executing-plans
executing-plans
Use when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available
Explorer les Skills
Productivité · SKILL.md
serialization-and-distribution
serialization-and-distributionSerialize/export Sonnet modules and reason about TensorFlow distribution, mixed precision, and backend limits.
Independent GitHub snapshot · Automated classification · Original documentation governs
Documentation de la Skill
GitHubname: serialization-and-distribution description: "Serialize/export Sonnet modules and reason about TensorFlow distribution, mixed precision, and backend limits." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0
Serialization and Distribution
Use this sub-skill for TensorFlow checkpoints, SavedModel export/load, pickle or Keras interoperability caveats, XLA, mixed precision, and snt.distribute helpers.
Start here
- references/serialization.md: checkpoint, SavedModel, Keras/pickle, XLA, and mixed precision recipes.
- references/distribution-and-backends.md:
Replicator,TpuReplicator,CrossReplicaBatchNorm, and backend limits. - references/troubleshooting.md: restored object identity, SavedModel signatures, cross-replica context, and accelerator-runtime failures.
- scripts/serialization_smoke.py: CPU checkpoint and SavedModel smoke.
Boundaries
- Module construction and lazy variables: ../module-authoring/SKILL.md.
- Training-loop checkpoint timing: ../training-and-optimization/SKILL.md.
- Functional transforms and device helpers: ../functional-transforms/SKILL.md.
Backend policy
CPU verifies checkpoints, SavedModel export/load, pickle caveats, and mixed precision policy mechanics. CUDA, TPU, XLA performance, and distributed replicas require a matching TensorFlow runtime. Do not claim accelerator verification from a CPU-only run.