fix permission issue for cache, and remove pooling

app.py

@@ -1,20 +1,27 @@
-from fastapi import FastAPI
+from fastapi import FastAPI, Request
 from pydantic import BaseModel
-from sentence_transformers import SentenceTransformer
+from transformers import AutoTokenizer, AutoModel
+import torch
 
 app = FastAPI()
 
-# Dùng bản hỗ trợ sentence-transformers
-model = SentenceTransformer("BAAI/bge-m3-v2")
+# Load model
+model_name = "BAAI/bge-m3"
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+model = AutoModel.from_pretrained(model_name)
 
 class InputText(BaseModel):
     text: str
 
-@app.post("/embed")
-def embed_text(data: InputText):
-    vector = model.encode(data.text, normalize_embeddings=True)
-    return {"embedding": vector.tolist()}
-    
 @app.get("/")
-def read_root():
-    return {"message": "BAAI/bge-m3 Sentence Embedding API is running."}
+def root():
+    return {"message": "BAAI/bge-m3 embedding API is running."}
+
+@app.post("/embed")
+def get_embedding(data: InputText):
+    inputs = tokenizer(data.text, return_tensors="pt", padding=True, truncation=True)
+    with torch.no_grad():
+        outputs = model(**inputs)
+        # Get CLS token or use pooling method
+        embedding = outputs.last_hidden_state[:, 0, :].squeeze().tolist()
+    return {"embedding": embedding}

requirements.txt

@@ -1,3 +1,4 @@
-sentence-transformers
+transformers==4.41.0
+torch
 fastapi
 uvicorn