app.py

from fastapi import FastAPI, Request, Form
from fastapi.responses import HTMLResponse
from fastapi.staticfiles import StaticFiles
from fastapi.templating import Jinja2Templates
import os
import torch

# =============================
# MODEL SETTINGS
# =============================
from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel

MODEL_ID = "DSDUDEd/Cass_beta1.0"

# Use writable cache directories
CACHE_DIR = "/tmp/hf_cache"
os.makedirs(CACHE_DIR, exist_ok=True)
os.environ["TRANSFORMERS_CACHE"] = CACHE_DIR
os.environ["HF_HOME"] = CACHE_DIR
os.environ["HF_DATASETS_CACHE"] = CACHE_DIR
os.environ["HF_METRICS_CACHE"] = CACHE_DIR
os.environ["TORCH_HOME"] = CACHE_DIR

# =============================
# FASTAPI APP INIT
# =============================
app = FastAPI()
app.mount("/static", StaticFiles(directory="."), name="static")
templates = Jinja2Templates(directory=".")

# =============================
# LOAD TOKENIZER + MODEL
# =============================
print("🚀 Loading Cass tokenizer and model...")
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, cache_dir=CACHE_DIR)

# Load base model first, then PEFT adapter
base_model = AutoModelForCausalLM.from_pretrained(
    MODEL_ID,
    cache_dir=CACHE_DIR,
    torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
    device_map="auto" if torch.cuda.is_available() else None,
    low_cpu_mem_usage=True
)
model = PeftModel.from_pretrained(base_model, MODEL_ID, cache_dir=CACHE_DIR)

device = "cuda" if torch.cuda.is_available() else "cpu"
model.to(device)
print(f"✅ Model loaded successfully on {device.upper()}")

# =============================
# CHAT HISTORY
# =============================
history = []

# =============================
# ROUTES
# =============================
@app.get("/", response_class=HTMLResponse)
async def get_chat(request: Request):
    return templates.TemplateResponse("index.html", {"request": request, "history": history})

@app.post("/", response_class=HTMLResponse)
async def post_chat(request: Request, user_input: str = Form(...)):
    history.append({"role": "user", "content": user_input})

    inputs = tokenizer(user_input, return_tensors="pt").to(device)
    outputs = model.generate(
        **inputs,
        max_length=250,
        pad_token_id=tokenizer.eos_token_id
    )
    ai_response = tokenizer.decode(outputs[0], skip_special_tokens=True)

    history.append({"role": "ai", "content": ai_response})
    return templates.TemplateResponse("index.html", {"request": request, "history": history})

# =============================
# RUN LOCAL (optional)
# =============================
if __name__ == "__main__":
    import uvicorn
    uvicorn.run(app, host="0.0.0.0", port=7860)