[Feature] add MiniMax as text LLM API provider (#1491)

* feat: add MiniMax as text LLM API provider

Add MiniMax (https://platform.minimaxi.com) as a text-based LLM provider
for evaluation, supporting MiniMax-M2.7, MiniMax-M2.5, and
MiniMax-M2.5-highspeed models via OpenAI-compatible chat completions API.

Changes:
- New vlmeval/api/minimax_api.py: MiniMaxAPI class extending BaseAPI
- Register MiniMaxAPI and 3 model configs in vlmeval/config.py
- Add MINIMAX_API_KEY to EN/CN quickstart docs
- Add 34 unit tests + 3 integration tests

* Delete tests/test_minimax_api.py

---------

Co-authored-by: PR Bot <pr-bot@minimaxi.com>
Co-authored-by: Ma Zerun <mzr1996@163.com>

Author: 3 people

docs/en/Quickstart.md

@@ -45,6 +45,8 @@ To infer with API models (GPT-4v, Gemini-Pro-V, etc.) or use LLM APIs as the **j
   HUNYUAN_SECRET_ID=
   # LMDeploy API
   LMDEPLOY_API_BASE=
+  # MiniMax API
+  MINIMAX_API_KEY=
   # You can also set a proxy for calling api models during the evaluation stage
   EVAL_PROXY=
   ```

docs/zh-CN/Quickstart.md

@@ -44,6 +44,8 @@ pip install -e .
   HUNYUAN_SECRET_ID=
   # LMDeploy API
   LMDEPLOY_API_BASE=
+  # MiniMax API
+  MINIMAX_API_KEY=
   # 你可以设置一个评估时代理，评估阶段产生的 API 调用将通过这个代理进行
   EVAL_PROXY=
   ```

vlmeval/api/__init__.py

@@ -30,6 +30,7 @@
 from .together import TogetherAPI
 from .gcp_vertex import GCPVertexAPI
 from .bedrock import BedrockAPI
+from .minimax_api import MiniMaxAPI
 
 __all__ = [
     'OpenAIWrapper', 'HFChatModel', 'GeminiWrapper', 'GPT4V', 'Gemini', 'QwenVLWrapper',
@@ -40,5 +41,5 @@
     'TaichuVLAPI', 'TaichuVLRAPI', 'DoubaoVL', "MUGUAPI", 'KimiVLAPIWrapper', 'KimiVLAPI',
     'RBdashMMChat3_API', 'RBdashChat3_5_API', 'RBdashMMChat3_78B_API', 'RBdashMMChat3_5_38B_API',
     'VideoChatOnlineV2API', 'TeleMM2_API', 'TeleMM2Thinking_API', 'TogetherAPI', 'GCPVertexAPI',
-    'BedrockAPI', 'SenseChatVisionV2API'
+    'BedrockAPI', 'SenseChatVisionV2API', 'MiniMaxAPI',
 ]

