Merge pull request #43 from OVINC-CN/feat_hunyuan

feat: support to use tencent cloud api for hunyuan #40

apps/chat/client.py

@@ -1,14 +1,9 @@
 import abc
-import base64
-import datetime
-import hashlib
-import hmac
 import json
 from typing import List, Union
 
 import google.generativeai as genai
 import qianfan
-import requests
 import tiktoken
 from django.conf import settings
 from django.contrib.auth import get_user_model
@@ -19,16 +14,12 @@
 from openai.types.chat.chat_completion_chunk import ChatCompletionChunk
 from openai.types.images_response import ImagesResponse
 from qianfan import QfMessages, QfResponse
-from requests import Response
 from rest_framework.request import Request
+from tencentcloud.common import credential as tencent_cloud_credential
+from tencentcloud.hunyuan.v20230901 import hunyuan_client
+from tencentcloud.hunyuan.v20230901 import models as hunyuan_models
 
-from apps.chat.constants import (
-    AI_API_REQUEST_TIMEOUT,
-    HUNYUAN_DATA_PATTERN,
-    GeminiRole,
-    OpenAIRole,
-)
-from apps.chat.exceptions import UnexpectedError
+from apps.chat.constants import GeminiRole, OpenAIRole
 from apps.chat.models import AIModel, ChatLog, HunYuanChuck, Message
 
 USER_MODEL = get_user_model()
@@ -185,22 +176,11 @@ def chat(self, *args, **kwargs) -> any:
         # call hunyuan api
         response = self.call_api()
         # explain completion
-        completion_text = bytes()
         for chunk in response:
-            completion_text += chunk
-            # match hunyuan data content
-            match = HUNYUAN_DATA_PATTERN.search(completion_text)
-            if match is None:
-                continue
-            # load content
-            resp_text = completion_text[match.regs[0][0] : match.regs[0][1]]
-            completion_text = completion_text[match.regs[0][1] :]
-            resp_text = json.loads(resp_text.decode()[6:])
-            chunk = HunYuanChuck.create(resp_text)
-            if chunk.error.code:
-                raise UnexpectedError(detail=chunk.error.message)
+            chunk = json.loads(chunk["data"])
+            chunk = HunYuanChuck.create(chunk)
             self.record(response=chunk)
-            yield chunk.choices[0].delta.content
+            yield chunk.Choices[0].Delta.Content
         if not self.log:
             return
         self.log.finished_at = int(timezone.now().timestamp() * 1000)
@@ -211,16 +191,16 @@ def chat(self, *args, **kwargs) -> any:
     def record(self, response: HunYuanChuck) -> None:
         # check log exist
         if self.log:
-            self.log.content += response.choices[0].delta.content
-            self.log.prompt_tokens = response.usage.prompt_tokens
-            self.log.completion_tokens = response.usage.completion_tokens
+            self.log.content += response.Choices[0].Delta.Content
+            self.log.prompt_tokens = response.Usage.PromptTokens
+            self.log.completion_tokens = response.Usage.CompletionTokens
             self.log.prompt_token_unit_price = self.model_inst.prompt_price
             self.log.completion_token_unit_price = self.model_inst.completion_price
             self.log.currency_unit = self.model_inst.currency_unit
             return
         # create log
         self.log = ChatLog.objects.create(
-            chat_id=response.id,
+            chat_id=response.Id,
             user=self.user,
             model=self.model,
             messages=self.messages,
@@ -229,35 +209,20 @@ def record(self, response: HunYuanChuck) -> None:
         )
         return self.record(response=response)
 
-    def call_api(self) -> Response:
-        data = {
-            "app_id": settings.QCLOUD_APP_ID,
-            "secret_id": settings.QCLOUD_SECRET_ID,
-            "timestamp": int(timezone.now().timestamp()),
-            "expired": int((timezone.now() + datetime.timedelta(minutes=5)).timestamp()),
-            "messages": self.messages,
-            "temperature": self.temperature,
-            "top_p": self.top_p,
-            "stream": 1,
-        }
-        message_string = ",".join(
-            [f"{{\"role\":\"{message['role']}\",\"content\":\"{message['content']}\"}}" for message in self.messages]
-        )
-        message_string = f"[{message_string}]"
-        params = {**data, "messages": message_string}
-        params = dict(sorted(params.items(), key=lambda x: x[0]))
-        url = (
-            settings.QCLOUD_HUNYUAN_API_URL.split("://", 1)[1]
-            + "?"
-            + "&".join([f"{key}={val}" for key, val in params.items()])
+    def call_api(self) -> hunyuan_models.ChatCompletionsResponse:
+        client = hunyuan_client.HunyuanClient(
+            tencent_cloud_credential.Credential(settings.QCLOUD_SECRET_ID, settings.QCLOUD_SECRET_KEY), ""
         )
-        signature = hmac.new(settings.QCLOUD_SECRET_KEY.encode(), url.encode(), hashlib.sha1).digest()
-        encoded_signature = base64.b64encode(signature).decode()
-        headers = {"Authorization": encoded_signature}
-        resp = requests.post(
-            settings.QCLOUD_HUNYUAN_API_URL, json=data, headers=headers, stream=True, timeout=AI_API_REQUEST_TIMEOUT
-        )
-        return resp
+        req = hunyuan_models.ChatCompletionsRequest()
+        params = {
+            "Model": self.model,
+            "Messages": [{"Role": message["role"], "Content": message["content"]} for message in self.messages],
+            "TopP": self.top_p,
+            "Temperature": self.temperature,
+            "Stream": True,
+        }
+        req.from_json_string(json.dumps(params))
+        return client.ChatCompletions(req)
 
