RuntimeError: The size of tensor a (812) must match the size of tensor b (6300) at non-singleton dimension 0

generated_ids = model.generate(**inputs, max_new_tokens=32,)为什么推理的时候报错，但是添加use_cache=False就可以不报错呢，，下面的报错

outputs_cd = self(
**model_inputs_cd,
return_dict=True,
output_attentions=output_attentions_wo_img,
output_hidden_states=output_hidden_states_wo_img,
)

Start by reproducing the failure from `model.generate(**inputs, max_new_tokens=32)` and compare it with `use_cache=False`. Trace the `self(**model_inputs_cd, ...)` call and inspect the tensor shapes at the point where dimension 0 is combined. Done means identifying the incompatible inputs or cache behavior and documenting a verified correction or minimal reproduction.

Written by the indexing model from the issue text.