From 326d6a4248a6b00f202cad6caff629881876e5ab Mon Sep 17 00:00:00 2001 From: Fangyin Cheng Date: Mon, 9 Jun 2025 16:23:10 +0800 Subject: [PATCH] feat(model): Support MiniCPM4-8B (#2754) --- README.ja.md | 2 ++ README.md | 2 ++ README.zh.md | 2 ++ .../src/dbgpt/model/adapter/hf_adapter.py | 20 +++++++++++++++++++ .../src/dbgpt/model/adapter/model_metadata.py | 13 +++++++++++- .../src/dbgpt/model/proxy/llms/siliconflow.py | 4 ++-- .../src/dbgpt/model/proxy/llms/tongyi.py | 2 +- 7 files changed, 41 insertions(+), 4 deletions(-) diff --git a/README.ja.md b/README.ja.md index 031081e76..61561ec21 100644 --- a/README.ja.md +++ b/README.ja.md @@ -162,6 +162,8 @@ DB-GPTのアーキテクチャは以下の図に示されています： 私たちは、LLaMA/LLaMA2、Baichuan、ChatGLM、Wenxin、Tongyi、Zhipuなど、オープンソースおよびAPIエージェントからの数十の大規模言語モデル（LLM）を含む幅広いモデルをサポートしています。 - ニュース + - 🔥🔥🔥 [DeepSeek-R1-0528](https://huggingface.co/deepseek-ai/DeepSeek-R1-0528) + - 🔥🔥🔥 [DeepSeek-V3-0324](https://huggingface.co/deepseek-ai/DeepSeek-V3-0324) - 🔥🔥🔥 [Qwen3-235B-A22B](https://huggingface.co/Qwen/Qwen3-235B-A22B) - 🔥🔥🔥 [Qwen3-30B-A3B](https://huggingface.co/Qwen/Qwen3-30B-A3B) - 🔥🔥🔥 [Qwen3-32B](https://huggingface.co/Qwen/Qwen3-32B) diff --git a/README.md b/README.md index eb1766e1a..be27a42b9 100644 --- a/README.md +++ b/README.md @@ -169,6 +169,8 @@ At present, we have introduced several key features to showcase our current capa We offer extensive model support, including dozens of large language models (LLMs) from both open-source and API agents, such as LLaMA/LLaMA2, Baichuan, ChatGLM, Wenxin, Tongyi, Zhipu, and many more. - News + - 🔥🔥🔥 [DeepSeek-R1-0528](https://huggingface.co/deepseek-ai/DeepSeek-R1-0528) + - 🔥🔥🔥 [DeepSeek-V3-0324](https://huggingface.co/deepseek-ai/DeepSeek-V3-0324) - 🔥🔥🔥 [Qwen3-235B-A22B](https://huggingface.co/Qwen/Qwen3-235B-A22B) - 🔥🔥🔥 [Qwen3-30B-A3B](https://huggingface.co/Qwen/Qwen3-30B-A3B) - 🔥🔥🔥 [Qwen3-32B](https://huggingface.co/Qwen/Qwen3-32B) diff --git a/README.zh.md b/README.zh.md index cbeab84d9..3410c6225 100644 --- a/README.zh.md +++ b/README.zh.md @@ -163,6 +163,8 @@ 海量模型支持，包括开源、API代理等几十种大语言模型。如LLaMA/LLaMA2、Baichuan、ChatGLM、文心、通义、智谱等。当前已支持如下模型: - 新增支持模型 + - 🔥🔥🔥 [DeepSeek-R1-0528](https://huggingface.co/deepseek-ai/DeepSeek-R1-0528) + - 🔥🔥🔥 [DeepSeek-V3-0324](https://huggingface.co/deepseek-ai/DeepSeek-V3-0324) - 🔥🔥🔥 [Qwen3-235B-A22B](https://huggingface.co/Qwen/Qwen3-235B-A22B) - 🔥🔥🔥 [Qwen3-30B-A3B](https://huggingface.co/Qwen/Qwen3-30B-A3B) - 🔥🔥🔥 [Qwen3-32B](https://huggingface.co/Qwen/Qwen3-32B) diff --git a/packages/dbgpt-core/src/dbgpt/model/adapter/hf_adapter.py b/packages/dbgpt-core/src/dbgpt/model/adapter/hf_adapter.py index 0c35c822e..51e7d4bad 100644 --- a/packages/dbgpt-core/src/dbgpt/model/adapter/hf_adapter.py +++ b/packages/dbgpt-core/src/dbgpt/model/adapter/hf_adapter.py @@ -1106,6 +1106,25 @@ class KimiVLAdapter(NewHFChatModelAdapter): return lower_model_name_or_path and "thinking" in lower_model_name_or_path +class MiniCPMAdapter(NewHFChatModelAdapter): + """ + https://huggingface.co/openbmb/MiniCPM4-8B + """ + + support_4bit: bool = True + support_8bit: bool = True + + def do_match(self, lower_model_name_or_path: Optional[str] = None): + return lower_model_name_or_path and "minicpm" in