yzj623/pplab/adalora-recovery-harness
Claude Code and agent skill by yzj623/pplab
- community
Checking that the SKILL.md file is still in the repository.
Run bounded soft-mode AdaLoRA recovery experiments with source-boundary logs, mechanism checks, and validation artifacts.
Install
npx vskill@latest install yzj623/pplab/adalora-recovery-harnessSKILL.md
The same SKILL.md was published earlier as vectorspacelab/arex-skill/adalora-recovery-harness.
AdaLoRA Recovery Harness
Use this skill to recover a fast AdaLoRA result when full DeBERTa or BART fine-tuning is blocked by runtime limits. The harness must be explicit that it is a reduced/proxy experiment.
Inputs
- Attempt directory containing
module_plan.jsonandenvironment/runtime_handoff.json. - Generated skill root with the SVD layer, rank allocator, and scheduler skills.
- Optional deterministic experiment configuration.
Outputs
recovery/logs/generated_data_item.json.recovery/logs/training_trace.jsonwithloss_before,loss_after,params_before, andparams_after.recovery/recovery_result.jsonwith paper target copied frommodule_plan.json.- Source manifest and generated-skill invocation log.
Workflow
- Read only allowed attempt artifacts and generated skill files; never read the original repository.
- Build a deterministic tiny linear adaptation problem.
- Invoke the generated SVD layer, budget scheduler, and rank allocator scripts or importable helpers.
- Perform an actual parameter update on trainable reduced parameters.
- Apply rank masking and compute mechanism booleans: SVD update used, optimizer changed parameters, budget respected, high-importance triplet retained, and loss decreased.
- Mark full-model booleans false when no real DeBERTa/BART training occurred.
- Run the recovery experiment validator.
Validation
Run the harness script from an attempt directory, then run validate_recovery_experiment.py.
Limitations
This is not a replacement for full paper reproduction. It is acceptable only in soft mode after full runtime blockers are recorded in runtime_handoff.json.
Security
- Security scanPassed, 90 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingResearchAgent-Skill/skill_library/paperbench/adaptive-pruning/skill/adalora_adaptive_budget/adalora_recovery_harness/SKILL.md
Popularity
- 7-day trend score
- 5
Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/yzj623/pplab/adalora-recovery-harness)