add prompt tokens, skip_spe for raccoon

Author: wushuo1

lightllm/server/api_tgi.py

@@ -169,6 +169,7 @@ async def stream_results() -> AsyncGenerator[bytes, None]:
                     "logprob": metadata.get("logprob", None),
                     "special": metadata.get("special", False),
                     "count_output_tokens": metadata.get("count_output_tokens", 0),
+                    "prompt_tokens": metadata.get("prompt_tokens", 0),
                 },
                 "generated_text": None,
                 "finished": finish_status.is_finished(),

lightllm/server/core/objs/sampling_params.py

@@ -300,6 +300,9 @@ def init(self, tokenizer, **kwargs):
         self.suggested_dp_index = kwargs.get("suggested_dp_index", -1)
 
         self.skip_special_tokens = kwargs.get("skip_special_tokens", True)
+        SKIP_SPECIAL_TOKENS = os.environ.get("SKIP_SPECIAL_TOKENS", True)
+        self.skip_special_tokens = SKIP_SPECIAL_TOKENS
+        
         self.add_special_tokens = kwargs.get("add_special_tokens", True)
         self.add_spaces_between_special_tokens = kwargs.get("add_spaces_between_special_tokens", True)
         self.print_eos_token = kwargs.get("print_eos_token", False)