Workbench: browser-based LLM intervention experiments on HuggingFace models, no code needed
davidbau · x · 2026-11-19
davidbau highlights Workbench (workbench.ndif.us), a visual tool built on NNterp and running on NDIF that lets researchers set up sophisticated activation-patching and intervention experiments on HuggingFace models with a few clicks and drags — no code required. The screenshot shows the Patch Lens feature: contrastive source/target prompts produce per-layer heatmaps, and activations can be dragged between positions to intervene and inspect shifted token probabilities (demoed on Llama-3.1-8B).
Related event: NNterp 2.0 Launches with Unified Interpretability API and Tooling Ecosystem(5 posts)→
More from Research
- tessera-small: a diffusion-LLM-based pixel-art text-to-image model — thejagstudio · 2026-11-19
- WorldRuler: rule-graph world models that honor user-defined game rules — AlayaLab · 2026-11-19
- NexSight: verifiable long-video tasks show top models score only 54.5% — nex-agi · 2026-11-19
- PReMind: benchmark + RL framework for hazard-aware personalized VLM assistants — SeoulNatlUniv · 2026-11-19
- Late Entry in looped LLMs delivers up to 12.5x lossy and 8.4x exact speedups — NanyangTechnologicalUniversity · 2026-11-19
- Optimization × AI Forum on Nov 20: two Peking University talks on AI-driven math discovery — aaron_defazio · 2026-11-19