Fix: forced_decoder_ids causes error on HF Inference Endpoints

#1
by nishantkumar - opened

Issue

When deploying this model to HuggingFace Inference Endpoints, inference fails with:

Error 400: "You have explicitly specified forced_decoder_ids. Please remove the forced_decoder_ids
argument in favour of input_ids or decoder_input_ids respectively."

Cause

The generation_config.json contains forced_decoder_ids which is deprecated in newer versions of
the transformers library used by Inference Endpoints.

Suggested Fix

In generation_config.json, change:

"forced_decoder_ids": [
  [
    1,
    null
  ],
  [
    2,
    50360
  ]
],                                

To:

  "forced_decoder_ids": null                                                                           

This will allow the model to work with HF Inference Endpoints while maintaining the same behavior.

Inference Error log:
image

aapoorv75 changed discussion status to closed

Sign up or log in to comment