Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,82 @@
|
|
| 1 |
---
|
| 2 |
license: mit
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
license: mit
|
| 3 |
+
language:
|
| 4 |
+
- en
|
| 5 |
+
tags:
|
| 6 |
+
- mechanistic-interpretability
|
| 7 |
+
- circuit-analysis
|
| 8 |
+
- transformers
|
| 9 |
+
- benchmark
|
| 10 |
+
pretty_name: AgenticInterpBench
|
| 11 |
+
size_categories:
|
| 12 |
+
- n<1K
|
| 13 |
---
|
| 14 |
+
|
| 15 |
+
# AgenticInterpBench
|
| 16 |
+
|
| 17 |
+
A benchmark for evaluating language-model agents on **transformer circuit explanation**. Given an already-localized circuit, the agent must recover what each component does: a functional role tag from a 5-class taxonomy, a task-specific natural-language note, and a derived description of the overall task.
|
| 18 |
+
|
| 19 |
+
The benchmark has **84 semi-synthetic circuits with 163 annotated components**, plus a manually annotated real-model circuit (three-operand addition in Llama-3-8B). It was introduced in *Can Language Model Agents be Helpful Circuit Explainers in Mechanistic Interpretability?* and is used to evaluate the HyVE (Hypothesize, Validate, Explain) agent framework.
|
| 20 |
+
|
| 21 |
+
## Contents
|
| 22 |
+
|
| 23 |
+
```text
|
| 24 |
+
annotations/case_{id}.json # 84 (task specifications, I/O examples, gold component roles)
|
| 25 |
+
real_circuits/llama3_abc_annotations.json # The AF1 Circuit / Llama-3-8B reference annotation (10 components)
|
| 26 |
+
```
|
| 27 |
+
|
| 28 |
+
Each `annotations/case_{id}.json` carries the task summary, up to five input–output examples, the originating RASP program, the target model, and the localized components. Every component has an `id`, a TransformerLens `hook`, and a gold `role` (`tag` + task-specific `note`), following the running `frac_prevs` example:
|
| 29 |
+
|
| 30 |
+
```json
|
| 31 |
+
[
|
| 32 |
+
{
|
| 33 |
+
"id": "L0_MLP",
|
| 34 |
+
"hook": "blocks.0.mlp.hook_post",
|
| 35 |
+
"role": {
|
| 36 |
+
"tag": "INDICATOR",
|
| 37 |
+
"note": "Computes per-position feature indicating whether the token at that position is 'x' or not."
|
| 38 |
+
},
|
| 39 |
+
"labels": ["is_x_3"]
|
| 40 |
+
},
|
| 41 |
+
{
|
| 42 |
+
"id": "L1H2_ATTN",
|
| 43 |
+
"hook": "blocks.1.attn.hook_result[2]",
|
| 44 |
+
"role": {
|
| 45 |
+
"tag": "AGGREGATOR",
|
| 46 |
+
"note": "Aggregates prefix fraction by attending over previous positions."
|
| 47 |
+
},
|
| 48 |
+
"labels": ["frac_prevs_1"]
|
| 49 |
+
}
|
| 50 |
+
]
|
| 51 |
+
```
|
| 52 |
+
|
| 53 |
+
`real_circuits/llama3_abc_annotations.json` holds reference roles for the 10-component AF1 circuit (Mamidanna et al., 2025), with per-component notes, and supporting findings.
|
| 54 |
+
|
| 55 |
+
## Role taxonomy
|
| 56 |
+
|
| 57 |
+
| Tag | Description |
|
| 58 |
+
|---|---|
|
| 59 |
+
| INDICATOR | Detects a property of the current token and emits a binary/predicate signal. |
|
| 60 |
+
| AGGREGATOR | Summarizes selected positions into a count, fraction, or accumulated quantity. |
|
| 61 |
+
| ROUTER | Moves content between positions via positional or index-based selection. |
|
| 62 |
+
| MAPPER | Transforms each position independently into a non-binary output. |
|
| 63 |
+
| COMBINER | Fuses two or more upstream signals into one output. |
|
| 64 |
+
|
| 65 |
+
## Statistics
|
| 66 |
+
| | |
|
| 67 |
+
|---|---|
|
| 68 |
+
| Circuits | 84 |
|
| 69 |
+
| Annotated components | 163 |
|
| 70 |
+
| MLP / attention components | 120 / 43 |
|
| 71 |
+
| Components per circuit (avg / min / max) | 1.94 / 1 / 10 |
|
| 72 |
+
| Tags (MAPPER / COMBINER / AGGREGATOR / INDICATOR / ROUTER) | 72 / 33 / 32 / 15 / 11 |
|
| 73 |
+
|
| 74 |
+
## Provenance & license
|
| 75 |
+
|
| 76 |
+
The annotation layer here is the authors' own work, released under the **MIT License**. The synthetic tasks and circuits derive from [InterpBench](https://huggingface.co/cybershiptrooper/InterpBench) (Gupta et al., 2025), which retrains Tracr-compiled RASP programs; the two IOI tasks are excluded. Model weights live in the InterpBench repo and are loaded separately at runtime by HyVE.
|
| 77 |
+
|
| 78 |
+
## Citation
|
| 79 |
+
|
| 80 |
+
```bibtex
|
| 81 |
+
TBD
|
| 82 |
+
```
|