File size: 2,089 Bytes
4283167 50a25ae |
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 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 |
from fastapi import FastAPI
from pydantic import BaseModel
from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel
import torch
app = FastAPI()
# === MODEL ===
MODEL_REPO = "sahil239/falcon-lora-chatbot" # replace with your HF repo
BASE_MODEL = "tiiuae/falcon-rw-1b"
# === Load tokenizer ===
tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL, trust_remote_code=True)
tokenizer.pad_token = tokenizer.eos_token # required to avoid padding error
# === Load base model and merge LoRA ===
base_model = AutoModelForCausalLM.from_pretrained(BASE_MODEL, trust_remote_code=True)
model = PeftModel.from_pretrained(base_model, MODEL_REPO)
model.eval()
# === Move to GPU if available ===
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)
# === Request Schema ===
class PromptRequest(BaseModel):
prompt: str
max_new_tokens: int = 200
temperature: float = 0.7
top_p: float = 0.95
@app.get("/")
def health_check():
return {"status": "running"}
@app.post("/generate")
async def generate_text(req: PromptRequest):
inputs = tokenizer(
req.prompt,
return_tensors="pt",
padding=True,
truncation=True,
max_length=200
)
inputs = {k: v.to(device) for k, v in inputs.items()}
with torch.no_grad():
outputs = model.generate(
input_ids=inputs["input_ids"],
attention_mask=inputs["attention_mask"],
max_new_tokens=req.max_new_tokens,
temperature=req.temperature,
top_p=req.top_p,
do_sample=True,
pad_token_id=tokenizer.eos_token_id,
eos_token_id=tokenizer.eos_token_id, # 🚨 Helps stop when sentence is "done"
repetition_penalty=1.2, # 🚫 Penalizes repeating phrases
no_repeat_ngram_size=3
)
generated_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
return {"response": generated_text[len(req.prompt):].strip()}
if __name__ == "__main__":
uvicorn.run(app, host="0.0.0.0", port=7860) |