 
 class GeminiClient(BaseClient):

apps/chat/constants.py

@@ -1,4 +1,3 @@
-import re
 import sys
 
 from django.utils.translation import gettext_lazy
@@ -16,8 +15,6 @@
 PRICE_DIGIT_NUMS = 20
 PRICE_DECIMAL_NUMS = 10
 
-HUNYUAN_DATA_PATTERN = re.compile(rb"data:\s\{.*\}\n\n")
-
 if "celery" in sys.argv:
     TOKEN_ENCODING = ""
 else:

apps/chat/models.py

@@ -1,3 +1,5 @@
+# pylint: disable=C0103
+
 from dataclasses import dataclass
 from typing import List
 
@@ -140,46 +142,38 @@ def authed_models(cls, user: USER_MODEL, model: str = None) -> QuerySet:
 
 @dataclass
 class HunYuanDelta:
-    content: str = ""
+    Role: str = ""
+    Content: str = ""
 
 
 @dataclass
 class HunYuanChoice:
-    finish_reason: str = ""
-    delta: HunYuanDelta = None
+    FinishReason: str = ""
+    Delta: HunYuanDelta = None
 
 
 @dataclass
 class HunYuanUsage:
-    prompt_tokens: int = 0
-    completion_tokens: int = 0
-    total_tokens: int = 0
-
-
-@dataclass
-class HunYuanError:
-    code: int = 0
-    message: str = ""
+    PromptTokens: int = 0
+    CompletionTokens: int = 0
+    TotalTokens: int = 0
 
 
 @dataclass
 class HunYuanChuck:
-    req_id: str = ""
-    note: str = ""
-    choices: List[HunYuanChoice] = None
-    created: str = ""
-    id: str = ""  # pylint: disable=C0103
-    usage: HunYuanUsage = None
-    error: HunYuanError = None
+    Note: str = ""
+    Choices: List[HunYuanChoice] = None
+    Created: str = ""
+    Id: str = ""
+    Usage: HunYuanUsage = None
 
     @classmethod
     def create(cls, data: dict) -> "HunYuanChuck":
         chuck = cls(**data)
-        chuck.usage = HunYuanUsage(**data.get("usage", {}))
-        chuck.error = HunYuanError(**data.get("error", {}))
-        chuck.choices = [
-            HunYuanChoice(finish_reason=choice.get("finish_reason", ""), delta=HunYuanDelta(**choice.get("delta", {})))
-            for choice in data.get("choices", [])
+        chuck.Usage = HunYuanUsage(**data.get("Usage", {}))
+        chuck.Choices = [
+            HunYuanChoice(FinishReason=choice.get("FinishReason", ""), Delta=HunYuanDelta(**choice.get("Delta", {})))
+            for choice in data.get("Choices", [])
         ]
         return chuck
 

entry/settings.py

@@ -204,12 +204,8 @@
 GEMINI_API_KEY = os.getenv("GEMINI_API_KEY", "")
 
 # QCLOUD
-QCLOUD_APP_ID = int(os.getenv("QCLOUD_APP_ID", "0"))
 QCLOUD_SECRET_ID = os.getenv("QCLOUD_SECRET_ID")
 QCLOUD_SECRET_KEY = os.getenv("QCLOUD_SECRET_KEY")
-QCLOUD_HUNYUAN_API_URL = os.getenv(
-    "QCLOUD_HUNYUAN_API_DOMAIN", "https://hunyuan.cloud.tencent.com/hyllm/v1/chat/completions"
-)
 
 # Baidu Qianfan
 QIANFAN_ACCESS_KEY = os.getenv("QIANFAN_ACCESS_KEY", "")

requirements.txt

@@ -28,3 +28,6 @@ google-ai-generativelanguage==0.4.0
 
 # Baidu
 qianfan==0.3.0
+
+# Hunyuan
+tencentcloud-sdk-python==3.0.1137