Issues

• Metadata-backed packed datasets fingerprint the gzip bytes, so rewriting identical boundaries breaks resume

  allenai/Olmo-core#917 · Filter by author abhishekraok opened on Oct 6, 2026

• MixtureDataLoader continuous stream: sources without get(index, epoch) reuse stale per-source epochs

  allenai/Olmo-core#909 · Filter by author jason718 opened on Oct 5, 2026

• Step-0 checkpoint can advance fresh Adam optimizer state before first update

  allenai/Olmo-core#908 · Filter by author wisoums opened on Oct 4, 2026

• Activation checkpointing re-draws dropout masks on main (#828 only landed on vision, and it does not count attention dropout); attention dropout is also active under model.eval()

  allenai/Olmo-core#905 · Filter by author Mazukiri opened on Oct 2, 2026

• Add Badgr deployment path for OLMo-2-0425-1B-Instruct

  allenai/Olmo-core#904 · Filter by author michaelmanly opened on Oct 2, 2026

• Molmo2 stage 1: make the v2 pointing data the default, as in mm_olmo

  allenai/Olmo-core#892 · Filter by author jason718 opened on Sep 29, 2026

• Upstream multimodal eval infrastructure from olmo-eval: KV-cached attention backend, export conversion, mm_olmo config translation

  allenai/Olmo-core#867 · Filter by author jason718 opened on Sep 16, 2026
  2

• Verify evals on Papers with Code

  allenai/Olmo-core#864 · Filter by author NielsRogge opened on Sep 14, 2026

• Free staged state dict during async checkpointer callbacks

  allenai/Olmo-core#856 · Filter by author ouillie opened on Sep 4, 2026

• [Bug] DionConfig incorrectly scales LM head learning rate for AdamW + Proposal for Dion3 integration

  allenai/Olmo-core#854 · Filter by author JenWei0312 opened on Sep 4, 2026
  1

• BeakerCallback fails without the [beaker] extra: olmo_core.launch.beaker imports gantry at module top

  allenai/Olmo-core#850 · Filter by author jason718 opened on Sep 2, 2026

• Molmo2 tokenizer chat template rejects system turns (roles must alternate user/assistant)

  allenai/Olmo-core#849 · Filter by author jason718 opened on Sep 2, 2026

• vision: compile_vision/compile_connector backward hits an inductor stride AssertionError when crop counts vary across batches

  allenai/Olmo-core#848 · Filter by author jason718 opened on Sep 2, 2026
  3

• vision: FlexAttention in eager mode (compile_model=false) OOMs at seq 16384 — Molmo2-Stage2.py smoke recipe

  allenai/Olmo-core#847 · Filter by author jason718 opened on Sep 2, 2026

• vision: Molmo2-Stage2.py smoke recipe OOMs on the step-0 pre-train checkpoint with HF init

  allenai/Olmo-core#846 · Filter by author jason718 opened on Sep 2, 2026
  2

• Drop OLMoE-1B-7B support

  allenai/Olmo-core#838 · Filter by author abhishekraok opened on Aug 19, 2026
  2

• FSDP2 post-backward reduce-scatter hangs on GDN hybrid models (Olmo-Hybrid-7B), 4/4 runs

  allenai/Olmo-core#829 · Filter by author abhishekraok opened on Aug 14, 2026
  2

• Missing all OLMo-3-1025-7B Stage 1 (pretraining) Olmo-core checkpoints

  allenai/Olmo-core#729 · Filter by author muchanem opened on Jun 16, 2026
  2

• CPU thread leaking

  allenai/Olmo-core#703 · Filter by author rwood-97 opened on Jun 4, 2026
  1

• newer transformers library versions break olmo3, fix provided in a PR

  allenai/Olmo-core#685 · Filter by author nurpax opened on May 26, 2026

• Question: Current thinking on Muon/Dion optimizer support for OLMo-core training runs

  allenai/Olmo-core#683 · Filter by author JenWei0312 opened on May 19, 2026

• olmo-hybrid pretraining data are not available

  allenai/Olmo-core#678 · Filter by author HydrogenSulfate opened on May 14, 2026
  2

• RFC: Engram-style Sparse Memory Retrieval in OLMo-core

  allenai/Olmo-core#673 · Filter by author JenWei0312 opened on May 12, 2026
  1

• Track progress on tilelang integration

  allenai/Olmo-core#668 · Filter by author AkshitaB opened on May 1, 2026

• Why is intra-document masking only applied for long-context training and not midtraining?

  allenai/Olmo-core#662 · Filter by author BrownianNotion opened on Apr 29, 2026
  2