app.py

from fastapi import FastAPI
from pydantic import BaseModel
from transformers import (
    pipeline,
    AutoTokenizer,
    AutoModelForCausalLM,
)
from langchain_huggingface import HuggingFacePipeline
from langchain_core.prompts import PromptTemplate
from langchain_core.runnables import RunnableSequence

# — Model setup —
MODEL_ID = "bigcode/starcoder2-3b"
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, trust_remote_code=True)
# Explicitly set pad_token_id to eos_token_id
tokenizer.pad_token_id = tokenizer.eos_token_id
model = AutoModelForCausalLM.from_pretrained(MODEL_ID, trust_remote_code=True)

# — Pipeline setup (remove unused parameters, set device explicitly) —
pipe = pipeline(
    "text-generation",
    model=model,
    tokenizer=tokenizer,
    device=-1,  # Explicitly use CPU; change to 0 or "cuda" if GPU available
    max_new_tokens=64,
    do_sample=False,
)
llm = HuggingFacePipeline(pipeline=pipe)

# — Prompt & chain (using RunnableSequence) —
prompt = PromptTemplate(
    input_variables=["description"],
    template=(
        "### Convert English description to an Emmet abbreviation\n"
        "Description: {description}\n"
        "Emmet:"
    ),
)
chain = RunnableSequence(prompt | llm)

# — FastAPI app —
app = FastAPI()

class Req(BaseModel):
    description: str

class Res(BaseModel):
    emmet: str

@app.get("/")
async def root():
    return {"message": "Welcome to the Emmet Generator API. Use POST /generate-emmet."}

@app.post("/generate-emmet", response_model=Res)
async def generate_emmet(req: Req):
    raw = chain.invoke(req.description)
    emmet = raw.strip().splitlines()[0]
    return {"emmet": emmet}