Issues

• [Perf] Training speed issue for HOPE-mid (very slow step time with 8×A100)

  kmccleary3301/nested_learning#13 · Filter by author Calvert0921 opened on Mar 18, 2026

• [Perf] Loss ~6 from 4k–10k steps on pilot_160m_multi.yaml (custom seq_len=1024, dim=712, batch=2); where are eval + baseline comparisons?

  kmccleary3301/nested_learning#12 · Filter by author nijagalanand1 opened on Feb 16, 2026
  4

• [Eval] Benchmark against NanoGPT hackers

  kmccleary3301/nested_learning#11 · Filter by author TomLucidor opened on Feb 6, 2026
  1

• [Faithfulness] Major fault in gradient propagation

  kmccleary3301/nested_learning#9 · Filter by author weipeilun opened on Dec 27, 2025
  1

• Relations between sequence, segment, batch and chunk

  kmccleary3301/nested_learning#8 · Filter by author embarrassing-noob-questions opened on Dec 25, 2025
  3

• creating artifacts/tokenizer/refinedweb_mix/spm_32000_unigram.model

  kmccleary3301/nested_learning#6 · Filter by author virtustate opened on Nov 29, 2025
  9