pipeline() error message incorrectly suggests translation tasks are supported in v5

[math-hiyoko]

System Info

Copy-and-paste the text below in your GitHub issue and FILL OUT the two last points.

• transformers version: 5.2.0.dev0
• Platform: Linux-6.6.105+-x86_64-with-glibc2.35
• Python version: 3.12.12
• Huggingface_hub version: 1.3.7
• Safetensors version: 0.7.0
• Accelerate version: 1.12.0
• Accelerate config: not found
• DeepSpeed version: not installed
• PyTorch version (accelerator?): 2.9.0+cpu (NA)
• Using distributed or parallel set-up in script?: parallel

Who can help?

@Rocketknight1

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

When calling transformers.pipeline() with an unknown task, the error message lists translation_XX_to_YY as an available task. However, in Transformers v5, the default pipeline() does not provide the "translation" pipeline in supported_tasks, so attempting to use translation_XX_to_YY leads to a KeyError: 'translation'.

Also, calling pipeline("translation") raises:
KeyError: "Invalid translation task translation, use 'translation_XX_to_YY' format"
but since "translation" is not actually supported, the expected error should be closer to:
KeyError: "Unknown task translation, available tasks are ..."

Colab link: https://colab.research.google.com/drive/1aKTGHuuDhpdxAJlJB8txOmcyAKeNoz2i?usp=sharing

Expected behavior

KeyError: "Unknown task hoge, available tasks are ['any-to-any', 'audio-classification', 'automatic-speech-recognition', 'depth-estimation', 'document-question-answering', 'feature-extraction', 'fill-mask', 'image-classification', 'image-feature-extraction', 'image-segmentation', 'image-text-to-text', 'image-to-image', 'keypoint-matching', 'mask-generation', 'ner', 'object-detection', 'question-answering', 'sentiment-analysis', 'table-question-answering', 'text-classification', 'text-generation', 'text-to-audio', 'text-to-speech', 'token-classification', 'video-classification', 'visual-question-answering', 'vqa', 'zero-shot-audio-classification', 'zero-shot-classification', 'zero-shot-image-classification', 'zero-shot-object-detection']

[Rocketknight1]

The translation pipeline has been removed in v5! This is because the quality of translations from Seq2Seq models was generally worse than using LLMs instead, and we didn't want pipelines to mislead users. We probably need to remove all the leftover translation references in the pipeline files entirely.

[math-hiyoko]

@Rocketknight1

Thanks for the clarification, I understand that the translation pipeline itself is removed in v5.

The actual issue here is:

• pipeline() still lists translation_XX_to_YY tasks in the "available tasks" message, which makes it look like translation is supported.

pipeline('hoge')

KeyError: "Unknown task hoge, available tasks are ['any-to-any', 'audio-classification', 'automatic-speech-recognition', 'depth-estimation', 'document-question-answering', 'feature-extraction', 'fill-mask', 'image-classification', 'image-feature-extraction', 'image-segmentation', 'image-text-to-text', 'image-to-image', 'keypoint-matching', 'mask-generation', 'ner', 'object-detection', 'question-answering', 'sentiment-analysis', 'table-question-answering', 'text-classification', 'text-generation', 'text-to-audio', 'text-to-speech', 'token-classification', 'video-classification', 'visual-question-answering', 'vqa', 'zero-shot-audio-classification', 'zero-shot-classification', 'zero-shot-image-classification', 'zero-shot-object-detection', 'translation_XX_to_YY']"

• When explicitly passing translation_XX_to_YY, the resulting error message is misleading.

pipeline('translation_en_to_fr')

---------------------------------------------------------------------------
KeyError                                  Traceback (most recent call last)
/tmp/ipython-input-124266547.py in <cell line: 0>()
----> 1 pipeline('translation_en_to_fr')

2 frames
/usr/local/lib/python3.12/dist-packages/transformers/pipelines/base.py in check_task(self, task)
   1349             tokens = task.split("_")
   1350             if len(tokens) == 4 and tokens[0] == "translation" and tokens[2] == "to":
-> 1351                 targeted_task = self.supported_tasks["translation"]
   1352                 task = "translation"
   1353                 return task, targeted_task, (tokens[1], tokens[3])

KeyError: 'translation'

PR #43826 addresses this by fixing the pipeline error message and preventing the misleading KeyError, which resolves the core confusion.

[Rocketknight1]

Yes, this cleanup is good, but there are a few other areas that need fixing too! I made a PR at #43869 which includes your PR at #43826 as well as some other changes. Can you take a look and tell me if there's anything else you think I missed?

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.

Please note that issues that do not follow the contributing guidelines are likely to be ignored.