app.py

import gradio as gr
import spaces
import torch

from peft import PeftConfig, PeftModel
from transformers import LlamaForCausalLM, AutoTokenizer, BitsAndBytesConfig

config = PeftConfig.from_pretrained("GGmorello/FLAMES-20k")
model = LlamaForCausalLM.from_pretrained(
    'codellama/codellama-7b-hf',
    quantization_config=BitsAndBytesConfig(
        load_in_4bit=True,
        bnb_4bit_quant_type="nf4",
        bnb_4bit_use_double_quant=True,
        bnb_4bit_compute_dtype=torch.bfloat16,
    ),
)
model = PeftModel.from_pretrained(model, "GGmorello/FLAMES-20k")


MAX_SEQ_LEN = 4096
tokenizer = AutoTokenizer.from_pretrained('codellama/codellama-7b-hf')
model.config.pad_token = tokenizer.pad_token = tokenizer.unk_token



zero = torch.Tensor([0]).cuda()
print(zero.device) # <-- 'cpu' 🤔

@spaces.GPU
def predict(text):
    model.to(zero.device)
    input_ids = tokenizer(text, return_tensors='pt')["input_ids"]
    input_ids.to(zero.device)
    generated_ids = model.generate(input_ids, max_new_tokens=256)
    filling = tokenizer.batch_decode(generated_ids[:, input_ids.shape[1]:], skip_special_tokens = True)[0]
    return filling


demo = gr.Interface(fn=predict, inputs=gr.Text(), outputs=gr.Text())
demo.launch()