transformers - 💡(How to fix) Fix [Bug] Incorrect error message for T5 model family's decoder input validation [1 pull requests]

System Info

• transformers version: 5.5.1
• Platform: Windows-11-10.0.26200-SP0
• Python version: 3.12.12
• Huggingface_hub version: 1.8.0
• Safetensors version: 0.7.0
• Accelerate version: 1.13.0
• Accelerate config: not found
• DeepSpeed version: not installed
• PyTorch version (accelerator?): 2.6.0+cu124 (CUDA)
• Using distributed or parallel set-up in script?: no
• Using GPU in script?: yes
• GPU type: NVIDIA GeForce RTX 3060 Ti

Who can help?

@ArthurZucker @Cyrilvallez

Information

• The official example scripts
• My own modified scripts

Tasks

• An officially supported task in the examples folder (such as GLUE/SQuAD, ...)
• My own task or dataset (give details below)

Reproduction

Steps to reproduce

1. Import and load a T5 model from transformers:

from transformers import T5ForConditionalGeneration

model = T5ForConditionalGeneration.from_pretrained("t5-small")

1. Extract the decoder:

t5_decoder = model.decoder

1. Case 1: Call decoder with both input_ids and inputs_embeds:

import torch

input_ids = torch.tensor([[1, 2, 3]])
inputs_embeds = torch.randn(1, 3, model.config.d_model)

t5_decoder(input_ids=input_ids, inputs_embeds=inputs_embeds)

1. Case 2: Call decoder with neither input_ids nor inputs_embeds:

t5_decoder()

Observed behavior

The error message is misleading and references incorrect argument names.

Example traceback:

Traceback (most recent call last):
  File "repro.py", line XX, in <module>
    t5_decoder(input_ids=input_ids, inputs_embeds=inputs_embeds)
  File ".../transformers/models/t5/modeling_t5.py", line XXX, in forward
    ...
ValueError: You cannot specify both decoder_input_ids and decoder_inputs_embeds at the same time

Note:

• The decoder was called with input_ids and inputs_embeds
• But the error refers to decoder_input_ids and decoder_inputs_embeds

Colab notebook

https://colab.research.google.com/drive/1eggqqICSgiKltsDoY_86iH9D6M8zHc8R?authuser=0#scrollTo=9Ehdgpq7S1p8

Expected behavior

Expected behavior

The error message should reference the actual arguments passed to the decoder, for example :

ValueError: You cannot specify both decoder's input_ids and decoder's inputs_embeds at the same time

Possible fix

In T5-related models, the error message prefix is currently defined as:

err_msg_prefix = "decoder_" if self.is_decoder else ""
raise ValueError(
    f"You cannot specify both {err_msg_prefix}input_ids and {err_msg_prefix}inputs_embeds at the same time"
)

This results in misleading messages such as:

ValueError: You cannot specify both decoder_input_ids and decoder_inputs_embeds at the same time

even when the decoder is directly called with input_ids and inputs_embeds.

A possible improvement is to make the message clearer by adjusting the prefix:

err_msg_prefix = "decoder's " if self.is_decoder else ""
raise ValueError(
    f"You cannot specify both {err_msg_prefix}input_ids and {err_msg_prefix}inputs_embeds at the same time"
)

This would produce:

ValueError: You cannot specify both decoder's input_ids and decoder's inputs_embeds at the same time