fix: fix meta instruction

modeling_internlm.py

@@ -777,13 +777,15 @@ class InternLMForCausalLM(InternLMPreTrainedModel):
             reordered_past += (tuple(past_state.index_select(0, beam_idx) for past_state in layer_past),)
         return reordered_past
 
-    def build_inputs(self, tokenizer, query: str, history: List[Tuple[str, str]] = []):
+    def build_inputs(self, tokenizer, query: str, history: List[Tuple[str, str]] = [], meta_instruction=""):
         prompt = ""
-        for record in history:
-            prompt += f"""<s><|User|>:{record[0]}<eoh>\n<|Bot|>:{record[1]}<eoa>\n"""
-        if len(prompt) == 0:
+        if meta_instruction:
+            prompt += f"""<s><|System|>:{meta_instruction}\n"""
+        else:
             prompt += "<s>"
-        prompt += f"""<|User|>:{query}<eoh>\n<|Bot|>:"""
+        for record in history:
+            prompt += f"""<|User|>:{record[0]}\n<|Bot|>:{record[1]}<eoa>\n"""
+        prompt += f"""<|User|>:{query}\n<|Bot|>:"""
         return tokenizer([prompt], return_tensors="pt")
 
     @torch.no_grad()
@@ -797,9 +799,12 @@ class InternLMForCausalLM(InternLMPreTrainedModel):
         do_sample: bool = True,
         temperature: float = 0.8,
         top_p: float = 0.8,
+        meta_instruction: str = "You are an AI assistant whose name is InternLM (书生·浦语).\n"
+"- InternLM (书生·浦语) is a conversational language model that is developed by Shanghai AI Laboratory (上海人工智能实验室). It is designed to be helpful, honest, and harmless.\n"
+"- InternLM (书生·浦语) can understand and communicate fluently in the language chosen by the user such as English and 中文.",
         **kwargs,
     ):
-        inputs = self.build_inputs(tokenizer, query, history)
+        inputs = self.build_inputs(tokenizer, query, history, meta_instruction)
         inputs = {k: v.to(self.device) for k, v in inputs.items() if torch.is_tensor(v)}
         outputs = self.generate(
             **inputs,