origin_inputs size problem for llama,pangu,baichuan models

[kfertakis]

Hi,

When trying to run either llama or pangu or baichuan models, I get the following error regarding the size of origin_inputs and seq_length configuration.

Traceback (most recent call last):
  File "~/mindrlhf/train.py", line 109, in <module>
    run_rlhf(args)
  File "~/mindrlhf/train.py", line 99, in run_rlhf
    trainer.make_experience(num_rollouts=ppo_config.num_rollouts)
  File "~/mindrlhf/mindrlhf/trainer/ppo_trainer.py", line 240, in make_experience
    samples, resposne_array, left_padding_prompt = self.generate(prompt_tensors)
  File "~/mindrlhf/mindrlhf/trainer/ppo_trainer.py", line 190, in generate
    outputs = self.ppo_model.policy_model.model.generate(input_ids_list, max_length=self.ppo_config.seq_length)
  File "~/venv/lib/python3.9/site-packages/mindformers/generation/text_generator.py", line 557, in generate
    output_ids = self._forward(
  File "~/venv/lib/python3.9/site-packages/mindformers/generation/text_generator.py", line 278, in _forward
    input_ids = self._pad_inputs_using_max_length(
  File "~/venv/lib/python3.9/site-packages/mindformers/generation/text_generator.py", line 179, in _pad_inputs_using_max_length
    raise ValueError(
ValueError: origin_inputs size is (1, 11000), you should increase the seq_length of the model 2048.

How should this be addressed? Thanks.