Saelens — Train sparse autoencoders to interpret model features
Saelens — Train sparse autoencoders to interpret model features
Start with meaning, then move to detail.
This lesson explains Saelens — Train sparse autoencoders to interpret model features as part of getting started with Hermes correctly. You will learn what it does, when it matters, and the smallest safe test that proves it works.
If you are new, do not memorize names. Focus on three questions: what problem does this solve, what access does it need, and how can you verify the result?
For practice, inspect the first example, identify its effects, run it on test data, and compare the result with the source claim.
For advanced readers, inspect Skill metadata, Reference: full SKILL.md, The Problem: Polysemanticity & Superposition, then verify failure modes and version compatibility.
No prior experience is required; follow the steps on a safe test setup first.
A clear outcome before you read.
- Understand Saelens — Train sparse autoencoders to interpret model features without assumed prior knowledge.
- Separate the source description from what still needs testing in your environment.
- Read the first command and identify its inputs and outputs before copying it.
Short definitions before the details.
- Provider
- The service that runs or provides access and authentication to a model.
- Skill
- An instruction bundle that teaches Hermes a repeatable workflow without necessarily adding an external service.
Train sparse autoencoders to interpret model features
What does the source say, and in what order?
- 01Skill metadata
Start here to understand the core idea or structure.
- 02Reference: full SKILL.md
Read this after the foundation, then connect it to the previous step.
- 03The Problem: Polysemanticity & Superposition
Read this after the foundation, then connect it to the previous step.
- 04When to Use SAELens
Read this after the foundation, then connect it to the previous step.
- 05Installation
Read this after the foundation, then connect it to the previous step.
- 06Core Concepts
Read this after the foundation, then connect it to the previous step.
- 07What SAEs Learn
Read this after the foundation, then connect it to the previous step.
- 08Key Validation (Anthropic Research)
Read this after the foundation, then connect it to the previous step.
- 09Workflow 1: Loading and Analyzing Pre-trained SAEs
Read this after the foundation, then connect it to the previous step.
- 10Step-by-Step
Finish here to verify the result and special cases.
Copy only after you understand the effect.
pip install sae-lensInput Activation → Encoder → Sparse Features → Decoder → Reconstructed Activation
(d_model) ↓ (d_sae >> d_model) ↓ (d_model)
sparsity reconstruction
penalty loss### Available Pre-trained SAEs
| Release | Model | Layers |
|---------|-------|--------|
| `gpt2-small-res-jb` | GPT-2 Small | Multiple residual streams |
| `gemma-2b-res` | Gemma 2B | Residual streams |
| Various on HuggingFace | Search tag `saelens` | Various |
### Checklist
- [ ] Load model with TransformerLens
- [ ] Load matching SAE for target layer
- [ ] Encode activations to sparse features
- [ ] Identify top-activating features per token
- [ ] Validate reconstruction quality
## Workflow 2: Training a Custom SAE
### Step-by-StepRead the first command and identify its inputs and outputs before copying it.
Match every command to your installed Hermes version, review the files and accounts it can reach, and use non-sensitive data for the first test. If this explanation differs from the source, the official source wins.