From 6e949ed6f82764234c326106087901fb9e079b32 Mon Sep 17 00:00:00 2001 From: octo-patch Date: Sun, 15 Mar 2026 00:53:46 +0800 Subject: [PATCH] feat: add MiniMax model provider with LLM and TTS support Add MiniMax as a new model provider supporting: - Chat LLM models: MiniMax-M2.5 and MiniMax-M2.5-highspeed (204K context) - TTS models: speech-2.8-hd and speech-2.8-turbo MiniMax LLM uses OpenAI-compatible API at api.minimax.io/v1. MiniMax TTS uses the native T2A v2 API with multiple voice options. --- README.md | 2 +- README_CN.md | 2 +- .../constants/model_provider_constants.py | 2 + .../impl/minimax_model_provider/__init__.py | 0 .../credential/__init__.py | 0 .../minimax_model_provider/credential/llm.py | 72 ++++++++++++++++ .../minimax_model_provider/credential/tts.py | 67 +++++++++++++++ .../minimax_model_provider/icon/__init__.py | 0 .../icon/minimax_icon_svg | 4 + .../minimax_model_provider.py | 57 +++++++++++++ .../minimax_model_provider/model/__init__.py | 0 .../impl/minimax_model_provider/model/llm.py | 23 ++++++ .../impl/minimax_model_provider/model/tts.py | 82 +++++++++++++++++++ 13 files changed, 309 insertions(+), 2 deletions(-) create mode 100644 apps/models_provider/impl/minimax_model_provider/__init__.py create mode 100644 apps/models_provider/impl/minimax_model_provider/credential/__init__.py create mode 100644 apps/models_provider/impl/minimax_model_provider/credential/llm.py create mode 100644 apps/models_provider/impl/minimax_model_provider/credential/tts.py create mode 100644 apps/models_provider/impl/minimax_model_provider/icon/__init__.py create mode 100644 apps/models_provider/impl/minimax_model_provider/icon/minimax_icon_svg create mode 100644 apps/models_provider/impl/minimax_model_provider/minimax_model_provider.py create mode 100644 apps/models_provider/impl/minimax_model_provider/model/__init__.py create mode 100644 apps/models_provider/impl/minimax_model_provider/model/llm.py create mode 100644 apps/models_provider/impl/minimax_model_provider/model/tts.py diff --git a/README.md b/README.md index d2ae0562f20..93f8e95ea70 100644 --- a/README.md +++ b/README.md @@ -16,7 +16,7 @@ MaxKB = Max Knowledge Brain, it is an open-source platform for building enterpri - **RAG Pipeline**: Supports direct uploading of documents / automatic crawling of online documents, with features for automatic text splitting, vectorization. This effectively reduces hallucinations in large models, providing a superior smart Q&A interaction experience. - **Agentic Workflow**: Equipped with a powerful workflow engine, function library and MCP tool-use, enabling the orchestration of AI processes to meet the needs of complex business scenarios. - **Seamless Integration**: Facilitates zero-coding rapid integration into third-party business systems, quickly equipping existing systems with intelligent Q&A capabilities to enhance user satisfaction. -- **Model-Agnostic**: Supports various large models, including private models (such as DeepSeek, Llama, Qwen, etc.) and public models (like OpenAI, Claude, Gemini, etc.). +- **Model-Agnostic**: Supports various large models, including private models (such as DeepSeek, Llama, Qwen, etc.) and public models (like OpenAI, Claude, Gemini, MiniMax, etc.). - **Multi Modal**: Native support for input and output text, image, audio and video. ## Quick start diff --git a/README_CN.md b/README_CN.md index 8be099addd7..da5db24891d 100644 --- a/README_CN.md +++ b/README_CN.md @@ -19,7 +19,7 @@ MaxKB = Max Knowledge Brain，是一个强大易用的企业级智能体平台 - **RAG 检索增强生成**：高效搭建本地 AI 知识库，支持直接上传文档 / 自动爬取在线文档，支持文本自动拆分、向量化，有效减少大模型幻觉，提升问答效果； - **灵活编排**：内置强大的工作流引擎、函数库和 MCP 工具调用能力，支持编排 AI 工作过程，满足复杂业务场景下的需求； - **无缝嵌入**：支持零编码快速嵌入到第三方业务系统，让已有系统快速拥有智能问答能力，提高用户满意度； -- **模型中立**：支持对接各种大模型，包括本地私有大模型（DeepSeek R1 / Qwen 3 等）、国内公共大模型（通义千问 / 腾讯混元 / 字节豆包 / 百度千帆 / 智谱 AI / Kimi 等）和国外公共大模型（OpenAI / Claude / Gemini 等）。 +- **模型中立**：支持对接各种大模型，包括本地私有大模型（DeepSeek R1 / Qwen 3 等）、国内公共大模型（通义千问 / 腾讯混元 / 字节豆包 / 百度千帆 / 智谱 AI / Kimi / MiniMax 等）和国外公共大模型（OpenAI / Claude / Gemini 等）。 MaxKB 三分钟视频介绍：https://www.bilibili.com/video/BV18JypYeEkj/ diff --git a/apps/models_provider/constants/model_provider_constants.py b/apps/models_provider/constants/model_provider_constants.py index be377b911cd..4533e0fca9b 100644 --- a/apps/models_provider/constants/model_provider_constants.py +++ b/apps/models_provider/constants/model_provider_constants.py @@ -23,6 +23,7 @@ from models_provider.impl.wenxin_model_provider.wenxin_model_provider import WenxinModelProvider from models_provider.impl.xf_model_provider.xf_model_provider import XunFeiModelProvider from models_provider.impl.xinference_model_provider.xinference_model_provider import XinferenceModelProvider +from models_provider.impl.minimax_model_provider.minimax_model_provider import MiniMaxModelProvider from models_provider.impl.zhipu_model_provider.zhipu_model_provider import ZhiPuModelProvider @@ -48,3 +49,4 @@ class ModelProvideConstants(Enum): model_anthropic_provider = AnthropicModelProvider() model_siliconCloud_provider = SiliconCloudModelProvider() model_regolo_provider = RegoloModelProvider() + model_minimax_provider = MiniMaxModelProvider() diff --git a/apps/models_provider/impl/minimax_model_provider/__init__.py b/apps/models_provider/impl/minimax_model_provider/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/apps/models_provider/impl/minimax_model_provider/credential/__init__.py b/apps/models_provider/impl/minimax_model_provider/credential/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/apps/models_provider/impl/minimax_model_provider/credential/llm.py b/apps/models_provider/impl/minimax_model_provider/credential/llm.py new file mode 100644 index 00000000000..56a191ccf91 --- /dev/null +++ b/apps/models_provider/impl/minimax_model_provider/credential/llm.py @@ -0,0 +1,72 @@ +# coding=utf-8 +from typing import Dict + +from django.utils.translation import gettext_lazy as _, gettext +from langchain_core.messages import HumanMessage + +from common import forms +from common.exception.app_exception import AppApiException +from common.forms import BaseForm, TooltipLabel +from models_provider.base_model_provider import BaseModelCredential, ValidCode +from common.utils.logger import maxkb_logger + + +class MiniMaxLLMModelParams(BaseForm): + temperature = forms.SliderField(TooltipLabel(_('Temperature'), + _('Higher values make the output more random, while lower values make it more focused and deterministic')), + required=True, default_value=1.0, + _min=0.01, + _max=1.0, + _step=0.01, + precision=2) + + max_tokens = forms.SliderField( + TooltipLabel(_('Output the maximum Tokens'), + _('Specify the maximum number of tokens that the model can generate')), + required=True, default_value=800, + _min=1, + _max=192000, + _step=1, + precision=0) + + +class MiniMaxLLMModelCredential(BaseForm, BaseModelCredential): + + def is_valid(self, model_type: str, model_name, model_credential: Dict[str, object], model_params, provider, + raise_exception=False): + model_type_list = provider.get_model_type_list() + if not any(list(filter(lambda mt: mt.get('value') == model_type, model_type_list))): + raise AppApiException(ValidCode.valid_error.value, + gettext('{model_type} Model