Issues

• Proposal: LoRA model handler and a canary-exposure metric — would PRs be welcome?

  locuslab/open-unlearning#202 · Filter by author RAFAELDCOELHO opened on Oct 5, 2026

• [TOFU Reproduction] Released Llama-3.2-1B-Instruct NPO forget10 checkpoint does not reproduce documented metrics

  bug
  locuslab/open-unlearning#199 · Filter by author asad073-ui opened on Aug 8, 2026

• Provenance and license clarification for the TOFU Llama-3.2-1B RMU forget10 checkpoint

  locuslab/open-unlearning#198 · Filter by author yangkaibo0609-source opened on Aug 6, 2026

• Tokenize the question (+ special tokens) and the response separately, then merge into chat_ids

  locuslab/open-unlearning#195 · Filter by author ummagumm-a opened on Jun 18, 2026

• Optimize Adam; Remove dead config; Make flash-attn optional

  locuslab/open-unlearning#190 · Filter by author filyp opened on May 12, 2026

• Evaluate during training

  locuslab/open-unlearning#187 · Filter by author chengjiali opened on Apr 6, 2026

• PEFT LoRA transformer layers not trained when gradient_checkpointing=True — enable_input_require_grads() missing

  bug
  locuslab/open-unlearning#178 · Filter by author Anya-wUw opened on Mar 5, 2026
  2

• Script Batch Size

  locuslab/open-unlearning#164 · Filter by author BaekJunehong opened on Dec 15, 2025
  1

• Adding Support for Hubble Suite

  locuslab/open-unlearning#148 · Filter by author aflah02 opened on Oct 8, 2025
  3

• Significant difference in results with different GPU counts on WMDP bio dataset using GradDiff

  bug
  locuslab/open-unlearning#144 · Filter by author ZeguanXiao opened on Sep 26, 2025
  1

• eos_token as pad_token Causes Incorrect Attention Masking in Multi-Turn Dialogue

  locuslab/open-unlearning#128 · Filter by author Chennz1 opened on Jul 2, 2025
  2

• SimNPO implementation question

  bug
  locuslab/open-unlearning#123 · Filter by author Anya-wUw opened on Jun 21, 2025
  3