feat: add MiniMax provider support

langkit/openai/__init__.py

@@ -8,6 +8,8 @@
     OpenAIGPT4,
     OpenAIDefault,
     OpenAILegacy,
+    MiniMaxDefault,
+    MiniMaxHighSpeed,
 )
 
 __ALL__ = [
@@ -20,4 +22,6 @@
     OpenAIDefault,
     OpenAIGPT4,
     OpenAILegacy,
+    MiniMaxDefault,
+    MiniMaxHighSpeed,
 ]

langkit/openai/openai.py

@@ -6,6 +6,21 @@
 
 
 openai.api_key = os.getenv("OPENAI_API_KEY")
+
+_minimax_default_model = "MiniMax-M2.7"
+_minimax_highspeed_model = "MiniMax-M2.7-highspeed"
+_minimax_default_base_url = "https://api.minimax.io/v1"
+
+
+def create_minimax_chat_completion(messages, api_key, base_url, **params):
+    openai_version = get_openai_version()
+    if openai_version.startswith("0."):
+        openai.api_base = base_url
+        openai.api_key = api_key
+        return openai.ChatCompletion.create(messages=messages, **params)
+    else:
+        client = openai.OpenAI(api_key=api_key, base_url=base_url)
+        return client.chat.completions.create(messages=messages, **params)
 _openai_llm_model = os.getenv("LANGKIT_OPENAI_LLM_MODEL_NAME") or "gpt-3.5-turbo"
 _llm_model_temperature = 0.9
 _llm_model_max_tokens = 1024
@@ -373,6 +388,58 @@ def send_prompt(self, prompt: str) -> ChatLog:
         )
 
 
+@dataclass
+class MiniMaxDefault(LLMInvocationParams):
+    """MiniMax chat model provider using the OpenAI-compatible API."""
+
+    model: str = field(default_factory=lambda: _minimax_default_model)
+    temperature: float = field(default_factory=lambda: 1.0)
+    max_tokens: int = field(default_factory=lambda: _llm_model_max_tokens)
+    frequency_penalty: float = field(default_factory=lambda: _llm_model_frequency_penalty)
+    presence_penalty: float = field(default_factory=lambda: _llm_model_presence_penalty)
+
+    def completion(self, messages: List[Dict[str, str]], **kwargs):
+        params = asdict(self)
+        api_key = os.getenv("MINIMAX_API_KEY")
+        base_url = os.getenv("MINIMAX_BASE_URL") or _minimax_default_base_url
+        return create_minimax_chat_completion(messages=messages, api_key=api_key, base_url=base_url, **params)
+
+    def copy(self) -> LLMInvocationParams:
+        return MiniMaxDefault(
+            model=self.model,
+            temperature=self.temperature,
+            max_tokens=self.max_tokens,
+            frequency_penalty=self.frequency_penalty,
+            presence_penalty=self.presence_penalty,
+        )
+
+
+@dataclass
+class MiniMaxHighSpeed(LLMInvocationParams):
+    """MiniMax high-speed chat model provider using the OpenAI-compatible API."""
+
+    model: str = field(default_factory=lambda: _minimax_highspeed_model)
+    temperature: float = field(default_factory=lambda: 1.0)
+    max_tokens: int = field(default_factory=lambda: _llm_model_max_tokens)
+    frequency_penalty: float = field(default_factory=lambda: _llm_model_frequency_penalty)
+    presence_penalty: float = field(default_factory=lambda: _llm_model_presence_penalty)
+
+    def completion(self, messages: List[Dict[str, str]], **kwargs):
+        params = asdict(self)
+        api_key = os.getenv("MINIMAX_API_KEY")
+        base_url = os.getenv("MINIMAX_BASE_URL") or _minimax_default_base_url
+        return create_minimax_chat_completion(messages=messages, api_key=api_key, base_url=base_url, **params)
+
+    def copy(self) -> LLMInvocationParams:
+        return MiniMaxHighSpeed(
+            model=self.model,
+            temperature=self.temperature,
+            max_tokens=self.max_tokens,
+            frequency_penalty=self.frequency_penalty,
+            presence_penalty=self.presence_penalty,
+        )
+
+
 # this is just for demonstration purposes
 def send_prompt(prompt: str) -> ChatLog:
     try:

