feat: swap openai for Groq via openai-compatible base_url

config.py

@@ -20,5 +20,7 @@ CHROME_PATH = 'C:/Program Files (x86)/Google/Chrome/Application/chrome.exe %s'
 # Токен Picovoice
 PICOVOICE_TOKEN = os.getenv('PICOVOICE_TOKEN')
 
-# Токен OpenAI
+# Токен Groq
-OPENAI_TOKEN = os.getenv('OPENAI_TOKEN')
+GROQ_TOKEN = os.getenv('GROQ_TOKEN')
+GROQ_BASE_URL = "https://api.groq.com/openai/v1"
+GROQ_MODEL = "llama-3.3-70b-versatile"

dev.env

@@ -1,2 +1,2 @@
 PICOVOICE_TOKEN=<тут ключ-picovoice>
-OPENAI_TOKEN=<тут ключ-openai>
+GROQ_TOKEN=<тут ключ-groq>

main.py

@@ -10,7 +10,7 @@ import time
 from ctypes import POINTER, cast
 
 import openai
-from openai import error
+from openai import OpenAI
 import pvporcupine
 import simpleaudio as sa
 import vosk
@@ -37,8 +37,7 @@ VA_CMD_LIST = yaml.safe_load(
 system_message = {"role": "system", "content": "Ты голосовой ассистент из железного человека."}
 message_log = [system_message]
 
-# init openai
+client = OpenAI(api_key=config.GROQ_TOKEN, base_url=config.GROQ_BASE_URL)
-openai.api_key = config.OPENAI_TOKEN
 
 # PORCUPINE
 porcupine = pvporcupine.create(
@@ -59,33 +58,29 @@ q = queue.Queue()
 def gpt_answer():
     global message_log
 
-    model_engine = "gpt-3.5-turbo"
-    max_tokens = 256  # default 1024
     try:
-        response = openai.ChatCompletion.create(
+        response = client.chat.completions.create(
-            model=model_engine,
+            model=config.GROQ_MODEL,
             messages=message_log,
-            max_tokens=max_tokens,
+            max_tokens=256,
             temperature=0.7,
             top_p=1,
-            stop=None
+            stop=None,
         )
-    except (error.TryAgain, error.ServiceUnavailableError):
+    except openai.BadRequestError as ex:
-        return "ChatGPT перегружен!"
+        code = getattr(ex, "code", None) or ""
-    except openai.OpenAIError as ex:
+        message = str(ex)
-        # если ошибка - это макс длина контекста, то возвращаем ответ с очищенным контекстом
+        if code == "context_length_exceeded" or "context_length_exceeded" in message:
-        if ex.code == "context_length_exceeded":
             message_log = [system_message, message_log[-1]]
             return gpt_answer()
-        else:
+        return "Запрос отклонён моделью."
-            return "OpenAI токен не рабочий."
+    except openai.RateLimitError:
+        return "Лимит запросов исчерпан."
+    except openai.APIConnectionError:
+        return "Не могу связаться с сервером."
+    except openai.APIError:
+        return "Groq токен не рабочий."
 
-    # Find the first response from the chatbot that has text in it (some responses may not have text)
-    for choice in response.choices:
-        if "text" in choice:
-            return choice.text
-
-    # If no response with text is found, return the first response's content (which may be empty)
     return response.choices[0].message.content