🚀🚀 「大模型」2小时完全从0训练64M的小参数GPT！🌏 Train a 64M-parameter GPT from scratch in just 2h!
COMMITS
January 6, 2026
J
[update] readme
jingyaogong committed
January 5, 2026
J
[update] simplify loader
jingyaogong committed
J
[update] readme
jingyaogong committed
J
[update] rename train tokenizer
jingyaogong committed
J
[update] readme
jingyaogong committed
J
[update] rename reason
jingyaogong committed
January 4, 2026
J
[fix] messages num
jingyaogong committed
January 2, 2026
J
[fix] dist cleanup
jingyaogong committed
January 1, 2026
J
[update] aux loss
jingyaogong committed
December 31, 2025
J
[fix] experts unused
jingyaogong committed
J
[fix] layers set 8
jingyaogong committed
J
[fix] moe unused
jingyaogong committed
J
[feat] get params
jingyaogong committed
J
[feat] get params
jingyaogong committed
J
[feat] update config
jingyaogong committed
J
[feat] update lr
jingyaogong committed
J
[feat] compatible tokenizer
jingyaogong committed
December 28, 2025
J
[feat] stream load data
jingyaogong committed
December 26, 2025
J
[feat] remove empty_cache
jingyaogong committed
December 23, 2025
J
[feat] explicit left padding
jingyaogong committed
December 22, 2025
J
[fix] lora weight
jingyaogong committed
J
Merge pull request #594 from whiteswordLI/fix/lora-load-ddp-weights
jingyaogong committed
W
Fix: support loading DDP-saved LoRA weights for inference
whitesword committed
December 14, 2025
J
[feat] adjust seq length
jingyaogong committed
December 11, 2025
J
[feat] update readme
jingyaogong committed
December 9, 2025
J
[fix] dtype & lr
jingyaogong committed
J
Merge pull request #571 from dyhuachi/dyhuachi-patch-1
jingyaogong committed
December 6, 2025
D
[fix] Refactor get_lr function to include min_lr calculation
dyhuachi committed
December 5, 2025
J
[fix] reduce aux_loss_alpha
jingyaogong committed
December 1, 2025
J
[fix] cuda memory #559
jingyaogong committed