Repair effective-batch normalization in a frozen Transformers label-smoothing path and produce a…
Repair effective-batch normalization in a frozen Transformers label-smoothing path and produce a…: a task in MiMo-V2.6-RL-oss: Agentic RL Environments (MiMo RL release). Repair effective-batch normalization A deterministic language-model training reproduction in…
The task
Repair effective-batch normalization A deterministic language-model training reproduction in /app/vendor/transformers/tools/effective_batch_case.py exposes a regression in the existing Trainer label-smoothing path. With gradient accumulation, microbatches can contain different…
Part of XiaomiMiMo/MiMo-V2.6-RL-oss.