langkit/tests/test_minimax.py

@@ -0,0 +1,108 @@
+"""Unit tests for MiniMax LLM provider classes."""
+from dataclasses import asdict
+from unittest.mock import MagicMock, patch
+from langkit.openai.openai import MiniMaxDefault, MiniMaxHighSpeed, _minimax_default_base_url
+
+
+def test_minimax_default_model():
+    provider = MiniMaxDefault()
+    assert provider.model == "MiniMax-M2.7"
+
+
+def test_minimax_highspeed_model():
+    provider = MiniMaxHighSpeed()
+    assert provider.model == "MiniMax-M2.7-highspeed"
+
+
+def test_minimax_default_temperature():
+    provider = MiniMaxDefault()
+    assert provider.temperature == 1.0
+
+
+def test_minimax_highspeed_temperature():
+    provider = MiniMaxHighSpeed()
+    assert provider.temperature == 1.0
+
+
+def test_minimax_default_base_url():
+    assert _minimax_default_base_url == "https://api.minimax.io/v1"
+
+
+def test_minimax_default_copy():
+    provider = MiniMaxDefault()
+    copy = provider.copy()
+    assert isinstance(copy, MiniMaxDefault)
+    assert copy.model == provider.model
+    assert copy.temperature == provider.temperature
+
+
+def test_minimax_highspeed_copy():
+    provider = MiniMaxHighSpeed()
+    copy = provider.copy()
+    assert isinstance(copy, MiniMaxHighSpeed)
+    assert copy.model == provider.model
+    assert copy.temperature == provider.temperature
+
+
+def test_minimax_default_completion(monkeypatch):
+    monkeypatch.setenv("MINIMAX_API_KEY", "test-key")
+
+    mock_response = MagicMock()
+    mock_response.choices = [MagicMock()]
+    mock_response.choices[0].message.content = "Hello!"
+    mock_response.usage.total_tokens = 10
+
+    with patch("langkit.openai.openai.create_minimax_chat_completion", return_value=mock_response) as mock_fn:
+        provider = MiniMaxDefault()
+        messages = [{"role": "user", "content": "Hi"}]
+        result = provider.completion(messages)
+
+        mock_fn.assert_called_once()
+        call_kwargs = mock_fn.call_args[1]
+        assert call_kwargs["api_key"] == "test-key"
+        assert call_kwargs["base_url"] == "https://api.minimax.io/v1"
+        assert call_kwargs["model"] == "MiniMax-M2.7"
+        assert call_kwargs["temperature"] == 1.0
+        assert result == mock_response
+
+
+def test_minimax_highspeed_completion(monkeypatch):
+    monkeypatch.setenv("MINIMAX_API_KEY", "test-key")
+
+    mock_response = MagicMock()
+    mock_response.choices = [MagicMock()]
+    mock_response.choices[0].message.content = "Hello fast!"
+    mock_response.usage.total_tokens = 5
+
+    with patch("langkit.openai.openai.create_minimax_chat_completion", return_value=mock_response) as mock_fn:
+        provider = MiniMaxHighSpeed()
+        messages = [{"role": "user", "content": "Hi"}]
+        result = provider.completion(messages)
+
+        mock_fn.assert_called_once()
+        call_kwargs = mock_fn.call_args[1]
+        assert call_kwargs["api_key"] == "test-key"
+        assert call_kwargs["base_url"] == "https://api.minimax.io/v1"
+        assert call_kwargs["model"] == "MiniMax-M2.7-highspeed"
+        assert call_kwargs["temperature"] == 1.0
+        assert result == mock_response
+
+
+def test_minimax_custom_base_url(monkeypatch):
+    monkeypatch.setenv("MINIMAX_API_KEY", "test-key")
+    monkeypatch.setenv("MINIMAX_BASE_URL", "https://custom.minimax.io/v1")
+
+    mock_response = MagicMock()
+    with patch("langkit.openai.openai.create_minimax_chat_completion", return_value=mock_response) as mock_fn:
+        provider = MiniMaxDefault()
+        messages = [{"role": "user", "content": "Hi"}]
+        provider.completion(messages)
+
+        call_kwargs = mock_fn.call_args[1]
+        assert call_kwargs["base_url"] == "https://custom.minimax.io/v1"
+
+
+def test_minimax_exported_from_openai_module():
+    from langkit.openai import MiniMaxDefault as MD, MiniMaxHighSpeed as MH
+    assert MD is MiniMaxDefault
+    assert MH is MiniMaxHighSpeed