[example] Update Llama Inference example (#5629)

* [example] add infernece benchmark llama3

* revise inference config - arg

* remove unused args

* add llama generation demo script

* fix init rope in llama policy

* add benchmark-llama3 - cleanup

colossalai/inference/modeling/policy/nopadding_llama.py

@@ -100,5 +100,5 @@ class NoPaddingLlamaModelInferPolicy(LlamaForCausalLMPolicy):
         return policy
 
     def postprocess(self):
-        init_to_get_rotary(self.model.model)
+        init_to_get_rotary(self.model.model, self.model.config.rope_theta)
         return self.model