sonder-memory-and-training - Manage Sonder memory and QLoRA training
Operates Sonder Runtime’s lesson memory, calibrated retrieval, outcome recording, and attended QLoRA training and deployment lifecycle.
Tags
Updated: 2026-10-07Capabilities
Typical Inputs
Typical Outputs
What this skill does
- Retrieve relevant lessons
- Calibrate retrieval thresholds
- Quarantine failing lessons
- Record interaction outcomes
- Export training data
- Train personal adapters
- Deploy trained adapters
- Roll back models
Inputs
- Task text
- Interaction identifier
- Outcome signal
- Lesson corpus
- Embedding model
- Training interactions
Outputs
- Retrieved lessons
- Recorded outcome
- Quarantined lesson status
- Calibrated threshold
- Training data export
- Personal adapter
- Deployment state
- Model rollback state
Requirements
- Sonder Runtime environment
- Accessible Sonder state storage
- Compatible embedded corpus
- Frozen 4-bit base model
- Attended training data