[model] support GLM-4.7-Flash SFT (#10173)

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

src/llamafactory/data/template.py

@@ -459,6 +459,18 @@ class ReasoningTemplate(Template):
         return [(encoded_messages[i], encoded_messages[i + 1]) for i in range(0, len(encoded_messages), 2)]
 
 
+@dataclass
+class Glm47ReasoningTemplate(ReasoningTemplate):
+    r"""GLM-4.7 uses only the closing </think> tag for empty thinking blocks."""
+
+    @override
+    def add_thought(self, content: str = "") -> str:
+        if not content:
+            return self.thought_words[1]
+
+        return self.thought_words[0] + content + self.thought_words[1]
+
+
 TEMPLATES: dict[str, "Template"] = {}
 
 
@@ -1049,6 +1061,23 @@ register_template(
 )
 
 
+# copied from glm4_moe template
+register_template(
+    name="glm4_7",
+    format_user=StringFormatter(slots=["<|user|>\n{{content}}<|assistant|>"]),
+    format_assistant=StringFormatter(slots=["\n{{content}}"]),
+    format_system=StringFormatter(slots=["<|system|>\n{{content}}"]),
+    format_function=FunctionFormatter(slots=["{{content}}"], tool_format="glm4_moe"),
+    format_observation=StringFormatter(slots=["<|observation|>\n{{content}}<|assistant|>"]),
+    format_tools=ToolFormatter(tool_format="glm4_moe"),
+    format_prefix=EmptyFormatter(slots=["[gMASK]<sop>"]),
+    stop_words=["<|user|>", "<|observation|>"],
+    thought_words=("<think>", "</think>"),
+    efficient_eos=True,
+    template_class=Glm47ReasoningTemplate,
+)
+
+
 # copied from glm4 template
 register_template(
     name="glmz1",

src/llamafactory/extras/constants.py

@@ -939,6 +939,17 @@ register_model_group(
 )
 
 
+register_model_group(
+    models={
+        "GLM-4.7-Flash": {
+            DownloadSource.DEFAULT: "zai-org/GLM-4.7-Flash",
+            DownloadSource.MODELSCOPE: "ZhipuAI/GLM-4.7-Flash",
+        },
+    },
+    template="glm4_7",
+)
+
+
 register_model_group(
     models={
         "GLM-Z1-0414-9B-Chat": {

src/llamafactory/model/model_utils/moe.py

@@ -77,6 +77,11 @@ def add_z3_leaf_module(model: "PreTrainedModel") -> None:
 
         _set_z3_leaf_modules(model, [Glm4MoeMoE])
 
+    if model_type == "glm4_moe_lite":
+        from transformers.models.glm4_moe_lite.modeling_glm4_moe_lite import Glm4MoeLiteMoE
+
+        _set_z3_leaf_modules(model, [Glm4MoeLiteMoE])
+
     if model_type == "glm4v_moe":
         from transformers.models.glm4v_moe.modeling_glm4v_moe import Glm4vMoeTextMoE