Skip to content
Back to skills

Serving And Adapters

ASecurity

"Plan LoRA merge commands and Gradio inference for Qwen-VL merged

  • 247 stars
  • 0 votes
  • 0 copies
  • 1 view
  • Added September 9, 2026
toolspythonbashbackend

Works with

  • cli

Security analysis

A100/100

Pro scans all 5 files and shows the line behind each finding

Scanned September 9, 2026

npx -y skills add VectorSpaceLab/AREX-Skill --skill serving-and-adapters --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Serving And Adapters?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Serving And Adapters
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/vectorspacelab-serving-and-adapters/badge)](https://www.skillsdirectory.com/skills/vectorspacelab-serving-and-adapters)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: "serving-and-adapters"
description: "Plan LoRA merge commands and Gradio inference for Qwen-VL merged
  or adapter-backed models."
metadata:
  disco-role: operating
disable-model-invocation: true
license: Apache 2.0
---

# Serving and Adapters

Use this sub-skill for LoRA merge planning, model loading, and Gradio multimodal inference.

## Covers

- `merge_lora_weights.py` command planning.
- LoRA adapter loading and merge behavior.
- Gradio multimodal inference.
- Device, quantization, and generation controls.
- Safe output-path planning for merged checkpoints.

## Excludes

- Training loops.
- Dataset schema design.
- Preference or classification training.

## Read first

- `../../references/workflow-map.md` for route confirmation.
- `../../references/cli-reference.md` for the serving and merge flag surface.
- `../../references/model-compatibility.md` for model and backend caveats.
- `../../references/troubleshooting.md` for serving and merge failures.
- `references/adapter-workflows.md` for merge/load details.
- `references/inference.md` for the Gradio launch pattern.
- `scripts/adapter_command.py` for a safe command builder.

## Typical user requests

- "How do I merge a LoRA adapter?"
- "How do I launch the Gradio demo?"
- "Should I load 4-bit or 8-bit for inference?"
- "How do I load a merged checkpoint versus an adapter checkpoint?"

## Workflow

1. Determine whether the user wants merge or inference.
2. Check whether the model is merged, adapter-backed, or a plain base checkpoint.
3. Decide whether quantization or a specific device should be used for serving.
4. Emit the command with the bundled command builder.
5. For merge tasks, verify the output path is explicit before execution.

## Decision rules

- Use merged weights for LoRA inference when the user wants the simplest serving path.
- Keep `model-base` explicit when loading adapter-backed checkpoints.
- Treat merge as an output-producing step that should not run accidentally.
- Gradio launch is a network-facing service; keep it as a deliberate action.

## Safe command builder

Start with the bundled helper:

```bash
python scripts/adapter_command.py merge --help
python scripts/adapter_command.py gradio --help
```

## If you need more detail

Read `references/adapter-workflows.md` for merge/load notes and `references/inference.md` for the serving path.

Files in this skill

  • SKILL.md2.3 KB
  • references/adapter-workflows.md1.3 KB
  • references/inference.md1 KB
  • references/troubleshooting.md1022 B
  • scripts/adapter_command.py3.8 KB

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.