Add embeddings api

openai_api.py

@@ -12,12 +12,12 @@
 from fastapi.middleware.cors import CORSMiddleware
 from contextlib import asynccontextmanager
 from typing import Any, Dict, List, Literal, Optional, Union
-from transformers import AutoTokenizer, AutoModel
+from transformers import AutoTokenizer, AutoModel, BertModel, BertTokenizer
 from sse_starlette.sse import ServerSentEvent, EventSourceResponse
 
 
 @asynccontextmanager
-async def lifespan(app: FastAPI): # collects GPU memory
+async def lifespan(app: FastAPI):  # collects GPU memory
     yield
     if torch.cuda.is_available():
         torch.cuda.empty_cache()
@@ -34,6 +34,7 @@ async def lifespan(app: FastAPI): # collects GPU memory
     allow_headers=["*"],
 )
 
+
 class ModelCard(BaseModel):
     id: str
     object: str = "model"
@@ -109,8 +110,8 @@ async def create_chat_completion(request: ChatCompletionRequest):
     history = []
     if len(prev_messages) % 2 == 0:
         for i in range(0, len(prev_messages), 2):
-            if prev_messages[i].role == "user" and prev_messages[i+1].role == "assistant":
-                history.append([prev_messages[i].content, prev_messages[i+1].content])
+            if prev_messages[i].role == "user" and prev_messages[i + 1].role == "assistant":
+                history.append([prev_messages[i].content, prev_messages[i + 1].content])
 
     if request.stream:
         generate = predict(query, history, request.model)
@@ -154,7 +155,6 @@ async def predict(query: str, history: List[List[str]], model_id: str):
         chunk = ChatCompletionResponse(model=model_id, choices=[choice_data], object="chat.completion.chunk")
         yield "{}".format(chunk.json(exclude_unset=True, ensure_ascii=False))
 
-
     choice_data = ChatCompletionResponseStreamChoice(
         index=0,
         delta=DeltaMessage(),
@@ -165,6 +165,43 @@ async def predict(query: str, history: List[List[str]], model_id: str):
     yield '[DONE]'
 
 
+class EmbeddingsRequest(BaseModel):
+    model: str
+    input: Union[str, List[str]]
+
+
+class EmbeddingsData(BaseModel):
+    object: str = "embedding"
+    embedding: List[float]
+    index: int = 0
+
+
+class EmbeddingsResponse(BaseModel):
+    object: str = "list"
+    data: List[EmbeddingsData]
+    model: str
+    usage: dict
+
+
+@app.post("/v1/embeddings", response_model=EmbeddingsResponse)
+async def create_chat_completion(request: EmbeddingsRequest):
+    global bert_model, bert_tokenizer
+    encoded_input = bert_tokenizer(request.input, padding=True, truncation=True, return_tensors='pt')
+    with torch.no_grad():
+        model_output = bert_model(**encoded_input)
+    sentence_embeddings = mean_pooling(model_output, encoded_input['attention_mask'])
+    return EmbeddingsResponse(
+        data=[EmbeddingsData(embedding=_.tolist(), index=i) for i, _ in enumerate(sentence_embeddings)],
+        model=request.model,
+        usage={},
+    )
+
+
+def mean_pooling(model_output, attention_mask):
+    token_embeddings = model_output[0]
+    input_mask_expanded = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
+    return torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(input_mask_expanded.sum(1), min=1e-9)
+
 
 if __name__ == "__main__":
     tokenizer = AutoTokenizer.from_pretrained("THUDM/chatglm2-6b", trust_remote_code=True)
@@ -173,5 +210,8 @@ async def predict(query: str, history: List[List[str]], model_id: str):
     # from utils import load_model_on_gpus
     # model = load_model_on_gpus("THUDM/chatglm2-6b", num_gpus=2)
     model.eval()
-
+    bert_name = 'GanymedeNil/text2vec-large-chinese'
+    bert_model = BertModel.from_pretrained(bert_name)
+    bert_tokenizer = BertTokenizer.from_pretrained(bert_name, model_max_length=512)
     uvicorn.run(app, host='0.0.0.0', port=8000, workers=1)
+