Issues

• Desync at Cooldown->Finished transition in local testnet (0,0,1 tiers)

  plugyawn/aftermath#11 · Filter by author plugyawn opened on Jan 15, 2026

• Add MatFormer manifest for tier-sliced checkpoints with auto-fetch and strict validation

  plugyawn/aftermath#10 · Filter by author plugyawn opened on Jan 15, 2026
  

• Add per-layer gradient norm logging

  plugyawn/aftermath#9 · Filter by author plugyawn opened on Jan 15, 2026

• Disallow mlp_bias with MatFormer tiers (assert)

  plugyawn/aftermath#8 · Filter by author plugyawn opened on Jan 15, 2026

• Block MatFormer helper mode (suffix sampling) until rotation/sparse alignment is wired

  plugyawn/aftermath#7 · Filter by author plugyawn opened on Jan 15, 2026

• Fix MatFormer tier metadata when saving checkpoints

  plugyawn/aftermath#6 · Filter by author plugyawn opened on Jan 15, 2026

• Standardize MatFormer tier metadata in exported checkpoints

  plugyawn/aftermath#5 · Filter by author plugyawn opened on Jan 15, 2026

• Wire helper-mode sparse indices into DisTrO aggregation

  plugyawn/aftermath#4 · Filter by author plugyawn opened on Jan 15, 2026

• Auto-load tiered checkpoints for MatFormer tiers (remote + local)

  plugyawn/aftermath#3 · Filter by author plugyawn opened on Jan 15, 2026

• Add tier-aware scheduling/weighting for heterogeneous training

  plugyawn/aftermath#2 · Filter by author plugyawn opened on Jan 15, 2026
  1

• Support MatFormer tiering with tensor parallelism

  plugyawn/aftermath#1 · Filter by author plugyawn opened on Jan 15, 2026
  1