Deprecation warning: "Passing a tuple of `past_key_values` is deprecated"

[corneliusroemer]

Bug report checklist

• [x] I provided code that demonstrates a minimal reproducible example.
• [x] I confirmed bug exists on the latest mainline of Chronos via source install.

Describe the bug When running the example code from readme using amazon/chronos-bolt-base I get:

Passing a tuple of `past_key_values` is deprecated and will be removed in Transformers v4.48.0. You should pass an instance of `EncoderDecoderCache` instead, e.g. `past_key_values=EncoderDecoderCache.from_legacy_cache(past_key_values)`.

Expected behavior No deprecation warning

To reproduce

pipeline: BaseChronosPipeline = BaseChronosPipeline.from_pretrained(
    "amazon/chronos-bolt-base",  # use "amazon/chronos-bolt-small" for the corresponding Chronos-Bolt model
    device_map="cpu",  # use "cpu" for CPU inference
    torch_dtype=torch.bfloat16,
)

quantiles, mean = pipeline.predict_quantiles(
    context=torch.tensor(df["#Passengers"][:-12]),
    prediction_length=12,
    quantile_levels=[0.1, 0.5, 0.9],
)

Environment description Operating system: macOS Python version: 3.12 CUDA version: NA PyTorch version: HuggingFace transformers version: HuggingFace accelerate version:

[lostella]

Hi @corneliusroemer, thanks for opening the issue. The warning is generated here https://github.com/huggingface/transformers/blob/69bc848480d5f19a537a70ce14f09816b00cd80f/src/transformers/models/t5/modeling_t5.py#L1024 which appears not to be checking for the None case (last elif clause never applies). I’m wondering if this is the real issue

[lostella]

A similar fix was applied here to other models https://github.com/huggingface/transformers/pull/33541 maybe all is needed is a similar one for T5

[trzysiek]

I'll try to fix within in a week if no one else wants to.

[abdulfatir]

@trzysiek are you actually facing problems/errors due to this? If yes, can you share an example?

[egvillalobos]

Hi @abdulfatir , I see this was closed, but I can confirm I'm still seeing this deprecation warning. It makes using the model in production very uncomfortable because it creates a warning every time the model runs, which is very noisy for our logs.