decoderesearch/circuit-tracer

• highabout#1

  Add a concise description to the repository's "About" section

  Why:

  COPY-PASTE FIX

  Copy

  A library for finding, visualizing, and intervening on circuits and attribution graphs in MLP transcoders and neural networks, enabling mechanistic interpretability.

  OpenDoneDismiss
• mediumreadme#2

  Refine the README's opening paragraph to clarify its unique value and category

  Why:

  CURRENT

  # circuit-tracer This library implements tools for finding circuits using features from (cross-layer) MLP transcoders, as originally introduced by Ameisen et al. (2025) and Lindsey et al. (2025). Our library performs three main tasks. 1. Given a model with pre-trained transcoders, it finds the circuit / attribution graph; i.e., it computes the direct effect that each non-zero transcoder feature, transcoder error node, and input token has on each other non-zero transcoder feature and output logit. 2. Given an attribution graph, it visualizes this graph and allows you to annotate these features. 3. Enables interventions on a model's transcoder features using the insights gained from the attribution graph; i.e. you can set features to arbitrary values, and observe how model output changes.

  COPY-PASTE FIX

  Copy

  # circuit-tracer This library provides advanced tools for **mechanistic interpretability** in neural networks, specifically focusing on **finding, visualizing, and intervening on computational circuits** within models like MLP transcoders. Unlike general interpretability methods, `circuit-tracer` allows you to directly map and manipulate the causal paths of features, as introduced by Ameisen et al. (2025) and Lindsey et al. (2025). Our library performs three main tasks. 1. Given a model with pre-trained transcoders, it finds the circuit / attribution graph; i.e., it computes the direct effect that each non-zero transcoder feature, transcoder error node, and input token has on each other non-zero transcoder feature and output logit. 2. Given an attribution graph, it visualizes this graph and allows you to annotate these features. 3. Enables interventions on a model's transcoder features using the insights gained from the attribution graph; i.e. you can set features to arbitrary values, and observe how model output changes.

  OpenDoneDismiss

Category GEO backends resolved for this scan: google/gemini-2.5-flash, deepseek/deepseek-v4-flash

• CATEGORY QUERY

  How to analyze and visualize internal feature attribution graphs in neural network models?

  you: not recommended
  AI recommended (in order):

  1. Captum
  2. SHAP
  3. LIME
  4. TensorBoard
  5. TF-Explain
  6. Lucid
  7. iNNvestigate

  AI recommended 7 alternatives but never named decoderesearch/circuit-tracer. This is the gap to close.

  Show full AI answer
• CATEGORY QUERY

  What tools enable direct intervention on neural network features to observe output changes?

  you: not recommended
  AI recommended (in order):

  1. Captum (facebookresearch/captum)
  2. Lucid (tensorflow/lucid)
  3. PyTorch Hooks
  4. TensorFlow Debugger (tfdbg)
  5. SHAP (SHapley Additive exPlanations) (shap/shap)
  6. LIME (Local Interpretable Model-agnostic Explanations) (marcotcr/lime)

  AI recommended 6 alternatives but never named decoderesearch/circuit-tracer. This is the gap to close.

  Show full AI answer

• Metadata completeness

  fail

  Suggestion:

• README presence

  pass

• Compared to common alternatives in this category, what is the core differentiator of decoderesearch/circuit-tracer?

  pass
  ⚠AI did not name decoderesearch/circuit-tracer — likely talking about a different project

  AI answers can be confidently wrong. Read for accuracy: does it match your actual tech stack, audience, and differentiator?

• If a team adopts decoderesearch/circuit-tracer in production, what risks or prerequisites should they evaluate first?

  pass
  ✓AI named decoderesearch/circuit-tracer explicitly

  AI answers can be confidently wrong. Read for accuracy: does it match your actual tech stack, audience, and differentiator?

• In one sentence, what problem does the repo decoderesearch/circuit-tracer solve, and who is the primary audience?

  pass
  ✓AI named decoderesearch/circuit-tracer explicitly

  AI answers can be confidently wrong. Read for accuracy: does it match your actual tech stack, audience, and differentiator?