[Bugfix] Fix GLM rotary_dim issue and support v1 (vllm-project#16912)

Signed-off-by: isotr0py <2037008807@qq.com>
Signed-off-by: Agata Dobrzyniewicz <adobrzyniewicz@habana.ai>

Author: Isotr0py

vllm/model_executor/models/glm.py

@@ -3,21 +3,20 @@
 from vllm.config import VllmConfig
 from vllm.model_executor.models.llama import LlamaForCausalLM
 
-from .interfaces import SupportsV0Only
 from .utils import PPMissingLayer
 
 
-class GlmForCausalLM(LlamaForCausalLM, SupportsV0Only):
+class GlmForCausalLM(LlamaForCausalLM):
 
     def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
+        vllm_config.model_config.hf_config.partial_rotary_factor = 0.5
         super().__init__(vllm_config=vllm_config, prefix=prefix)
         # Hack Llama model to fit HF format GLM implementation
         # Attention difference between GLM and Llama:
         # 1. Half partial rotary_dim and no Neox style.
         # 2. There is no bias for o_proj in attention
         for layer in self.model.layers:
             if not isinstance(layer, PPMissingLayer):
-                layer.self_attn.rotary_emb.rotary_dim //= 2
                 layer.self_attn.rotary_emb.is_neox_style = False
                 layer.self_attn.o_proj.bias = None
                 layer.self_attn.o_proj.skip_bias_add = True