fix the offset during streaming chat (#142)

lmdeploy/model.py

@@ -113,7 +113,7 @@ conversation"""  # noqa: E501
 
     def get_prompt(self, prompt, sequence_start=True):
         if sequence_start:
-            return f'<bos>{self.system}\n' \
+            return f'<BOS>{self.system}\n' \
                    f'{self.user}:{prompt}{self.eoh}\n' \
                    f'{self.assistant}:'
         else:

lmdeploy/serve/client.py

@@ -37,7 +37,7 @@ def main(tritonserver_addr: str, model_name: str, session_id: int = 1):
             chatbot.end(session_id)
         else:
             request_id = f'{session_id}-{nth_round}'
-            for status, res, tokens in chatbot.stream_infer(
+            for status, res, n_token in chatbot.stream_infer(
                     session_id,
                     prompt,
                     request_id=request_id,

lmdeploy/serve/turbomind/chatbot.py

@@ -381,13 +381,16 @@ class Chatbot:
                                           request_output_len, sequence_start,
                                           sequence_end, preseq_length, cancel))
         producer.start()
-        for state, res, tokens in self.stream_consumer(
-                self.postprocess, que, session, preseq_length, cancel, logger,
-                self.display, self.profile_generation, self.eos_id):
+        for state, res, tokens in self.stream_consumer(self.postprocess, que,
+                                                       session, input_tokens,
+                                                       preseq_length, cancel,
+                                                       logger, self.display,
+                                                       self.profile_generation,
+                                                       self.eos_id):
             if state.value < 0:
                 yield state, res, 0
             else:
-                yield state, res, tokens - input_tokens
+                yield state, res, tokens
         producer.join()
         self._session = que.get()
         curseq_length = self._session.sequence_length
@@ -477,8 +480,9 @@ class Chatbot:
         que.put(None)
 
     @staticmethod
-    def stream_consumer(postprocess, res_queue, session, preseq_length, cancel,
-                        logger, display, profile_generation, eos_id):
+    def stream_consumer(postprocess, res_queue, session, n_input_token,
+                        preseq_length, cancel, logger, display,
+                        profile_generation, eos_id):
         """Consume the response from the triton inference server.
 
         Args:
@@ -486,6 +490,7 @@ class Chatbot:
                 the generated tokens
             res_queue (multiprocessing.Queue): response queue
             session (Session): an instance of a session
+            n_input_token (int): token number of input prompt
             preseq_length (int): the history sequence length
             cancel (bool): indicator for cancelling the session
             logger (util.Logger):
@@ -496,12 +501,12 @@ class Chatbot:
         Yields:
             tuple: status, text, generated token number
         """
+        offset = n_input_token + preseq_length
         while True:
             result = res_queue.get()
             if result is None:
-                yield StatusCode.TRITON_STREAM_END, \
-                      session.response[len(session.prompt):], \
-                      session.sequence_length - preseq_length
+                yield (StatusCode.TRITON_STREAM_END, session.response,
+                       session.sequence_length - offset)
                 session.status = StatusCode.TRITON_STREAM_END
                 break
             if 'errcode' in result:
@@ -521,7 +526,7 @@ class Chatbot:
                 output_ids = result.as_numpy('output_ids')
 
                 session.sequence_length = sequence_length.squeeze()
-                sequence_length = sequence_length - preseq_length
+                sequence_length = sequence_length - offset
                 last_token_id = output_ids[-1][-1][session.sequence_length - 1]
                 if last_token_id == eos_id:
                     session.sequence_length = session.sequence_length - 1
@@ -536,23 +541,15 @@ class Chatbot:
                            'postprocessing is ignored during profiling '
                            'token generation', sequence_length.squeeze())
                     continue
-                output_str = postprocess(output_ids[:, :, preseq_length:],
+                output_str = postprocess(output_ids[:, :, offset:],
                                          sequence_length)
                 text = output_str[0].decode()
                 if display:
-                    if len(text) > len(session.prompt):
-                        if session.status == StatusCode.TRITON_SESSION_READY:
-                            new_text = text[len(session.prompt):]
-                            session.status = StatusCode.TRITON_STREAM_ING
-                        else:
-                            new_text = text[len(session.response):]
-                        print(new_text, end='', flush=True)
+                    new_text = text[len(session.response):]
+                    print(new_text, end='', flush=True)
                 session.response = text
-                if len(session.response) > len(session.prompt):
-                    session.status = StatusCode.TRITON_STREAM_ING
-                    yield (StatusCode.TRITON_STREAM_ING,
-                           session.response[len(session.prompt):],
-                           sequence_length.squeeze())
+                yield (StatusCode.TRITON_STREAM_ING, session.response,
+                       sequence_length.squeeze())
             except Exception as e:
                 logger.error(f'catch exception: {e}')