import torch

def generate(model, input_ids, generation_config, **kwargs):    
    generation_config = generation_config or model.generation_config  # default to the model generation config
    cur_length = input_ids.shape[1]
    max_length = generation_config.max_length or cur_length + generation_config.max_new_tokens
    
    while cur_length < max_length:
        logits = model(input_ids).logits
        next_token_logits = logits[:, -1, :]
        next_tokens = torch.argmax(next_token_logits, dim=-1)
        input_ids = torch.cat((input_ids, next_tokens[:, None]), dim=-1)
        cur_length += 1

    return input_ids