Issues

• Questions about communication delays and recovery semantics in prime-diloco

  PrimeIntellect-ai/prime-diloco#266 · Filter by author David2Williams opened on Oct 13, 2026

• send/recv state dict reuse the same tensor

  PrimeIntellect-ai/prime-diloco#245 · Filter by author 3outeille opened on Apr 23, 2025
  2

• nnode value correct?

  PrimeIntellect-ai/prime-diloco#240 · Filter by author HariSeldon11988 opened on Apr 3, 2025

• Running prime with different number of GPUs for different nodes

  PrimeIntellect-ai/prime-diloco#227 · Filter by author HariSeldon11988 opened on Mar 26, 2025

• demo idea for crowdsourcing compute: contribute by simply duplicating a HF space

  PrimeIntellect-ai/prime-diloco#213 · Filter by author rawwerks opened on Feb 6, 2025

• Could prime support up to 70B llama3 model?

  PrimeIntellect-ai/prime-diloco#203 · Filter by author lizdongkun opened on Jan 19, 2025
  1

• Distributed testing

  PrimeIntellect-ai/prime-diloco#173 · Filter by author hellangleZ opened on Dec 7, 2024
  5

• Live checkpoint breaks if node leaves before recovery is complete

  PrimeIntellect-ai/prime-diloco#156 · Filter by author Jackmin801 opened on Nov 12, 2024

• Remove fake all reduce step for onboard

  PrimeIntellect-ai/prime-diloco#118 · Filter by author samsja opened on Oct 17, 2024

• allow to retry when hf dataset fails

  PrimeIntellect-ai/prime-diloco#100 · Filter by author samsja opened on Oct 12, 2024
  1

• Have the nodes ping out their dataloader state before the all-reduce.

  PrimeIntellect-ai/prime-diloco#98 · Filter by author Jackmin801 opened on Oct 11, 2024

• allow to load from diloco ckpt from a different diloco shard

  PrimeIntellect-ai/prime-diloco#91 · Filter by author samsja opened on Oct 10, 2024