D24 v6
Public project index for the complete D24 v6 training lineage.
| Stage | Loadable model | Retained full-state checkpoints |
|---|---|---|
| Pretrain | sfanm/d24-v6-pretrain |
23 (iter_0004000 through iter_0084527) |
| Midtrain | sfanm/d24-v6-midtrain |
12 (iter_0002000 through iter_0023842) |
| SFT | sfanm/d24-v6-sft |
2 (iter_0001600, iter_0001773) |
Lineage: nominal ClimbMix-400B pretraining (354,530,270,862 consumed training tokens) → 100,000,595,968 tokens of replay-free OLMo-3/Dolmino midtraining → no-GSM8K simple-chat SFT (475,114,114 packed tokens).
Each stage repository exposes its terminal BF16 Transformers model at the root
and stores every retained resumable Megatron checkpoint under megatron/.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support