OpenInterpretability — notebooks

Open notebooks for training SAEs and reproducing 2024–2026 mech-interp papers on Gemma, Qwen, and Llama. Apache-2.0.

· openinterp.org/train

The 5-repo ecosystem

---

Core ladder — train your first SAE

After you train — close the loop

Before you train — reduce friction

More models — same recipe, different architectures

Research-grade — replicate published results

Safety + production preview

Circuits — attribution graphs between SAE features

All circuit notebooks emit JSON consumed directly by the Circuit Canvas on openinterp.org.

Leaderboard — InterpScore v0.0.1

Lenses — classic layer-wise prediction tools

Probing — the supervised baselines SAE features must beat

Hallucination — detection & steering arc

The full research arc behind the 2026-04-25 blog post on hallucination in 27B reasoning models. Notebooks 24 → 28b shipped 2026-04-25 → 26.

Crosscoders — cross-model + cross-stage

The methodology behind paper-1's Pearson causal-equivalence (Pearson_CE) finding. First per-feature causal-equivalence test in the crosscoder literature.

Guards — product reproducers

Each notebook reproduces an exact metric behind a shipped openinterp Guard (SDK on PyPI, demo on HF, landing on openinterp.org/products/X). Drop-in pip install openinterp and you have these probes.

Each reproducer ships:

• probe.joblib + meta.json to HF dataset (drop-in for the SDK)
• verdict.json with raw numbers
• headline.png for landing pages / posts
• All artifacts pushed to caiovicentino1/<ProductName>-linearprobe-qwen36-27b (HF dataset)

---

Shared recipe (every training tier)

All tiers use the same research-grade protocol; hyperparameters scale:

• TopK activation (Gao et al. 2024) — hard top-k, no L1 penalty
• AuxK auxiliary loss — dead-feature revival (α=1/32, k_aux=d/2, dead_threshold=10M tokens)
• Geometric-median b_dec init (Weiszfeld) — robust to heavy-tailed residuals
• Decoder column renorm every step — keeps features interpretable
• Cosine LR + warmup — non-zero floor for continued dead-feature revival
• HuggingFace streaming checkpoints — crash-safe, never lose more than 5-10 min
• sae_lens-compatible export — safetensors + cfg.json

---

Notebook constraints

Use dtype=torch.bfloat16 (not the deprecated torch_dtype=) and attn_implementation='sdpa' (not flash-attn — reproducibility + install pain across Colab/Kaggle). HF_TOKEN goes through Colab/Kaggle secrets, never hard-coded. Stream checkpoints to HF every 5–10M tokens — Drive-only checkpoints die with the kernel. Use the multimodal layer-access fallback (getattr(model.model, 'layers', None) or model.model.language_model.layers), not a hard-coded .layers[N]. Report honest var_expl, L0, and dead-feature percentage — not cherry-picked seeds. CI checks all of these.

---

Contributing

Three common PR patterns, full rules in CONTRIBUTING.md:

1. Port a notebook to a new model — pick an existing notebook at your tier and swap MODEL_ID, LAYER, D_MODEL. Name it NN_<tier>_<model>_<platform>.ipynb.
2. Replicate a 2024–2026 paper — title cell with arxiv link, pinned install, paper hyperparameters, inline implementation, validation cell that matches the paper's headline metric within tolerance.
3. Add a platform (TPU/ROCm/MPS) — write a _platform_<name>.py helper with pick_device() / get_dtype(), patch one notebook as PoC, open a draft PR and tag @caiovicentino for design review.

Before opening a PR, validate JSON: python3 -c "import json; json.load(open('notebooks/YOUR.ipynb'))". CI runs nbformat.validate. If you have a GPU, dry-run with jupyter nbconvert --to notebook --execute --ExecutePreprocessor.timeout=300 — expect heavy training cells to time out; you're just catching import + dtype bugs.

---

Output schemas other tools consume

If your notebook emits a JSON that the website consumes, match the schema:

---

Where to go next

Your SAE is an asset. Put it to work:

• Trace Theater — 10 scenarios, view + share
• InterpScore — public leaderboard, submit your SAE
• Sandbox (Q2 2026) — drag-and-drop steering
• Expeditions (Q3 2026) — turn your run into a tutorial

---

Community

• Discussions — "which notebook should I use for X?"
• Good-first-issues — start here
• Contributor guide — full workflow
• hi@openinterp.org

---

Built on

SAELens (checkpoint format) · Gemma Scope (reference at-scale SAE suite) · Gao et al. 2024 (TopK + AuxK) · Bussmann et al. 2024 (BatchTopK) · Neuronpedia.

Apache-2.0 · openinterp.org