DiLoCo Distributed Training
Follow the data, decisions, feedback, and validation boundaries before writing the full system.
How to read this diagram
Read left to right for the forward path: raw information becomes a representation, passes through the project’s main computational ideas, and produces an output that can be measured. Then follow the feedback path back toward the trainable or decision-making components.
Inner/outer optimizers
Inner/outer optimizers defines one of the project’s main information transformations. Understand its input representation, objective, numerical invariants, computational cost, and failure modes before relying on a library implementation.
Boundary check: document its accepted input, output shape, mutable state, failure modes, and the metric that proves this stage is correct before connecting it downstream.
Pseudo-gradients
Pseudo-gradients defines one of the project’s main information transformations. Understand its input representation, objective, numerical invariants, computational cost, and failure modes before relying on a library implementation.
Boundary check: document its accepted input, output shape, mutable state, failure modes, and the metric that proves this stage is correct before connecting it downstream.
Non-IID sharding
Non-IID data means clients or shards follow different distributions. It tests whether aggregation remains stable when local gradients disagree and whether reported global accuracy hides poor subgroup behavior.
Boundary check: document its accepted input, output shape, mutable state, failure modes, and the metric that proves this stage is correct before connecting it downstream.
Architecture review checklist
- Every arrow has a documented shape, dtype, unit, or schema.
- Training and evaluation paths cannot leak information into each other.
- Randomness is seeded and captured in experiment metadata.
- Expensive stages expose timing, memory, throughput, and error metrics.
- Each feedback loop has a stop condition and a rollback strategy.
- Small reference implementations exist for numerical comparisons.