Algorithm Overview¶

CircuitKit ships 13 discovery algorithms across 4 backends. This page explains how to choose the right one. Note that "ships" is not "validated": only 2 are validated at production scale — eap and eap-ig (Stable). acdc and ibcircuit are Experimental (GPT-2 scale; ibcircuit OOMs above ~3B), and the remaining 9 are Research (validated only on GPT-2 IOI).

EAP Family Stable

Gradient-based edge attribution patching. Default eap-ig is validated across GPT-2 through Llama-3B and Gemma-4B.

EAP variants
ACDC Experimental

Greedy edge-pruning. Produces minimal circuits at GPT-2 scale.

ACDC details
IBCircuit Experimental

Information-bottleneck noise model. No paired corruption data needed.

IBCircuit details
CD-T Research

Contextual decomposition through transformers. GPT-2 only.

CD-T details

Quick selection guide¶

Goal	Algorithm	Why
New to CircuitKit, any model	`eap-ig`	Stable, fast, validated across model families
Speed over precision	`eap`	~30% faster; slightly noisier
GPT-2 exploratory	`acdc`	Minimal circuits via greedy edge-pruning
Information-flow analysis	`ibcircuit`	No paired data needed
Large model (3B+)	`eap-ig`	Only Stable-tier validated at this scale

All 13 algorithms¶

EAP family¶

Algorithm	Tier	Description
`eap-ig`	Stable	EAP + Integrated Gradients — default
`eap`	Stable	Vanilla EAP — fast baseline
`eap-ig-activations`	Research	IG over node activations
`eap-clean-corrupted`	Research	EAP with both clean/corrupted passes
`eap-exact`	Research	Exact EAP (quadratic cost)
`atp-gd`	Research	Attribution Patching with GradDrop (AtP+GD)
`eap-gp`	Research	EAP-GP / GradPath
`relp`	Research	Relevance Patching (LRP-style)
`peap`	Research	Position-aware EAP (PEAP)
`eap-ifr`	Research	Information Flow Routes (IFR)

When to use: Start with eap-ig. Use eap if speed is the bottleneck. Use Research variants only for algorithm comparison studies.

ACDC¶

Algorithm	Tier	Description
`acdc`	Experimental	Greedy edge-pruning; GPT-2 scale

When you want a minimal circuit and you're working at GPT-2 scale.

IBCircuit¶

Algorithm	Tier	Description
`ibcircuit`	Experimental	Noise-model approach; clean-only data

When paired (clean, corrupted) examples are hard to construct.

OOM risk above ~3B

IBCircuit trains a noise model end-to-end, doubling memory. Known to OOM above ~3B parameters.

CD-T¶

Algorithm	Tier	Description
`cdt`	Research	Frozen-RoPE attention approximation; GPT-2 IOI only

Only for research replication.

Model compatibility¶

Algorithm	GPT-2	Llama 3.x	Gemma 2/3	Qwen 2.5
`eap-ig`	✅	✅	✅	✅
`eap`	✅	✅	✅	✅
`eap-ig-activations`	✅	❌	❌	❌
`eap-clean-corrupted`	✅	❌	❌	❌
`acdc`	✅	⚠️	⚠️	⚠️
`ibcircuit`	✅	⚠️	❌	❌
`cdt`	✅	❌	❌	❌
Research tier	✅	❌	❌	❌

✅ Validated ⚠️ Experimental ❌ Not validated

Algorithm-specific config keys¶

Algorithm	Key	Default	Description
`eap-ig`	`ig_steps`	`5`	Integration steps for IG
`acdc`	`tao_bases`	`[1, 3, 5, 7, 9]`	Bases for the tao threshold sweep
`acdc`	`tao_exps`	`[-5, -4, -3, -2]`	Exponents for the tao threshold sweep
`acdc`	`faithfulness_target`	`kl_div`	Metric optimized during pruning (`kl_div` or `mse`)
`ibcircuit`	`num_epochs`	`1000`	Training epochs for noise model
`ibcircuit`	`beta`	`0.001`	IB regularization weight