Description
System Info
transformersversion: 4.39.3- Platform: macOS-13.4-arm64-arm-64bit
- Python version: 3.10.13
- Huggingface_hub version: 0.22.2
- Safetensors version: 0.4.3
- Accelerate version: 0.29.2
- Accelerate config: not found
- PyTorch version (GPU?): 2.2.2 (False)
- Tensorflow version (GPU?): 2.16.1 (False)
- Flax version (CPU?/GPU?/TPU?): not installed (NA)
- Jax version: not installed
- JaxLib version: not installed
- Using GPU in script?: no
- Using distributed or parallel set-up in script?: no
Who can help?
No response
Information
- The official example scripts
- My own modified scripts
Tasks
- An officially supported task in the
examplesfolder (such as GLUE/SQuAD, ...) - My own task or dataset (give details below)
Reproduction
Running this example prints a warning for every loop, despite there being no padding.
from transformers import AutoTokenizer, GPT2Model, GPT2ForSequenceClassification
model = GPT2ForSequenceClassification.from_pretrained("gpt2")
assert isinstance(model, GPT2ForSequenceClassification)
model.config.pad_token_id = model.config.eos_token_id
tokenizer = AutoTokenizer.from_pretrained("gpt2")
text = ["Hello, my dog is cute."]
inp = tokenizer(text, return_tensors="pt")
embeds = model.get_input_embeddings()(inp["input_ids"])
for i in range(3):
out = model(inputs_embeds=embeds)Output:
GPT2ForSequenceClassification will not detect padding tokens in `inputs_embeds`. Results may be unexpected if using padding tokens in conjunction with `inputs_embeds.`
GPT2ForSequenceClassification will not detect padding tokens in `inputs_embeds`. Results may be unexpected if using padding tokens in conjunction with `inputs_embeds.`
GPT2ForSequenceClassification will not detect padding tokens in `inputs_embeds`. Results may be unexpected if using padding tokens in conjunction with `inputs_embeds.`
Expected behavior
The warning is printed once, or at least there is some way to disable the warning.
I was split between whether this should be a bug report or a feature request. It makes sense to display this warning, but in my project I need to run on embeddings often and the warning really spams the logs.
For a while I was only running batches of size 1 so I was suppressing the warning by temporarily setting model.config.pad_token_id = None. The problem with this is that I then can't run batches of size >1 even if I'm careful to make them the same length with no padding tokens.
I'm not sure of the best way to handle this, but either using the warnings library to make it only print once and/or allow it to be suppressed would help, or having some flag to disable the warning.
The earliest instance of the string will not detect padding tokens in in the codebase I could find was from #7501.