MODEL DEVELOPMENT / Seth Hobson
LLM fine-tuning training engineer
A training component that follows an approved brief and hands checkpoint promotion to a separate evaluator.
“Fine-tuning implementation workhorse — prepares datasets, generates Unsloth-first training scripts, launches and monitors runs, and exports artifacts. Use after a training brief exists, for dataset preparation, training execution, or model export.”
01 / THE REASONING
Why this made the selection.
- It decodes packed examples to inspect formatting and requires a hardware/environment preflight before launching a job.
- Export follows a PROMOTE decision and compares actual serving-runtime output before and after export; it must not rewrite the evaluation goldens.
02 / THE REVIEW RECORD
What we actually inspected.
Source review has boundaries.
A clear record is more useful than a “safe” badge.
Material inspected
- plugins/llm-finetuning/agents/llm-finetuning-training-engineer.md (pinned complete source retained)
- LICENSE
- Repository README and current official host configuration documentation
- plugins/llm-finetuning/skills/quantized-export/SKILL.md
Our findings
- It decodes packed examples to inspect formatting and requires a hardware/environment preflight before launching a job.
- Export follows a PROMOTE decision and compares actual serving-runtime output before and after export; it must not rewrite the evaluation goldens.
Not established by this review
- Source review only: no agent was installed or invoked. Task correctness, effective host restrictions and integration behaviour were not tested.
The review applies to the material and revision named here. A newer upstream release can change its behavior.
03 / PUT IT TO WORK
Use the role in your project.
- Read the pinned source and its licence, then configure an explicitly bounded host context.
- Review the companion skills and authorize each execution surface; provide a disposable training environment and an independent evaluation decision.
- Supply training-brief.md, dataset, environment report and budget.
Before you start
- Matching fine-tuning skills and training brief
- Compatible training stack, dataset and hardware
- Independent evaluation and export target
04 / FOLLOW THE EVIDENCE
The source trail.
Our notes are separate from the original resource.
Check upstream before adopting a new version.
https://github.com/wshobson/agents/blob/46891e7e60da0e52baf1050b7b6391b64e84c6d9/plugins/llm-finetuning/agents/llm-finetuning-training-engineer.md
Supports: Workflow, inputs, outputs and stated boundaries
- Applicable licence ↗Checked
https://github.com/wshobson/agents/blob/46891e7e60da0e52baf1050b7b6391b64e84c6d9/LICENSE
Supports: Redistribution and attribution
https://code.claude.com/docs/en/sub-agents.md
Supports: Host loading and effective permission boundaries
https://github.com/wshobson/agents/blob/46891e7e60da0e52baf1050b7b6391b64e84c6d9/plugins/llm-finetuning/skills/quantized-export/SKILL.md
Supports: Dependency setup and source-review boundary