type is not supported').format(model_type=model_type)) + + for key in ['api_key', 'api_base']: + if key not in model_credential: + if raise_exception: + raise AppApiException(ValidCode.valid_error.value, gettext('{key} is required').format(key=key)) + else: + return False + try: + model = provider.get_model(model_type, model_name, model_credential, **model_params) + model.invoke([HumanMessage(content=gettext('Hello'))]) + except Exception as e: + maxkb_logger.error(f'Exception: {e}', exc_info=True) + if isinstance(e, AppApiException): + raise e + if raise_exception: + raise AppApiException(ValidCode.valid_error.value, + gettext( + 'Verification failed, please check whether the parameters are correct: {error}').format( + error=str(e))) + else: + return False + return True + + def encryption_dict(self, model: Dict[str, object]): + return {**model, 'api_key': super().encryption(model.get('api_key', ''))} + + api_base = forms.TextInputField('API URL', required=True, + default_value='https://api.minimax.io/v1') + api_key = forms.PasswordInputField('API Key', required=True) + + def get_model_params_setting_form(self, model_name): + return MiniMaxLLMModelParams() diff --git a/apps/models_provider/impl/minimax_model_provider/credential/tts.py b/apps/models_provider/impl/minimax_model_provider/credential/tts.py new file mode 100644 index 00000000000..827795bd168 --- /dev/null +++ b/apps/models_provider/impl/minimax_model_provider/credential/tts.py @@ -0,0 +1,67 @@ +# coding=utf-8 +from typing import Dict + +from django.utils.translation import gettext_lazy as _, gettext + +from common import forms +from common.exception.app_exception import AppApiException +from common.forms import BaseForm, TooltipLabel +from models_provider.base_model_provider import BaseModelCredential, ValidCode +from common.utils.logger import maxkb_logger + + +class MiniMaxTTSModelGeneralParams(BaseForm): + voice_id = forms.SingleSelect( + TooltipLabel(_('Voice'), + _('Select a voice for speech synthesis')), + required=True, default_value='English_Graceful_Lady', + text_field='value', + value_field='value', + option_list=[ + {'text': 'English_Graceful_Lady', 'value': 'English_Graceful_Lady'}, + {'text': 'English_Insightful_Speaker', 'value': 'English_Insightful_Speaker'}, + {'text': 'English_radiant_girl', 'value': 'English_radiant_girl'}, + {'text': 'English_Persuasive_Man', 'value': 'English_Persuasive_Man'}, + {'text': 'English_Lucky_Robot', 'value': 'English_Lucky_Robot'}, + ]) + + +class MiniMaxTTSModelCredential(BaseForm, BaseModelCredential): + api_base = forms.TextInputField('API URL', required=True, + default_value='https://api.minimax.io/v1') + api_key = forms.PasswordInputField('API Key', required=True) + + def is_valid(self, model_type: str, model_name, model_credential: Dict[str, object], model_params, provider, + raise_exception=False): + model_type_list = provider.get_model_type_list() + if not any(list(filter(lambda mt: mt.get('value') == model_type, model_type_list))): + raise AppApiException(ValidCode.valid_error.value, + gettext('{model_type} Model type is not supported').format(model_type=model_type)) + + for key in ['api_base', 'api_key']: + if key not in model_credential: + if raise_exception: + raise AppApiException(ValidCode.valid_error.value, gettext('{key} is required').format(key=key)) + else: + return False + try: + model = provider.get_model(model_type, model_name, model_credential, **model_params) + model.check_auth() + except Exception as e: + maxkb_logger.error(f'Exception: {e}', exc_info=True) + if isinstance(e, AppApiException): + raise e + if raise_exception: + raise AppApiException(ValidCode.valid_error.value, + gettext( + 'Verification failed, please check whether the parameters are correct: {error}').format( + error=str(e))) + else: + return False + return True + + def encryption_dict(self, model: Dict[str, object]): + return {**model, 'api_key': super().encryption(model.get('api_key', ''))} + + def get_model_params_setting_form(self, model_name): + return MiniMaxTTSModelGeneralParams() diff --git a/apps/models_provider/impl/minimax_model_provider/icon/__init__.py b/apps/models_provider/impl/minimax_model_provider/icon/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/apps/models_provider/impl/minimax_model_provider/icon/minimax_icon_svg b/apps/models_provider/impl/minimax_model_provider/icon/minimax_icon_svg new file mode 100644 index 00000000000..553af21e691 --- /dev/null +++ b/apps/models_provider/impl/minimax_model_provider/icon/minimax_icon_svg @@ -0,0 +1,4 @@ + + + M + diff --git a/apps/models_provider/impl/minimax_model_provider/minimax_model_provider.py b/apps/models_provider/impl/minimax_model_provider/minimax_model_provider.py new file mode 100644 index 00000000000..55bfb968c1e --- /dev/null +++ b/apps/models_provider/impl/minimax_model_provider/minimax_model_provider.py @@ -0,0 +1,57 @@ +# coding=utf-8 +import os + +from common.utils.common import get_file_content +from models_provider.base_model_provider import IModelProvider, ModelProvideInfo, ModelInfo, ModelTypeConst, \ + ModelInfoManage +from models_provider.impl.minimax_model_provider.credential.llm import MiniMaxLLMModelCredential +from models_provider.impl.minimax_model_provider.credential.tts import MiniMaxTTSModelCredential +from models_provider.impl.minimax_model_provider.model.llm import MiniMaxChatModel +from models_provider.impl.minimax_model_provider.model.tts import MiniMaxTextToSpeech +from maxkb.conf import PROJECT_DIR +from django.utils.translation import gettext_lazy as _ + +minimax_llm_model_credential = MiniMaxLLMModelCredential() +minimax_tts_model_credential = MiniMaxTTSModelCredential() + +minimax_m2_5 = ModelInfo('MiniMax-M2.5', + _('Peak Performance. Ultimate Value. 204K context window'), + ModelTypeConst.LLM, + minimax_llm_model_credential, MiniMaxChatModel) + +minimax_m2_5_highspeed = ModelInfo('MiniMax-M2.5-highspeed', + _('Same performance, faster and more agile. 204K context window'), + ModelTypeConst.LLM, + minimax_llm_model_credential, MiniMaxChatModel) + +minimax_tts_hd = ModelInfo('speech-2.8-hd', + _('Perfecting tonal nuances with maximized timbre similarity'), + ModelTypeConst.TTS, + minimax_tts_model_credential, MiniMaxTextToSpeech) + +minimax_tts_turbo = ModelInfo('speech-2.8-turbo', + _('Faster, more affordable TTS model'), + ModelTypeConst.TTS, + minimax_tts_model_credential, MiniMaxTextToSpeech) + +model_info_manage = ( + ModelInfoManage.builder() + .append_model_info(minimax_m2_5) + .append_model_info(minimax_m2_5_highspeed) + .append_default_model_info(minimax_m2_5) + .append_model_info(minimax_tts_hd) + .append_model_info(minimax_tts_turbo) + .append_default_model_info(minimax_tts_hd) + .build() +) + + +class MiniMaxModelProvider(IModelProvider): + + def get_model_info_manage(self): + return model_info_manage + + def get_model_provide_info(self): + return ModelProvideInfo(provider='model_minimax_provider', name='MiniMax', icon=get_file_content( + os.path.join(PROJECT_DIR, "apps", 'models_provider', 'impl', 'minimax_model_provider', 'icon', + 'minimax_icon_svg'))) diff --git a/apps/models_provider/impl/minimax_model_provider/model/__init__.py