Fix decoder_attention_mask None handling in generation utils

[damodharg6]

Summary

This PR fixes a potential issue in generation/utils.py where decoder_attention_mask could be accessed before proper validation.

Changes

• Replaced direct dictionary access with:
  model_kwargs.get("decoder_attention_mask", None)
• Added a None check before applying torch.cat(...)

Motivation

This prevents failures during generation/evaluation workflows when decoder_attention_mask is not provided.

[Rocketknight1]

Hey, can you give us some sample code that shows how the issue can be triggered? We get a lot of agent fixes that don't fix actual bugs, so we'd like to see a reproducer!

[damodharg6]

Thanks for reviewing! Here’s a minimal reproducer for the issue: from transformers import AutoTokenizer, AutoModelForCausalLM model = AutoModelForCausalLM.from_pretrained("gpt2") tokenizer = AutoTokenizer.from_pretrained("gpt2") inputs = tokenizer("Hello", return_tensors="pt") outputs = model.generate( **inputs, decoder_attention_mask=None, max_new_tokens=5, ) Before the fix, generation utilities fail when "decoder_attention_mask=None" is propagated internally. With the fix applied, generation works normally without errors.

…

On Fri, 15 May 2026, 6:55 pm Matt, ***@***.***> wrote: *Rocketknight1* left a comment (huggingface/transformers#45985) <#45985 (comment)> Hey, can you give us some sample code that shows how the issue can be triggered? We get a lot of agent fixes that don't fix actual bugs, so we'd like to see a reproducer! — Reply to this email directly, view it on GitHub <#45985 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BRCZIHDEHP4GD4K5B36OWXL424LGNAVCNFSM6AAAAACY62757OVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHM2DINRQGEZTENZQGQ> . Triage notifications on the go with GitHub Mobile for iOS <https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675> or Android <https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub>. You are receiving this because you authored the thread.Message ID: ***@***.***>

[Rocketknight1]

That code snippet runs fine for me.

[damodharg6]

Thanks for checking! I’ll investigate further and see if it’s environment specific on my side.

…

On Fri, May 15, 2026 at 8:12 PM Matt ***@***.***> wrote: *Rocketknight1* left a comment (huggingface/transformers#45985) <#45985 (comment)> That code snippet runs fine for me. — Reply to this email directly, view it on GitHub <#45985 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BRCZIHENZEYCMYIMVLTFNGT424UG7AVCNFSM6AAAAACY62757OVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHM2DINRQG4ZDOMJWGQ> . Triage notifications on the go with GitHub Mobile for iOS <https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675> or Android <https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub>. You are receiving this because you authored the thread.Message ID: ***@***.***>