add encode for opencompass (#828)

* add encode for opencompass

* doc

* remove **kwargs

docs/en/restful_api.md

@@ -12,7 +12,7 @@ The user can open the http url print by the following command in a browser.
 lmdeploy serve api_server ./workspace --server_name 0.0.0.0 --server_port ${server_port} --instance_num 64 --tp 1
 ```
 
-We provide four restful api in total. Three of them are in OpenAI format.
+We provide some RESTful APIs. Three of them are in OpenAI format.
 
 - /v1/chat/completions
 - /v1/models

docs/zh_cn/restful_api.md

@@ -12,7 +12,7 @@
 lmdeploy serve api_server ./workspace 0.0.0.0 --server_port ${server_port} --instance_num 64 --tp 1
 ```
 
-我们一共提供四个 restful api，其中三个仿照 OpenAI 的形式。
+我们提供的 restful api，其中三个仿照 OpenAI 的形式。
 
 - /v1/chat/completions
 - /v1/models

lmdeploy/serve/openai/api_client.py

@@ -28,6 +28,7 @@ class APIClient:
         self.chat_completions_v1_url = f'{api_server_url}/v1/chat/completions'
         self.completions_v1_url = f'{api_server_url}/v1/completions'
         self.models_v1_url = f'{api_server_url}/v1/models'
+        self.encode_v1_url = f'{api_server_url}/v1/encode'
         self._available_models = None
 
     @property
@@ -43,6 +44,31 @@ class APIClient:
             return self._available_models
         return None
 
+    def encode(self,
+               input: Union[str, List[str]],
+               do_preprocess: Optional[bool] = False,
+               add_bos: Optional[bool] = True):
+        """Encode prompts.
+
+        Args:
+            input: the prompt to be encoded. In str or List[str] format.
+            do_preprocess: whether do preprocess or not. Default to False.
+            add_bos: True when it is the beginning of a conversation. False
+                when it is not. Default to True.
+        Return: (input_ids, length)
+        """
+        headers = {'content-type': 'application/json'}
+        response = requests.post(self.encode_v1_url,
+                                 headers=headers,
+                                 json=dict(input=input,
+                                           do_preprocess=do_preprocess,
+                                           add_bos=add_bos),
+                                 stream=False)
+        if hasattr(response, 'text'):
+            output = json.loads(response.text)
+            return output['input_ids'], output['length']
+        return None, None
+
     def chat_completions_v1(self,
                             model: str,
                             messages: Union[str, List[Dict[str, str]]],

lmdeploy/serve/openai/api_server.py

@@ -18,8 +18,9 @@ from lmdeploy.serve.openai.protocol import (  # noqa: E501
     ChatCompletionStreamResponse, ChatMessage, CompletionRequest,
     CompletionResponse, CompletionResponseChoice,
     CompletionResponseStreamChoice, CompletionStreamResponse, DeltaMessage,
-    EmbeddingsRequest, ErrorResponse, GenerateRequest, GenerateResponse,
-    ModelCard, ModelList, ModelPermission, UsageInfo)
+    EmbeddingsRequest, EncodeRequest, EncodeResponse, ErrorResponse,
+    GenerateRequest, GenerateResponse, ModelCard, ModelList, ModelPermission,
+    UsageInfo)
 
 
 class VariableInterface:
@@ -393,6 +394,37 @@ async def create_embeddings(request: EmbeddingsRequest,
                                  'Unsupported by turbomind.')
 
 
+@app.post('/v1/encode')
+async def encode(request: EncodeRequest, raw_request: Request = None):
+    """Encode prompts.
+
+    The request should be a JSON object with the following fields:
+    - input: the prompt to be encoded. In str or List[str] format.
+    - do_preprocess: whether do preprocess or not. Default to False.
+    - add_bos: True when it is the beginning of a conversation. False when it
+        is not. Default to True.
+    """
+
+    def encode(prompt: str, do_preprocess: bool, add_bos: bool):
+        if do_preprocess:
+            prompt = VariableInterface.async_engine.model.get_prompt(
+                prompt, sequence_start=add_bos)
+        input_ids = VariableInterface.async_engine.tokenizer.encode(
+            prompt, add_bos=add_bos)
+        return input_ids
+
+    if isinstance(request.input, str):
+        encoded = encode(request.input, request.do_preprocess, request.add_bos)
+        return EncodeResponse(input_ids=encoded, length=len(encoded))
+    else:
+        encoded, length = [], []
+        for prompt in request.input:
+            ids = encode(prompt, request.do_preprocess, request.add_bos)
+            encoded.append(ids)
+            length.append(len(ids))
+        return EncodeResponse(input_ids=encoded, length=length)
+
+
 @app.post('/generate',
           tags=['deprecated'],
           description='please use /v1/chat/interactive')

lmdeploy/serve/openai/protocol.py

@@ -191,6 +191,19 @@ class EmbeddingsResponse(BaseModel):
     usage: UsageInfo
 
 
+class EncodeRequest(BaseModel):
+    """Encode request."""
+    input: Union[str, List[str]]
+    do_preprocess: Optional[bool] = False
+    add_bos: Optional[bool] = True
+
+
+class EncodeResponse(BaseModel):
+    """Encode response."""
+    input_ids: Union[List[int], List[List[int]]]
+    length: Union[int, List[int]]
+
+
 class GenerateRequest(BaseModel):
     """Generate request."""
     prompt: Union[str, List[Dict[str, str]]]