feat(model): Support GLM-4-0414 (#2630)

packages/dbgpt-core/pyproject.toml

@@ -147,6 +147,9 @@ hf_qwen_omni = [
     # uv pip install git+https://github.com/huggingface/transformers@v4.51.3-Qwen2.5-Omni-preview
     # "transformers>=4.52.0",
 ]
+hf_glm4 = [
+    "transformers>=4.51.3",
+]
 hf_kimi = [
     "transformers<4.51.3",
 ]
@@ -165,6 +168,10 @@ conflicts = [
         { extra = "hf_qwen_omni" },
         { extra = "hf_kimi" },
     ],
+    [
+        { extra = "hf_glm4" },
+        { extra = "hf_kimi" },
+    ],
 ]
 [tool.hatch.build.targets.wheel]
 packages = ["src/dbgpt"]

packages/dbgpt-core/src/dbgpt/model/adapter/hf_adapter.py

@@ -15,7 +15,6 @@ from dbgpt.model.adapter.model_metadata import (
     COMMON_HF_QWEN25_MODELS,
 )
 from dbgpt.model.base import ModelType
-from dbgpt.model.utils.media_utils import parse_messages
 from dbgpt.util.i18n_utils import _
 
 logger = logging.getLogger(__name__)
@@ -295,6 +294,8 @@ class NewHFChatModelAdapter(LLMModelAdapter, ABC):
             messages: The messages to load.
             tokenizer: The tokenizer to use.
         """
+        from dbgpt.model.utils.media_utils import parse_messages
+
         results = parse_messages(messages)
         if "images" in results and results["images"]:
             params["images"] = results["images"]
@@ -879,6 +880,38 @@ class GLM4Adapter(NewHFChatModelAdapter):
         )
 
 
+class GLM40414Adapter(NewHFChatModelAdapter):
+    """
+    https://huggingface.co/collections/THUDM/glm-4-0414-67f3cbcb34dd9d252707cb2e
+    """
+
+    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 "glm-4" in lower_model_name_or_path
+            and "0414" in lower_model_name_or_path
+            and "base" not in lower_model_name_or_path
+        ) or (lower_model_name_or_path and "glm-z1" in lower_model_name_or_path)
+
+    def use_fast_tokenizer(self) -> bool:
+        return True
+
+    def is_reasoning_model(
+        self,
+        deploy_model_params: LLMDeployModelParameters,
+        lower_model_name_or_path: Optional[str] = None,
+    ) -> bool:
+        if (
+            deploy_model_params.reasoning_model is not None
+            and deploy_model_params.reasoning_model
+        ):
+            return True
+        return lower_model_name_or_path and "z1" in lower_model_name_or_path
+
+
 class Codegeex4Adapter(GLM4Adapter):
     """
     https://huggingface.co/THUDM/codegeex4-all-9b
@@ -974,6 +1007,7 @@ register_model_adapter(PhiAdapter)
 register_model_adapter(SQLCoderAdapter)
 register_model_adapter(OpenChatAdapter)
 register_model_adapter(GLM4Adapter, supported_models=COMMON_HF_GLM_MODELS)
+register_model_adapter(GLM40414Adapter)
 register_model_adapter(Codegeex4Adapter)
 register_model_adapter(Qwen2Adapter, supported_models=COMMON_HF_QWEN25_MODELS)
 register_model_adapter(Internlm2Adapter)

packages/dbgpt-core/src/dbgpt/model/adapter/loader.py

@@ -216,7 +216,7 @@ def _hf_try_load_default_quantization_model(
         if model:
             logger.info(f"Load default quantization model {model_name} success")
             return _hf_handle_model_and_tokenizer(
-                model, tokenizer, device, num_gpus, model_params
+                model, tokenizer, device, num_gpus, model_params, to=False
             )
         return None, None
     except Exception as e:
@@ -232,11 +232,13 @@ def _hf_handle_model_and_tokenizer(
     device: str,
     num_gpus: int,
     model_params: HFLLMDeployModelParameters,
+    to: bool = True,
 ):
     if (device == "cuda" and num_gpus == 1) or device == "mps" and tokenizer:
         # TODO: Check cpu_offloading
         try:
-            model.to(device)
+            if to:
+                model.to(device)
         except ValueError:
             pass
         except AttributeError:

pyproject.toml

@@ -29,12 +29,12 @@ explicit = true
 
 [tool.uv.workspace]
 members = [
-  "packages/dbgpt-app", 
+  "packages/dbgpt-app",
   "packages/dbgpt-client",
-  "packages/dbgpt-core", 
+  "packages/dbgpt-core",
-  "packages/dbgpt-ext", 
+  "packages/dbgpt-ext",
-  "packages/dbgpt-serve", 
+  "packages/dbgpt-serve",
-  "packages/dbgpt-accelerator/dbgpt-acc*"
+  "packages/dbgpt-accelerator/dbgpt-acc*",
 ]
 
 [tool.uv]