Metraipolis › archive › dc_def81fa07ed3

NeMo-DCR: Bit-Exact Delta-Compressed Refit for Scalable Agentic RL at Trillion-Parameter Scale — arXiv:2610.08430

Source: arXiv (en)
Original: https://arxiv.org/abs/2610.08430
License (as archived): CC0-1.0
Archived: 2026-10-07T11:21:46.838Z
SHA-256 of the archived text: 2bdd22ac87bbd4c5e6d4cfe67ac82c34812c039343de75dc38729b2e17e30411
Stable URL of this page: https://metraipolis.com/documento/dc_def81fa07ed3

How to cite

NeMo-DCR: Bit-Exact Delta-Compressed Refit for Scalable Agentic RL at Trillion-Parameter Scale — arXiv:2610.08430 — arXiv (en), https://arxiv.org/abs/2610.08430 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_def81fa07ed3 (Metraipolis archive dc_def81fa07ed3, sha256 2bdd22ac87bbd4c5e6d4cfe67ac82c34812c039343de75dc38729b2e17e30411)

BibTeX

@misc{dcdef81fa07ed3,
  title = {NeMo-DCR: Bit-Exact Delta-Compressed Refit for Scalable Agentic RL at Trillion-Parameter Scale — arXiv:2610.08430},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.08430},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_def81fa07ed3; sha256 2bdd22ac87bbd4c5e6d4cfe67ac82c34812c039343de75dc38729b2e17e30411},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.08430. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Agentic reinforcement learning (RL) disaggregates training from rollout, so each policy update must reach the rollout clusters before the next batch. Transferring a full 1T checkpoint for such weight synchronization (refit) takes 87.5 min between two AWS regions. Measurements of BF16 training show that about 1% of weights change their stored values per step. Recent systems exploit this sparsity but fall short on placement, exactness, or efficiency: they reimplement placement rules, assemble full tensors, rebuild values arithmetically, or use a cross-cluster collective, and none fully recovers from mid-refit failures. We present NeMo-DCR (Delta-Compressed Refit), which sends only changes yet is bit-exact: receivers obtain the same parameter and buffer bits as a dense refit. For placement, fixed affine mappings project changes from training shards into the checkpoint's canonical coordinates, residual conversion covers the other changes, and the serving runtime's native loader places all changes in receiver storage. For exactness, compressible XOR masks carry affine changes whose projection and loader preserve stored bits, and overwrites carry the others. Receivers apply both in place, retries overwrite partial writes, and a joint commit binds the policy to the baseline for the next delta. For efficiency, object storage or a relay tree streams payloads during delta construction, without a cross-cluster collective. Even at 3% and 5% change rates, NeMo-DCR refits of 30B-1T models are 12-40$\times$ faster than a transport-only full-checkpoint reference. A 1T relay-tree refit at 3% takes 150 s instead of 87.5 min, making refits practical for cross-cluster agentic RL at trillion-parameter scale.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno).
Categorie arXiv: Distributed, Parallel, and Cluster Computing (cs.DC) ; Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2610.08430 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-07T11:15:24.827Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).