
Sparse Autoencoder Training
FreeTrain and analyze Sparse Autoencoders for interpretable features.
Free · Opens the source repo
What Sparse Autoencoder Training does
Sparse Autoencoder Training provides a comprehensive framework for training and analyzing Sparse Autoencoders (SAEs) using the SAELens library. This skill is particularly useful for researchers and developers interested in mechanistic interpretability of neural networks. SAEs decompose complex neural activations into sparse, interpretable features, enabling users to better understand the learned representations within models. By leveraging the SAELens library, users can explore how individual neurons contribute to the overall model behavior, making it easier to identify and analyze polysemantic features.
The primary challenge addressed by this skill is the polysemanticity of neural network activations, where individual neurons can represent multiple distinct concepts. This often complicates the interpretability of models, especially in applications like language processing. Sparse Autoencoders tackle this issue by focusing on monosemantic representations, which ensures that only a small number of features are activated for any given input. This leads to clearer insights into what concepts a model has learned and how it processes information.
Users can utilize this skill in various scenarios, such as feature discovery, analyzing superposition in neural networks, and studying safety-relevant features like bias and harmful content. The provided workflows detail how to load pre-trained SAEs, analyze model activations, and even train custom SAEs tailored to specific datasets. With clear step-by-step instructions and example code snippets, this skill is designed to facilitate effective experimentation and analysis.
However, it is important to note that this skill is not intended for basic activation analysis or causal intervention experiments; users seeking those functionalities should consider alternatives like TransformerLens or pyvene. Overall, Sparse Autoencoder Training is a valuable tool for those looking to deepen their understanding of neural network interpretability and feature representation.
When to use it
Use this skill when you need to analyze and interpret the features learned by neural networks, particularly when working with language models and seeking to understand their internal representations.
When not to use it
Avoid this skill if you require basic activation analysis or causal intervention capabilities, as those are better served by other tools like TransformerLens.
What you can build with it
Feature Discovery in Language Models
Analyze the features learned by a language model to understand how it processes and represents different concepts.
Studying Model Interpretability
Use Sparse Autoencoders to gain insights into the interpretability of neural network activations, particularly in complex models.
Training Custom SAEs
Train your own Sparse Autoencoders on specific datasets to tailor feature extraction to your research needs.
How to install Sparse Autoencoder Training
View source1. Install with the skills CLI
npx skills add davila7/claude-code-templates/mechanistic-interpretability-saelens --agent claude-code2. Or install it manually
Download the skill folder and drop it into ~/.claude/skills/ for all projects, or .claude/skills/ to scope it to one repo. Restart Claude Code so it picks up the new skill.
Anthropic's agentic coding CLI, and the reference implementation of Agent Skills. Drop a skill folder into ~/.claude/skills and Claude Code loads it automatically whenever a task matches the skill's description. Claude Code docs
Inside SKILL.md
Written by davila7SAELens: Sparse Autoencoders for Mechanistic Interpretability
SAELens is the primary library for training and analyzing Sparse Autoencoders (SAEs) - a technique for decomposing polysemantic neural network activations into sparse, interpretable features. Based on Anthropic's groundbreaking research on monosemanticity.
GitHub: jbloomAus/SAELens (1,100+ stars)
The Problem: Polysemanticity & Superposition
Individual neurons in neural networks are polysemantic - they activate in multiple, semantically distinct contexts. This happens because models use superposition to represent more features than they have neurons, making interpretability difficult.
SAEs solve this by decomposing dense activations into sparse, monosemantic features - typically only a small number of features activate for any given input, and each feature corresponds to an interpretable concept.
When to Use SAELens
Use SAELens when you need to:
- Discover interpretable features in model activations
- Understand what concepts a model has learned
- Study superposition and feature geometry
- Perform feature-based steering or ablation
- Analyze safety-relevant features (deception, bias, harmful content)
Consider alternatives when:
- You need basic activation analysis → Use TransformerLens directly
- You want causal intervention experiments → Use pyvene or TransformerLens
- You need production steering → Consider direct activation engineering
Installation
pip install sae-lens
Requirements: Python 3.10+, transformer-lens>=2.0.0
Core Concepts
What SAEs Learn
SAEs are trained to reconstruct model activations through a sparse bottleneck:
Input Activation → Encoder → Sparse Features → Decoder → Reconstructed Activation
(d_model) ↓ (d_sae >> d_model) ↓ (d_model)
sparsity reconstruction
penalty loss
Loss Function: MSE(original, reconstructed) + L1_coefficient × L1(features)
Key Validation (Anthropic Research)
In "Towards Monosemanticity", human evaluators found 70% of SAE features genuinely interpretable. Features discovered include:
- DNA sequences, legal language, HTTP requests
- Hebrew text, nutrition statements, code syntax
