Training
Knowledge editing (ROME / MEMIT)
Edit or adapt a modelWeight editingReferenceLocate where a fact is stored and rewrite it in place with a closed-form weight update, no fine-tuning.
- Objective
- Edit or adapt a model
- Targets
- Weights
- Format
- Rank-one / batched MLP weight update
- Granularity
- Targeted MLP layers per fact
- Lifecycle
- PTQ · no gradients
- Calibration
- Small calibration set
- Compression
- None — edits behavior, not size
- Quality
- Surgical — rewrites specific facts, aims to leave the rest intact
- Hardware
- None — offline weight edit, runs on any runtime
- Runtimes
A worked Spark recipe for this method hasn't been written yet — it lives here as a reference point in the ontology.