ZeroLock decouples LLM fine-tuning into independently updated chunks using local objectives and a pipeline system, achieving moderate memory and throughput gains, but its convergence proof targets a surrogate objective rather than the true global loss.
PipeDream: Generalized pipeline parallelism for DNN training,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.LG 1years
2026 1verdicts
CONDITIONAL 1roles
background 1polarities
background 1representative citing papers
citing papers explorer
-
ZeroLock: Concurrent Memory-Efficient LLM Training via Modular Update Decoupling
ZeroLock decouples LLM fine-tuning into independently updated chunks using local objectives and a pipeline system, achieving moderate memory and throughput gains, but its convergence proof targets a surrogate objective rather than the true global loss.