quantized-export
Export a promoted fine-tuned model in the right deployment format
by wshobsonFree
Description
quantized-export is an agent skill from the wshobson/agents collection for exporting a promoted fine-tuned model in the format its deployment target needs. It is the last step after the checkpoint-promotion gate: a checkpoint that cleared the four-stage gate is not deployed until it has been exported in the right format for its runtime and shown to still work after export. The skill covers the common export formats: merged safetensors, which folds the adapter weights into the base model; LoRA-only output, which keeps the adapter separate; GGUF with an importance matrix (imatrix), a quantized format used by many local inference tools; and FP8, a lower-precision format for accelerated hardware. It explains which format fits which target runtime, and it requires a verification pass after export, because quantization and conversion can change behavior even when the original checkpoint was sound. A REJECT verdict from the gate never reaches this skill, so export only starts from a promoted checkpoint. The skill is for teams that have trained and promoted a model and need it running in production or on a device. Export and quantization details depend on the exact tools and versions, so the agent's commands should be checked against them. The skill is installed with the skills CLI from the wshobson/agents repository. Its skills.sh page showed about 2.4K installs and passing results from three security audits at the time of the scan.…
Highlights
- Export of promoted fine-tuned checkpoints
- Merged safetensors and LoRA-only output
- GGUF with imatrix and FP8 formats
- Format choice by target runtime
- Post-export verification step
- Installs with the skills CLI
Ratings & Reviews
No ratings yet
Be the first to rate quantized-export.
People also visit
quantized-export FAQ
When should I use it?
After a checkpoint has passed the checkpoint-promotion gate and needs to be deployed.
Which formats does it cover?
Merged safetensors, LoRA-only, GGUF with imatrix, and FP8.
Does it test the exported model?
Yes. The skill requires a check that the model still works after export.
What happens to a rejected checkpoint?
A REJECT verdict never reaches this skill, so only promoted checkpoints are exported.
How do I install it?
Run npx skills add https://github.com/wshobson/agents --skill quantized-export.
What security audits does it have?
skills.sh lists three audits: Gen Agent Trust Hub Pass, Socket Pass and Snyk Pass.
What are the best quantized-export alternatives?
Popular alternatives to quantized-export include dataset-curation, eval-harness-first and finetuning-method-selection. See all quantized-export alternatives