app.py

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig

quant_config = BitsAndBytesConfig(load_in_8bit=True, llm_int8_skip_modules=["temporal_block"])
tokenizer = AutoTokenizer.from_pretrained("alpindale/recurrentgemma-9b-it")
model = AutoModelForCausalLM.from_pretrained(
    "alpindale/recurrentgemma-9b-it",
    device_map="auto", torch_dtype=torch.float16,
    quantization_config=quant_config
)

model.push_to_hub("recurrentgemma-9b-it-8bit")