Download Latest Version v0.3.22 source code.zip (1.5 MB)
Email in envelope

Get an email when there's a new version of optillm

Home / v0.3.22
Name Modified Size InfoDownloads / Week
Parent folder
README.md 2026-07-18 955 Bytes
0
v0.3.22 source code.tar.gz 2026-07-18 1.3 MB
0
v0.3.22 source code.zip 2026-07-18 1.5 MB
0
Totals: 3 Items   2.8 MB 0

What's Changed

Fix: autothink, thinkdeeper and deepconf for transformers >= 5 (#320)

Since transformers 5, apply_chat_template(return_tensors="pt") returns a BatchEncoding rather than a plain tensor, so passing the result straight to model(input_ids=...) raised. The tensor is now unwrapped via .input_ids:

  • optillm/thinkdeeper.py
  • optillm/autothink/processor.py
  • optillm/deepconf/processor.py

The unwrap is chained with the existing device transfer (.input_ids.to(self.model.device)) so tokens still land on the model's device — Tensor.to() returns a new tensor rather than mutating in place, so the assignment matters on CUDA/MPS.

Thanks to @jacquerie for diagnosing and reporting this.

README: dropped the GitHub stars badge and switched the downloads badge to pepy.tech, which reports actual download counts.

Full Changelog: https://github.com/algorithmicsuperintelligence/optillm/compare/v0.3.21...v0.3.22

Source: README.md, updated 2026-07-18