deepspeed-zero3
Partitioning weights, gradients, and optimizer states.
SKILL.mdskills/training-tuning/deepspeed-zero3/SKILL.md
GitHubDeepspeed Zero3
Overview
Partitioning weights, gradients, and optimizer states.
When to Use
- Standard workflow tasks related to deepspeed zero3.
When NOT to Use
- Out-of-scope scenarios.
Prompt Template
Execute the deepspeed-zero3 process targeting the parameters below.
Parameters: {{parameters}}
Evaluation Checklist
- Task matches structural schemas.
- Output is clean and formatted.