b/apps/models_provider/impl/minimax_model_provider/model/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/apps/models_provider/impl/minimax_model_provider/model/llm.py b/apps/models_provider/impl/minimax_model_provider/model/llm.py new file mode 100644 index 00000000000..27a14cc115a --- /dev/null +++ b/apps/models_provider/impl/minimax_model_provider/model/llm.py @@ -0,0 +1,23 @@ +# coding=utf-8 +from typing import Dict + +from models_provider.base_model_provider import MaxKBBaseModel +from models_provider.impl.base_chat_open_ai import BaseChatOpenAI + + +class MiniMaxChatModel(MaxKBBaseModel, BaseChatOpenAI): + + @staticmethod + def is_cache_model(): + return False + + @staticmethod + def new_instance(model_type, model_name, model_credential: Dict[str, object], **model_kwargs): + optional_params = MaxKBBaseModel.filter_optional_params(model_kwargs) + + return MiniMaxChatModel( + model=model_name, + openai_api_base=model_credential.get('api_base') or 'https://api.minimax.io/v1', + openai_api_key=model_credential.get('api_key'), + extra_body=optional_params + ) diff --git a/apps/models_provider/impl/minimax_model_provider/model/tts.py b/apps/models_provider/impl/minimax_model_provider/model/tts.py new file mode 100644 index 00000000000..8ba3eb81090 --- /dev/null +++ b/apps/models_provider/impl/minimax_model_provider/model/tts.py @@ -0,0 +1,82 @@ +# coding=utf-8 +from typing import Dict + +import requests + +from django.utils.translation import gettext as _ + +from common.utils.common import _remove_empty_lines +from models_provider.base_model_provider import MaxKBBaseModel +from models_provider.impl.base_tts import BaseTextToSpeech + + +class MiniMaxTextToSpeech(MaxKBBaseModel, BaseTextToSpeech): + api_base: str + api_key: str + model: str + params: dict + + def __init__(self, **kwargs): + super().__init__(**kwargs) + self.api_key = kwargs.get('api_key') + self.api_base = kwargs.get('api_base') + self.model = kwargs.get('model') + self.params = kwargs.get('params') + + @staticmethod + def is_cache_model(): + return False + + @staticmethod + def new_instance(model_type, model_name, model_credential: Dict[str, object], **model_kwargs): + optional_params = {'params': {'voice_id': 'English_Graceful_Lady'}} + for key, value in model_kwargs.items(): + if key not in ['model_id', 'use_local', 'streaming']: + optional_params['params'][key] = value + return MiniMaxTextToSpeech( + model=model_name, + api_base=model_credential.get('api_base') or 'https://api.minimax.io/v1', + api_key=model_credential.get('api_key'), + **optional_params, + ) + + def check_auth(self): + self.text_to_speech(_('Hello')) + + def text_to_speech(self, text): + text = _remove_empty_lines(text) + api_base = self.api_base.rstrip('/') + url = f'{api_base}/t2a_v2' + + voice_id = self.params.get('voice_id', 'English_Graceful_Lady') + + payload = { + 'model': self.model, + 'text': text, + 'stream': False, + 'voice_setting': { + 'voice_id': voice_id, + }, + 'audio_setting': { + 'format': 'mp3', + }, + } + + headers = { + 'Authorization': f'Bearer {self.api_key}', + 'Content-Type': 'application/json', + } + + response = requests.post(url, json=payload, headers=headers, timeout=60) + response.raise_for_status() + + result = response.json() + if result.get('base_resp', {}).get('status_code', 0) != 0: + error_msg = result.get('base_resp', {}).get('status_msg', 'Unknown error') + raise Exception(f'MiniMax TTS API error: {error_msg}') + + audio_hex = result.get('data', {}).get('audio', '') + if not audio_hex: + raise Exception('MiniMax TTS API returned empty audio data') + + return bytes.fromhex(audio_hex)