vlmeval/api/minimax_api.py

@@ -0,0 +1,123 @@
+"""
+MiniMax API support for text-based LLM evaluation.
+OpenAI-compatible chat completions endpoint.
+Set MINIMAX_API_KEY or pass key=...
+
+Models: MiniMax-M2.7, MiniMax-M2.5, MiniMax-M2.5-highspeed
+API docs: https://platform.minimaxi.com/document/guides/chat-model/text-generation
+"""
+import json
+import os
+
+import requests
+
+from ..smp import get_logger
+from .base import BaseAPI
+
+MINIMAX_API_BASE = "https://api.minimax.io/v1/chat/completions"
+
+
+class MiniMaxAPI(BaseAPI):
+    """Text LLM API using MiniMax (OpenAI-compatible)."""
+
+    is_api: bool = True
+
+    def __init__(
+        self,
+        model: str = "MiniMax-M2.7",
+        key: str = None,
+        api_base: str = None,
+        retry: int = 10,
+        wait: int = 1,
+        system_prompt: str = None,
+        verbose: bool = True,
+        temperature: float = 0,
+        max_tokens: int = 2048,
+        timeout: int = 300,
+        **kwargs,
+    ):
+        self.model = model
+        self.key = key or os.environ.get("MINIMAX_API_KEY")
+        self.api_base = api_base or os.environ.get("MINIMAX_API_BASE", MINIMAX_API_BASE)
+        self.temperature = temperature
+        self.max_tokens = max_tokens
+        self.timeout = timeout
+
+        if not self.key:
+            raise ValueError(
+                "MiniMax API key is required. Set MINIMAX_API_KEY or pass key=..."
+            )
+
+        super().__init__(
+            retry=retry,
+            wait=wait,
+            system_prompt=system_prompt,
+            verbose=verbose,
+            **kwargs,
+        )
+
+        self.logger.info(f"MiniMaxAPI: model={self.model}, api_base={self.api_base}")
+
+    def _prepare_messages(self, inputs):
+        """Build OpenAI-style messages from VLMEvalKit input format."""
+        messages = []
+        if self.system_prompt:
+            messages.append({"role": "system", "content": self.system_prompt})
+
+        # Handle multi-turn chat format
+        if inputs and isinstance(inputs[0], dict) and "role" in inputs[0]:
+            for item in inputs:
+                content_parts = item.get("content", [])
+                text = "\n".join(
+                    x["value"] for x in content_parts if x["type"] == "text"
+                )
+                messages.append({"role": item["role"], "content": text or ""})
+        else:
+            # Single-turn: extract text from inputs
+            text = "\n".join(
+                x["value"] for x in inputs if x["type"] == "text"
+            )
+            messages.append({"role": "user", "content": text or ""})
+
+        return messages
+
+    def generate_inner(self, inputs, **kwargs):
+        temperature = kwargs.pop("temperature", self.temperature)
+        max_tokens = kwargs.pop("max_tokens", self.max_tokens)
+
+        messages = self._prepare_messages(inputs)
+        payload = {
+            "model": self.model,
+            "messages": messages,
+            "temperature": temperature,
+            "max_tokens": max_tokens,
+        }
+
+        try:
+            response = requests.post(
+                self.api_base,
+                headers={
+                    "Authorization": f"Bearer {self.key}",
+                    "Content-Type": "application/json",
+                },
+                data=json.dumps(payload),
+                timeout=self.timeout * 1.1,
+            )
+        except Exception as err:
+            if self.verbose:
+                self.logger.error(f"{type(err).__name__}: {err}")
+            return -1, self.fail_msg, str(err)
+
+        ret_code = response.status_code
+        ret_code = 0 if (200 <= ret_code < 300) else ret_code
+        answer = self.fail_msg
+
+        try:
+            data = response.json()
+            answer = data["choices"][0]["message"]["content"].strip()
+        except Exception as err:
+            if self.verbose:
+                self.logger.error(f"{type(err).__name__}: {err}")
+                self.logger.error(response.text)
+
+        return ret_code, answer, response

vlmeval/config.py

@@ -465,6 +465,28 @@
         max_tokens=2048,
         retry=10,
     ),
+    # MiniMax (set MINIMAX_API_KEY)
+    "MiniMax-M2.7": partial(
+        MiniMaxAPI,
+        model="MiniMax-M2.7",
+        temperature=0,
+        max_tokens=2048,
+        retry=10,
+    ),
+    "MiniMax-M2.5": partial(
+        MiniMaxAPI,
+        model="MiniMax-M2.5",
+        temperature=0,
+        max_tokens=2048,
+        retry=10,
+    ),
+    "MiniMax-M2.5-highspeed": partial(
+        MiniMaxAPI,
+        model="MiniMax-M2.5-highspeed",
+        temperature=0,
+        max_tokens=2048,
+        retry=10,
+    ),
     # Claude
     "Claude3V_Opus": partial(
         Claude3V, model="claude-3-opus-20240229", temperature=0, retry=10, verbose=False