Export
The link always opens the latest version of this page.
Ask an AI about it
Open in ChatGPTA new chat that reads this entry from the site Open in ClaudeThe same, in ClaudeShare the link instead of a file, and they will always see the latest version.
Files are a snapshot of today. The live page has anything newer.
PyTorch Monarch
Distributed GPU training runtime with fault-tolerance via actor model and hierarchical supervision
- Vendor
- Stellar Alpina 1 source
- Category
- distributed training runtime 1 source
PyTorch Monarch is a distributed GPU training runtime designed to enable fault-tolerant execution of machine learning workloads across multiple GPUs.
The system implements fault-tolerance through an actor-model architecture combined with hierarchical supervision 1. This design allows training jobs to recover from node or GPU failures without manual intervention or complete restart.
Monarch supports execution on both CUDA and ROCm platforms 1, enabling deployment across NVIDIA and AMD GPU hardware. The runtime is developed and maintained by Stellar Alpina 2.
The hierarchical supervision model organizes worker processes into a tree structure, where parent actors monitor child actors and coordinate recovery when failures occur. This approach reduces the overhead of centralized fault detection while maintaining consistency guarantees needed for distributed training.
Facts 2
| Attribute | Value | Confidence | Sources | Cited |
|---|---|---|---|---|
| Vendor | Stellar Alpina | low | 1 | 2 |
| Category | distributed training runtime | low | 1 | 1 |
Platform 1
Developed by 1
Official links
Sources 2
- Bringing PyTorch Monarch to AMD GPUs · Hacker News (YC)
- Exclusive: Stellar Alpina Unveils Monarch Orbiter and Morpho Lunar Lander · Payload Space