adamw-PT-adamw-MT-a10-g100-mdr
The PT checkpoint after 10,000 steps (20.97B tokens) of mid-training with FineMath-4plus: MT budget 10%, Two-Stage data schedule with final MT ratio 100% (MT-only data from step 9,000), constant LR 5e-4 until step 9,000, then linear decay to 1e-5 (MDR). AdamW throughout. Subset (32M-token) losses at step 10,000: DCLM 2.9726, FineMath 1.5696. Reproduction of arXiv 2610.29348 with Megatron-LM.