Add GLM-4 TextGeneration Model support for SGLang (#1736)

README.md

@@ -303,6 +303,7 @@ You can view the full example [here](https://github.com/sgl-project/sglang/tree/
 - MiniCPM / MiniCPM 3
 - XVERSE / XVERSE MoE
 - SmolLM
+- GLM-4
 
 **Embedding Models**
 

python/sglang/srt/models/chatglm.py

@@ -303,7 +303,7 @@ class GLMTransformer(nn.Module):
         return hidden_states
 
 
-class ChatGLMModel(nn.Module):
+class ChatGLMM(nn.Module):
     def __init__(
         self,
         config,
@@ -366,7 +366,7 @@ class ChatGLMForCausalLM(nn.Module):
         self.config: ChatGLMConfig = config
         self.quant_config = quant_config
         self.max_position_embeddings = getattr(config, "max_sequence_length", 8192)
-        self.transformer = ChatGLMModel(config, cache_config, quant_config)
+        self.transformer = ChatGLMM(config, cache_config, quant_config)
         self.lm_head = self.transformer.output_layer
         self.logits_processor = LogitsProcessor(config)
 
@@ -401,4 +401,4 @@ class ChatGLMModel(ChatGLMForCausalLM):
     pass
 
 
-EntryClass = [ChatGLMForCausalLM, ChatGLMModel]
+EntryClass = [ChatGLMModel]

test/srt/models/test_generation_models.py

@@ -57,6 +57,7 @@ ALL_OTHER_MODELS = [
     ModelCase("Qwen/Qwen2.5-14B-Instruct"),
     ModelCase("HuggingFaceTB/SmolLM-135M-Instruct", skip_long_prompt=True),
     ModelCase("allenai/OLMo-1B-0724-hf", decode_tolerance=8e-2, skip_long_prompt=True),
+    ModelCase("THUDM/glm-4-9b-chat"),
 ]
 
 TORCH_DTYPES = [torch.float16]