8 papers
Agora: Collective and Permissionless Internet-Scale Pretraining of Large Language Models
Gil Avraham, Violetta Shevchenko, Hadi Mohaghegh Dolatabadi +9
Agora is a system that enables large language models to be trained collectively across many heterogeneous, internet‑connected GPUs, using pipeline‑parallel sharding and fault‑toler…
Factored Gossip DiLoCo: Reducing Blocking Communication in DiLoCo
Chamin Hewa Koneputugodage, Thalaiyasingam Ajanthan, Sameera Ramasinghe +7
To make large-scale distributed training practical outside high-bandwidth datacenters, we must reduce blocking, high-volume synchronization. While DiLoCo communicates infrequently,…
Taming Curvature: Architecture Warm-Up for Stable Transformer Training
Sameera Ramasinghe, Ajanthan Thalaiyasingam, Hadi Mohaghegh Dolatabadi +6
Training billion-parameter Transformers is often brittle, with transient loss spikes and divergence that waste compute. Even though the recently developed Edge of Stability (EoS) t…
Mixtures of Subspaces for Bandwidth Efficient Context Parallel Training
Sameera Ramasinghe, Ajanthan Thalaiyasingam, Hadi Mohaghegh Dolatabadi +5
Pretraining language models with extended context windows enhances their ability to leverage rich information during generation. Existing methods split input sequences into chunks,…
Unextractable Protocol Models: Collaborative Training and Inference without Weight Materialization
Alexander Long, Chamin Hewa Koneputugodage, Thalaiyasingam Ajanthan +5
We consider a decentralized setup in which the participants collaboratively train and serve a large neural network, and where each participant only processes a subset of the model.…
NuMuon: Nuclear-Norm-Constrained Muon for Compressible LLM Training
Hadi Mohaghegh Dolatabadi, Thalaiyasingam Ajanthan, Sameera Ramasinghe +7
The rapid progress of large language models (LLMs) is increasingly constrained by memory and deployment costs, motivating compression methods for practical deployment. Many state-o…