sushruthsam's picture
Update main.py
6478b34 verified
raw
history blame
644 Bytes
from fastapi import FastAPI
from pydantic import BaseModel
from ctransformers import AutoModelForCausalLM # Ensure correct import for your model
# Model loading
llm = AutoModelForCausalLM.from_pretrained("sqlcoder-34b-alpha.Q4_K_S.gguf",
max_tokens = 1024, threads = 3)
# Pydantic object for request validation
class Validation(BaseModel):
prompt: str
# Initialize FastAPI app
app = FastAPI()
# Endpoint for SQL query generation
@app.post("/generate_sql")
async def generate_sql(item: Validation):
# Use the provided prompt directly
completion = llm(item.prompt)
return completion