], process_time) return msgs @@ -121,6 +132,5 @@ class Gemini(SimpleService): 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) + character_length -= len(memory.pop(0)) return memory