feat(pytorch): add glm5.3-flash support

lmdeploy/archs.py

@@ -97,6 +97,7 @@ def check_vl_llm(backend: str, config: dict) -> bool:
         'Gemma3ForConditionalGeneration', 'Llama4ForConditionalGeneration', 'InternVLForConditionalGeneration',
         'InternS1ForConditionalGeneration', 'InternS1ProForConditionalGeneration',
         'InternS1_1_ForConditionalGeneration', 'Glm4vForConditionalGeneration',
+        'Glm5NextForConditionalGeneration',
         'InternS2MobiusForConditionalGeneration', 'InternS2MobiusForCausalLM',
         'InternS2PreviewForConditionalGeneration', 'InternS2PreviewForCausalLM',
         'KimiK25ForConditionalGeneration', 'Kimi_K25ForConditionalGeneration',

lmdeploy/hf_configs/__init__.py

@@ -11,7 +11,10 @@
 @lru_cache
 def register_config(model_type: str):
     """Register an LMDeploy-owned Transformers config when available."""
-    if model_type == 'kimi_k2':
+    if model_type == 'glm5_next':
+        from .configuration_glm5_next import Glm5NextConfig
+        AutoConfig.register(Glm5NextConfig.model_type, Glm5NextConfig)
+    elif model_type == 'kimi_k2':
         # Standalone Kimi EAGLE checkpoints do not provide an auto_map.
         from .configuration_kimi_k2 import KimiK2Config
         AutoConfig.register(KimiK2Config.model_type, KimiK2Config)