@@ -17,6 +17,7 @@ from ml_model.services.midjourney import Midjourney
 from ml_model.services.mistral import Mistral
 from ml_model.services.musicgen import Musicgen
 from ml_model.services.pulid import Pulid
+from ml_model.services.qwen import Qwen
 from ml_model.services.recraft import Recraft
 from ml_model.services.sdxlemoji import Sdxlemoji
 from ml_model.services.stablediffusion import Stablediffusion
@@ -0,0 +1,113 @@
+import time
+from datetime import timedelta
+from decimal import Decimal
+from typing import Iterator, Any
+
+from messages.models import Message
+from ml_model.models import ModelCategory, ModelVersion, ModelInput, ModelParameter
+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
+
+
+class Qwen(SimpleService):
+    """
+    Qwen Service
+    contains abstract method make, which makes a generation
+    """
+
+    title = 'Qwen'
+    description = 'Нейросеть, способная генерировать еще больше текста из вашего текста'
+    category = ModelCategory(title='Чат-боты', slug='chat-bots')
+    versions = [
+        ModelVersion(name='QwQ 32B', default=True, slug='qwq-32b'),
+        # ModelVersion(name='QwQ 32B', slug='qwq-32b:free'),
+    ]
+    inputs = [ModelInput(type=ModelInput.TypeChoices.TEXT, required=True), ]
+    parameters = []
+    TOKENS_COST = {
+        'qwq-32b': {'input': Decimal('24'), 'output': Decimal('36')},  # 1M tokens
+        'qwq-32b:free': {'input': Decimal('0'), 'output': Decimal('0')},  # 1M tokens
+    }
+
+    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
+        )
+        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()
+        version = f'qwen/{input_message.info.pop('version', 'qwq-32b')}'
+        callback_data = {
+            'provider': {
+                'order': ['DeepInfra']
+            },
+            **input_message.info
+        }
+        messages = self.get_chat_history()
+        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
@@ -6,6 +6,7 @@ from io import BytesIO
 from typing import IO, Any, Dict
 
 import deepl
+import httpx
 import replicate
 import requests
 from celery import shared_task
@@ -112,6 +113,37 @@ def replicate_run(callback_url: str, payload: dict[str, Any]):
     )
 
 
+@shared_task
+def openrouter_run(version: str, messages: list, callback_data: dict, model_name: str):
+    with httpx.Client(
+            base_url='https://openrouter.ai/api/v1',
+            headers={'Authorization': f'Bearer {settings.OPENROUTER_API_KEY}'},
+    ) 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]:
     content = requests.post(
@@ -3,6 +3,7 @@ SECRET_KEY=testtest
 DEBUG=true
 
 # NEURON MODELS
+OPENROUTER_API_KEY=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
 OPENAI_API_KEY=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
 STABLE_DIFFUSION_API_KEY=sk-fztQxZobaL0SD7PgpmK7XMQlyNivpKFZNJnqAVG2CcbvAP6Z
 REPLICATE_API_KEY=r8_HBk6Ts5UJU60nDOUl1V6Uej4ihAAxUc3HAZLO