Inference Extrapolation
Activation Patching
Introduced by Ghandeharioun et al. in Patchscopes: A Unifying Framework for Inspecting Hidden Representations of Language ModelsActivation patching studies the model's computation by altering its latent representations, the token embeddings in transformer-based language models, during the inference process
Source: Patchscopes: A Unifying Framework for Inspecting Hidden Representations of Language ModelsPapers
| Paper | Code | Results | Date | Stars |
|---|
Tasks
| Task | Papers | Share |
|---|---|---|
| Time Series Forecasting | 6 | 10.71% |
| Language Modelling | 3 | 5.36% |
| Vulnerability Detection | 2 | 3.57% |
| Question Answering | 2 | 3.57% |
| Safety Alignment | 2 | 3.57% |
| Image Classification | 2 | 3.57% |
| Multivariate Time Series Forecasting | 2 | 3.57% |
| C++ code | 1 | 1.79% |
| Code Translation | 1 | 1.79% |
Usage Over Time
Components
| Component | Type |
|
|---|---|---|
| 🤖 No Components Found | You can add them if they exist; e.g. Mask R-CNN uses RoIAlign |
