CurvDelta: Low-Curvature Delta Modeling For Surgical Learnin
The story so far
CurvDelta and CurvLoRA enhance capability retention by constraining LLM updates within a low-curvature preserving space while maintaining target adaptation capabilities. These methods are applicable to retention-constrained adaptation tasks such as reasoning, code generation, and selective forgetting. On mathematical reasoning and code generation tasks, CurvDelta and CurvLoRA achieve downstream performance comparable to baseline models while better retaining general capabilities, providing a more optimal acquisition-retention trade-off.
AI-written from the reports · Updated 11 hours ago
Coverage timeline
Follow the coverage to see every side of the story.
- Hugging Face Daily PapersCurvDelta: Low-Curvature Delta Modeling For Surgical Learning and Unlearning
CurvDelta and CurvLoRA enhance capability retention by constraining LLM updates within a low-curvature preserving space while maintaining target adaptation capabilities. These methods are applicable to retention-constrained adaptation tasks such as reasoning, code generation, and selective forgetting. On mathematical reasoning and code generation tasks, CurvDelta and CurvLoRA achieve downstream performance comparable to baseline models while better retaining general capabilities, providing a more optimal acquisition-retention trade-off.
Heat over time
- Now
- 5
- Peak
- 719 Nov, 22:00
- 24h change
- –
Hover or tap the chart to see each hour; use the arrow keys on a keyboard.