mixed-precision-trainer

Mixed Precision Trainer - Auto-activating skill for ML Training. Triggers on: mixed precision trainer, mixed precision trainer Part of the ML Training skill category.

1.07x

Quality

Does it follow best practices?

Impact

96%

1.07x

Average score across 3 eval scenarios

Securityby

Passed

No known issues

Optimize this skill with Tessl

npx tessl skill review --optimize ./planned-skills/generated/07-ml-training/mixed-precision-trainer/SKILL.md

Evaluation results

98%

Accelerating Model Training at Frostline Analytics

PyTorch AMP training loop best practices

Criteria

Without context

With context

GradScaler used

100%

autocast context manager

100%

Scaled loss backward

100%

scaler.step() used

100%

scaler.update() called

100%

float16 or bfloat16 dtype

100%

Step ordering correct

70%

80%

CPU fallback

100%

Per-epoch metrics reported

100%

Production-ready structure

100%

Without context: $0.4697 · 3s · 1 turns · 3 in / 38 out tokens

With context: $0.7392 · 3s · 1 turns · 3 in / 86 out tokens

95%

Optimizing Deep Learning Infrastructure at Meridian Health Tech

TensorFlow mixed precision configuration

Criteria

Without context

With context

Mixed precision policy set

100%

Policy set before model build

100%

LossScaleOptimizer used

75%

100%

float32 output layer

100%

TensorFlow 2.x API

100%

Per-epoch metrics

100%

Policy mentioned in notes

100%

Memory benefit explained

100%

No large dataset download

100%

Compute dtype vs variable dtype

100%

Production-ready structure

57%

28%

Without context: $0.4173 · 2m 18s · 21 turns · 21 in / 5,404 out tokens

With context: $0.9752 · 2s · 1 turns · 3 in / 49 out tokens

97%

19%

End-to-End Training Pipeline for Arclight Research

End-to-end mixed precision training pipeline

Criteria

Without context

With context

AMP in pipeline

83%

GradScaler in pipeline

90%

Data preparation step

100%

Multiple hyperparameter configs

100%

experiments.json structure

100%

Per-epoch metrics logged

100%

Best config identified

100%

CPU fallback

100%

Validation step

100%

Production code structure

100%

No large files left

100%

Hyperparams in experiments.json

100%

Without context: $0.5132 · 1s · 1 turns · 3 in / 21 out tokens

With context: $0.7917 · 1s · 1 turns · 3 in / 25 out tokens

Repository: jeremylongshore/claude-code-plugins-plus-skills
Commit: f17dd51

Evaluated: 18 days ago
Agent: Claude Code
Model: Claude Sonnet 4.6

Table of Contents

Accelerating Model Training at Frostline Analytics Optimizing Deep Learning Infrastructure at Meridian Health Tech End-to-End Training Pipeline for Arclight Research

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.