File size: 1,735 Bytes
ac47d9a f82f890 ac47d9a f82f890 ac47d9a 0ab7ce5 ac47d9a 0ab7ce5 ac47d9a |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 |
from fastapi import FastAPI
from pydantic import BaseModel
from transformers import T5Tokenizer, T5ForConditionalGeneration
from sentence_transformers import SentenceTransformer
from pinecone import Pinecone
device = 'cpu'
# Initialize Pinecone instance
pc = Pinecone(api_key='your-pinecone-api-key')
# Initialize FastAPI app
app = FastAPI()
# Initialize the models
def load_models():
retriever = SentenceTransformer("flax-sentence-embeddings/all_datasets_v3_mpnet-base")
tokenizer = T5Tokenizer.from_pretrained('t5-small')
generator = T5ForConditionalGeneration.from_pretrained('t5-base').to(device)
return retriever, generator, tokenizer
retriever, generator, tokenizer = load_models()
class QueryInput(BaseModel):
input: str
@app.post("/predict")
def predict(query: QueryInput):
query_text = query.input
# Query Pinecone
xq = retriever.encode([query_text]).tolist()
xc = index.query(vector=xq, top_k=1, include_metadata=True)
if 'matches' in xc and isinstance(xc['matches'], list):
context = [m['metadata']['Output'] for m in xc['matches']]
context_str = " ".join(context)
formatted_query = f"answer the question: {query_text} context: {context_str}"
else:
context_str = ""
formatted_query = f"answer the question: {query_text} context: {context_str}"
# Generate answer using T5 model
inputs = tokenizer.encode(formatted_query, return_tensors="pt", max_length=512, truncation=True).to(device)
ids = generator.generate(inputs, num_beams=2, min_length=10, max_length=60, repetition_penalty=1.2)
answer = tokenizer.decode(ids[0], skip_special_tokens=True, clean_up_tokenization_spaces=False)
return {"response": answer}
|