🚨 Remove DoLa decoding strategy #40082
Conversation
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
There was a problem hiding this comment.
Looks mostly good to me, added two minor nits :) I've also checked the repo itself, and I'm quite happy with it!
👉 Sanity-check question: if you add trust_remote_code=True on the mixin test (test_dola_decoding_sample), do the tests still pass? Let's confirm this 🤗 (I see you did it in the integration tests 👍 )
👉 merging: we'll need to move the repo under transformers-community before merging, let's take care of it tomorrow!
manueldeprada
changed the title
|
@manueldeprada happy with the new test! Feel free to merge 🤗 (I've also added DoLa to the collection in |
|
[For maintainers] Suggested jobs to run (before merge) run-slow: aria, csm, deepseek_v3, dia, gemma, gemma3n, git, granitemoe, granitemoeshared, llama, mistral, qwen2_5_omni, recurrent_gemma |
3 participants
Removes Decoding by Contrasting Layers (DoLa) generation strategy from the codebase. Directs users to the
transformers-community/dolarepository.It has been a warning for a few releases, but now
trust_remote_code=Trueis required to run DoLa generation.