lower_model_name_or_path + + def load(self, model_path: str, from_pretrained_kwargs: dict): + if not from_pretrained_kwargs: + from_pretrained_kwargs = {} + if "trust_remote_code" not in from_pretrained_kwargs: + from_pretrained_kwargs["trust_remote_code"] = True + return super().load(model_path, from_pretrained_kwargs) + + # The following code is used to register the model adapter # The last registered model adapter is matched first register_model_adapter(CommonModelAdapter) # For all of hf models can be matched @@ -1137,3 +1156,4 @@ register_model_adapter(Qwen2VLAdapter) register_model_adapter(Internlm2Adapter) register_model_adapter(DeepseekV3R1Adapter, supported_models=COMMON_HF_DEEPSEEK__MODELS) register_model_adapter(KimiVLAdapter) +register_model_adapter(MiniCPMAdapter) diff --git a/packages/dbgpt-core/src/dbgpt/model/adapter/model_metadata.py b/packages/dbgpt-core/src/dbgpt/model/adapter/model_metadata.py index bb0d09182..f65f69909 100644 --- a/packages/dbgpt-core/src/dbgpt/model/adapter/model_metadata.py +++ b/packages/dbgpt-core/src/dbgpt/model/adapter/model_metadata.py @@ -46,7 +46,18 @@ COMMON_HF_DEEPSEEK__MODELS = [ function_calling=True, ), ModelMetadata( - model=["deepseek-ai/DeepSeek-V3"], + model=[ + "deepseek-ai/DeepSeek-R1-0528", + "deepseek-ai/DeepSeek-R1-0528-Qwen3-8B", + ], + context_length=128 * 1024, + max_output_length=64 * 1024, + description="DeepSeek-R1 by DeepSeek", + link="https://huggingface.co/deepseek-ai/DeepSeek-R1-0528", + function_calling=True, + ), + ModelMetadata( + model=["deepseek-ai/DeepSeek-V3", "deepseek-ai/DeepSeek-V3-0324"], context_length=128 * 1024, max_output_length=8 * 1024, description="DeepSeek-V3 by DeepSeek", diff --git a/packages/dbgpt-core/src/dbgpt/model/proxy/llms/siliconflow.py b/packages/dbgpt-core/src/dbgpt/model/proxy/llms/siliconflow.py index 808e42c41..154cec475 100644 --- a/packages/dbgpt-core/src/dbgpt/model/proxy/llms/siliconflow.py +++ b/packages/dbgpt-core/src/dbgpt/model/proxy/llms/siliconflow.py @@ -148,7 +148,7 @@ register_proxy_model_adapter( model=["deepseek-ai/DeepSeek-V3", "Pro/deepseek-ai/DeepSeek-V3"], context_length=64 * 1024, max_output_length=8 * 1024, - description="DeepSeek-V3 by DeepSeek", + description="DeepSeek-V3 by DeepSeek(DeepSeek-V3-0324)", link="https://siliconflow.cn/zh-cn/models", function_calling=True, ), @@ -156,7 +156,7 @@ register_proxy_model_adapter( model=["deepseek-ai/DeepSeek-R1", "Pro/deepseek-ai/DeepSeek-R1"], context_length=64 * 1024, max_output_length=8 * 1024, - description="DeepSeek-R1 by DeepSeek", + description="DeepSeek-R1 by DeepSeek(DeepSeek-R1-0528)", link="https://siliconflow.cn/zh-cn/models", function_calling=True, ), diff --git a/packages/dbgpt-core/src/dbgpt/model/proxy/llms/tongyi.py b/packages/dbgpt-core/src/dbgpt/model/proxy/llms/tongyi.py index d229f55c6..52b988be0 100644 --- a/packages/dbgpt-core/src/dbgpt/model/proxy/llms/tongyi.py +++ b/packages/dbgpt-core/src/dbgpt/model/proxy/llms/tongyi.py @@ -143,7 +143,7 @@ register_proxy_model_adapter( function_calling=True, ), ModelMetadata( - model="deepseek-r1", + model=["deepseek-r1", "deepseek-r1-0528"], context_length=64 * 1024, max_output_length=8 * 1024, description="DeepSeek-R1 by DeepSeek",