update version

requirements-rocm.txt

@@ -15,5 +15,5 @@ setuptools_scm>=8
 torch == 2.4.1
 triton == 3.0.0
 flash_attn == 2.6.1
-lmslim == 0.2.0  
+lmslim == 0.3.0  
 numa

setup.py

@@ -488,10 +488,10 @@ def get_version_add(sha: Optional[str] = None) -> str:
             if sha is None:
                 sha = get_sha(vllm_root)
             if (major, minor) == ('2', '4'):
-                version = 'das.opt1.' + sha[:7]
+                version = 'das.opt1.cust1.' + sha[:7]
     else:
         if (major, minor) == ('2', '4'):
-            version = 'das.opt1'
+            version = 'das.opt1.cust1'
 
 
     # dtk version

vllm/model_executor/model_loader/utils.py

@@ -79,7 +79,8 @@ def get_model_architecture(
         model_config: ModelConfig) -> Tuple[Type[nn.Module], str]:
     architectures = getattr(model_config.hf_config, "architectures", [])
     visions = getattr(model_config.hf_config, "visual", []) or getattr(model_config.hf_config, "vision_config", [])
-    support_nn_architectures = ['LlamaForCausalLM', 'QWenLMHeadModel', 'Qwen2ForCausalLM', 'Qwen2VLForConditionalGeneration', 'Qwen2_5_VLForConditionalGeneration', 'Qwen2MoeForCausalLM', 'ChatGLMModel', 'ChatGLMForConditionalGeneration', 
+    # TODO: support deepseek distillation series models ( 'LlamaForCausalLM', 'Qwen2ForCausalLM' )
+    support_nn_architectures = ['QWenLMHeadModel', 'Qwen2VLForConditionalGeneration', 'Qwen2_5_VLForConditionalGeneration', 'Qwen2MoeForCausalLM', 'ChatGLMModel', 'ChatGLMForConditionalGeneration', 
                                 'BaichuanForCausalLM', 'BloomForCausalLM', 'MedusaModel', 'MixtralForCausalLM', 'MLPSpeculatorPreTrainedModel', 'FalconForCausalLM', 'DeepseekV2ForCausalLM', 'DeepseekV3ForCausalLM']  
     if any(arch in architectures for arch in support_nn_architectures): 
         if os.getenv('LLAMA_NN') != '0':