[Feature] add prompt_tokens and completion_tokens (#3504)

fastdeploy/entrypoints/openai/protocol.py

@@ -139,6 +139,8 @@ class ChatMessage(BaseModel):
     completion_token_ids: Optional[List[int]] = None
     text_after_process: Optional[str] = None
     raw_prediction: Optional[str] = None
+    prompt_tokens: Optional[str] = None
+    completion_tokens: Optional[str] = None
 
 
 class ChatCompletionResponseChoice(BaseModel):
@@ -198,6 +200,8 @@ class DeltaMessage(BaseModel):
     tool_calls: Optional[List[DeltaToolCall | ToolCall]] = None
     text_after_process: Optional[str] = None
     raw_prediction: Optional[str] = None
+    prompt_tokens: Optional[str] = None
+    completion_tokens: Optional[str] = None
 
 
 class ChatCompletionResponseStreamChoice(BaseModel):
@@ -236,6 +240,8 @@ class CompletionResponseChoice(BaseModel):
     completion_token_ids: Optional[List[int]] = None
     text_after_process: Optional[str] = None
     raw_prediction: Optional[str] = None
+    prompt_tokens: Optional[str] = None
+    completion_tokens: Optional[str] = None
     arrival_time: Optional[float] = None
     logprobs: Optional[CompletionLogprobs] = None
     reasoning_content: Optional[str] = None
@@ -280,6 +286,8 @@ class CompletionResponseStreamChoice(BaseModel):
     completion_token_ids: Optional[List[int]] = None
     text_after_process: Optional[str] = None
     raw_prediction: Optional[str] = None
+    prompt_tokens: Optional[str] = None
+    completion_tokens: Optional[str] = None
     reasoning_content: Optional[str] = None
     finish_reason: Optional[Literal["stop", "length", "tool_calls"]] = None
     tool_calls: Optional[List[DeltaToolCall | ToolCall]] = None

fastdeploy/entrypoints/openai/serving_chat.py

@@ -237,6 +237,7 @@ class OpenAIServingChat:
                             if request.return_token_ids:
                                 choice.delta.prompt_token_ids = list(prompt_token_ids)
                                 choice.delta.text_after_process = text_after_process
+                                choice.delta.prompt_tokens = text_after_process
                             chunk = ChatCompletionStreamResponse(
                                 id=request_id,
                                 object=chunk_object_type,
@@ -308,6 +309,7 @@ class OpenAIServingChat:
                     if request.return_token_ids:
                         choice.delta.completion_token_ids = list(output["token_ids"])
                         choice.delta.raw_prediction = output.get("raw_prediction")
+                        choice.delta.completion_tokens = output.get("raw_prediction")
                     if include_continuous_usage:
                         chunk.usage = UsageInfo(
                             prompt_tokens=num_prompt_tokens,
@@ -442,7 +444,9 @@ class OpenAIServingChat:
             prompt_token_ids=prompt_token_ids if request.return_token_ids else None,
             completion_token_ids=completion_token_ids if request.return_token_ids else None,
             text_after_process=text_after_process if request.return_token_ids else None,
+            prompt_tokens=text_after_process if request.return_token_ids else None,
             raw_prediction=output.get("raw_prediction") if request.return_token_ids else None,
+            completion_tokens=output.get("raw_prediction") if request.return_token_ids else None,
         )
         logprobs_full_res = None
         if logprob_contents:

fastdeploy/entrypoints/openai/serving_completion.py

@@ -343,6 +343,7 @@ class OpenAIServingCompletion:
                                         text="",
                                         prompt_token_ids=list(prompt_batched_token_ids[idx]),
                                         text_after_process=text_after_process_list[idx],
+                                        prompt_tokens=text_after_process_list[idx],
                                         completion_token_ids=None,
                                     )
                                 ],
@@ -393,6 +394,7 @@ class OpenAIServingCompletion:
                             completion_token_ids=output.get("token_ids") if request.return_token_ids else None,
                             tool_calls=None,
                             raw_prediction=output.get("raw_prediction") if request.return_token_ids else None,
+                            completion_tokens=output.get("raw_prediction") if request.return_token_ids else None,
                             reasoning_content=output.get("reasoning_content"),
                             arrival_time=arrival_time,
                             logprobs=logprobs_res,
@@ -511,7 +513,9 @@ class OpenAIServingCompletion:
                 prompt_token_ids=prompt_token_ids if request.return_token_ids else None,
                 completion_token_ids=completion_token_ids if request.return_token_ids else None,
                 raw_prediction=output.get("raw_prediction") if request.return_token_ids else None,
+                completion_tokens=output.get("raw_prediction") if request.return_token_ids else None,
                 text_after_process=text_after_process_list[idx] if request.return_token_ids else None,
+                prompt_tokens=text_after_process_list[idx] if request.return_token_ids else None,
                 reasoning_content=output.get("reasoning_content"),
                 tool_calls=output.get("tool_call"),
                 logprobs=aggregated_logprobs,