-
Fix CUDA OOM on T4: cast model to amp_dtype directly, batch_size=2, safe RMSNorm
-
Fix CUDA OOM on T4: cast model to amp_dtype directly, batch_size=2, safe RMSNorm
-
feat: wipe-proof HF persistence (full ckpt + live log + newest-2 cleanup)
-
docs: DDP verify RUN#3 FULL PASS — ready for 50k pretrain
-
fix: exclude non-5col schema paths from HF stream (wikipedia/grammar/gsm8k CastError on shuffle)
-
fix: probe OOM reserve 4GB + expandable_segments + natural-mode stream shuffle (Hindi-only trap)
-
docs & scripts: add dedicated grammar stream milestone (145.93GB/105.1B tokens)
-
docs & scripts: add full-scale 5-domain milestone (145.93GB/105B tokens)
-
init: 28L MoE 241M audit-fixed, smoke pass
-
initial commit