update

swift/llm/model/model/glm.py

@@ -134,11 +134,11 @@ def get_model_tokenizer_glm4(model_dir: str,
                 Model('ZhipuAI/GLM-4-Z1-32B-0414', 'THUDM/GLM-4-Z1-32B-0414'),
             ])
         ],
-        TemplateType.glm4,
+        TemplateType.glm4_z1,
         get_model_tokenizer_with_flash_attn,
         architectures=['Glm4ForCausalLM'],
         model_arch=ModelArch.chatglm,
-        requires=['transformers>4.5.1'],
+        requires=['transformers>4.51'],
     ))
 
 register_model(

swift/llm/template/constant.py

@@ -35,6 +35,7 @@ class LLMTemplateType:
 
     chatglm2 = 'chatglm2'
     glm4 = 'glm4'
+    glm4_z1 = 'glm4_z1'
     codegeex4 = 'codegeex4'
     longwriter_llama = 'longwriter_llama'
 

swift/llm/template/template/glm.py

@@ -17,6 +17,16 @@ class GLMTemplateMeta(TemplateMeta):
     auto_add_bos: bool = True
 
 
+class GLM4Z1Template(Template):
+
+    def _swift_encode(self, inputs: StdTemplateInputs):
+        if not self.is_training:
+            for message in inputs.messages:
+                if message['role'] == 'assistant' and isinstance(message['content'], str):
+                    message['content'] = message['content'].split('</think>')[-1].strip()
+        return super()._swift_encode(inputs)
+
+
 register_template(
     GLMTemplateMeta(
         LLMTemplateType.chatglm2,
@@ -27,17 +37,23 @@ class GLMTemplateMeta(TemplateMeta):
 
 @dataclass
 class GLM4TemplateMeta(GLMTemplateMeta):
-    prefix: Prompt = field(default_factory=lambda: ['[gMASK]<sop>'])
+    prefix: Prompt = field(default_factory=lambda: [])
     prompt: Prompt = field(default_factory=lambda: ['<|user|>\n{{QUERY}}<|assistant|>\n'])
     chat_sep: Optional[Prompt] = field(default_factory=list)
     suffix: Prompt = field(default_factory=lambda: ['<|user|>'])
-    system_prefix: Optional[Prompt] = field(default_factory=lambda: ['[gMASK]<sop><|system|>\n{{SYSTEM}}'])
+    system_prefix: Optional[Prompt] = field(default_factory=lambda: ['<|system|>\n{{SYSTEM}}'])
 
     default_tools_prompt: str = 'glm4'
     tool_prompt: Optional[Prompt] = field(default_factory=lambda: ['<|observation|>\n{{QUERY}}<|assistant|>\n'])
     stop_words: List[Word] = field(default_factory=lambda: ['<|endoftext|>', '<|user|>', '<|observation|>'])
 
 
+@dataclass
+class GLM4Z1TemplateMeta(GLMTemplateMeta):
+    prefix: Prompt = field(default_factory=lambda: ['[gMASK]<sop>'])
+    system_prefix: Optional[Prompt] = field(default_factory=lambda: ['[gMASK]<sop><|system|>\n{{SYSTEM}}'])
+
+
 class GLM4VTemplate(Template):
 
     def replace_tag(self, media_type: Literal['image', 'video', 'audio'], index: int,
@@ -80,6 +96,8 @@ def _data_collator(self, batch: List[Dict[str, Any]], *, padding_to: Optional[in
 
 register_template(GLM4TemplateMeta(LLMTemplateType.glm4))
 
+register_template(GLM4Z1TemplateMeta(LLMTemplateType.glm4_z1, template_cls=GLM4Z1Template))
+
 codegeex4_system = '你是一位智能编程助手，你叫CodeGeeX。你会为用户回答关于编程、代码、计算机方面的任何问题，并提供格式规范、可以执行、准确安全的代码，并在必要时提供详细的解释。'
 
 register_template(GLM4TemplateMeta(LLMTemplateType.codegeex4, default_system=codegeex4_system))