- Sentiment, named entities, grammatical structures
Workflow 1: Loading and Analyzing Pre-trained SAEs
Step-by-Step
from transformer_lens import HookedTransformer
from sae_lens import SAE
# 1. Load model and pre-trained SAE
model = HookedTransformer.from_pretrained("gpt2-small", device="cuda")
sae, cfg_dict, sparsity = SAE.from_pretrained(
release="gpt2-small-res-jb",
sae_id="blocks.8.hook_resid_pre",
device="cuda"
)
# 2. Get model activations
tokens = model.to_tokens("The capital of France is Paris")
_, cache = model.run_with_cache(tokens)
activations = cache["resid_pre", 8] # [batch, pos, d_model]
# 3. Encode to SAE features
sae_features = sae.encode(activations) # [batch, pos, d_sae]
print(f"Active features: {(sae_features > 0).sum()}")
# 4. Find top features for each position
for pos in range(tokens.shape[1]):
top_features = sae_features[0, pos].topk(5)
token = model.to_str_tokens(tokens[0, pos:pos+1])[0]
print(f"Token '{token}': features {top_features.indices.tolist()}")
# 5. Reconstruct activations
reconstructed = sae.decode(sae_features)
reconstruction_error = (activations - reconstructed).norm()
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-Step
from sae_lens import SAE, LanguageModelSAERunnerConfig, SAETrainingRunner
# 1. Configure training
cfg = LanguageModelSAERunnerConfig(
# Model
model_name="gpt2-small",
hook_name="blocks.8.hook_resid_pre",
hook_layer=8,
d_in=768, # Model dimension
# SAE architecture
architecture="standard", # or "gated", "topk"
d_sae=768 * 8, # Expansion factor of 8
activation_fn="relu",
# Training
lr=4e-4,
l1_coefficient=8e-5, # Sparsity penalty
l1_warm_up_steps=1000,
train_batch_size_tokens=4096,
training_tokens=100_000_000,
# Data
dataset_path="monology/pile-uncopyrighted",
context_size=128,
# Logging
log_to_wandb=True,
wandb_project="sae-training",
# Checkpointing
checkpoint_path="checkpoints",
n_checkpoints=5,
)
# 2. Train
trainer = SAETrainingRunner(cfg)
sae = trainer.run()
# 3. Evaluate
print(f"L0 (avg active features): {trainer.metrics['l0']}")
print(f"CE Loss Recovered: {trainer.metrics['ce_loss_score']}")
Key Hyperparameters
| Parameter | Typical Value | Effect |
|---|---|---|
d_sae | 4-16× d_model | More features, higher capacity |
l1_coefficient | 5e-5 to 1e-4 | Higher = sparser, less accurate |
lr | 1e-4 to 1e-3 | Standard optimizer LR |
l1_warm_up_steps | 500-2000 | Prevents early feature death |
Evaluation Metrics
| Metric | Target | Meaning |
|---|---|---|
| L0 | 50-200 | Average active features per token |
| CE Loss Score | 80-95% | Cross-entropy recovered vs original |
| Dead Features | <5% | Features that never activate |
| Explained Variance | >90% | Reconstruction quality |
Checklist
- Choose target layer and hook point
- Set expansion factor (d_sae = 4-16× d_model)
- Tune L1 coefficient for desired sparsity
- Enable L1 warm-up to prevent dead features
- Monitor metrics during training (W&B)
- Validate L0 and CE loss recovery
- Check dead feature ratio
Workflow 3: Feature Analysis and Steering
Analyzing Individual Features
from transformer_lens import HookedTransformer
from sae_lens import SAE
import torch
model = HookedTransformer.from_pretrained("gpt2-small", device="cuda")
sae, _, _ = SAE.from_pretrained(
release="gpt2-small-res-jb",
sae_id="blocks.8.hook_resid_pre",
device="cuda"
)
# Find what activates a specific feature
feature_idx = 1234
test_texts = [
"The scientist conducted an experiment",
"I love chocolate cake",
"The code compiles successfully",
"Paris is beautiful in spring",
]
for text in test_texts:
tokens = model.to_tokens(text)
_, cache = model.run_with_cache(tokens)
features = sae.encode(cache["resid_pre", 8])
activation = features[0, :, feature_idx].max().item()
print(f"{activation:.3f}: {text}")
Feature Steering
def steer_with_feature(model, sae, prompt, feature_idx, strength=5.0):
"""Add SAE feature direction to residual stream."""
tokens = model.to_tokens(prompt)
# Get feature direction from decoder
feature_direction = sae.W_dec[feature_idx] # [d_model]
def steering_hook(activation, hook):
# Add scaled feature direction at all positions
activation += strength * feature_direction
return activation
# Generate with steering
output = model.generate(
tokens,
max_new_tokens=50,
fwd_hooks=[("blocks.8.hook_resid_pre", steering_hook)]
)
return model.to_string(output[0])
Feature Attribution
# Which features most affect a specific output?
tokens = model.to_tokens("The capital of France is")
_, cache = model.run_with_cache(tokens)
# Get features at final position
features = sae.encode(cache["resid_pre", 8])[0, -1] # [d_sae]
# Get logit attribution per feature
# Feature contribution = feature_activation × decoder_weight × unembedding
W_dec = sae.W_dec # [d_sae, d_model]
W_U = model.W_U # [d_model, vocab]
# Contribution to "Paris" logit
paris_token = model.to_single_token(" Paris")
feature_contributions = features * (W_dec @ W_U[:, paris_token])
top_features = feature_contributions.topk(10)
print("Top features for 'Paris' prediction:")
for idx, val in zip(top_features.indices, top_features.values):
print(f" Feature {idx.item()}: {val.item():.3f}")
Common Issues & Solutions
Issue: High dead feature ratio
# WRONG: No warm-up, features die early
cfg = LanguageModelSAERunnerConfig(
l1_coefficient=1e-4,
l1_warm_up_steps=0, # Bad!
)
# RIGHT: Warm-up L1 penalty
cfg = LanguageModelSAERunnerConfig(
l1_coefficient=8e-5,
l1_warm_up_steps=1000, # Gradually increase
use_ghost_grads=True, # Revive dead features
)
Issue: Poor reconstruction (low CE recovery)
# Reduce sparsity penalty
cfg = LanguageModelSAERunnerConfig(
l1_coefficient=5e-5, # Lower = better reconstruction
d_sae=768 * 16, # More capacity
)
Issue: Features not interpretable
# Increase sparsity (higher L1)
cfg = LanguageModelSAERunnerConfig(
l1_coefficient=1e-4, # Higher = sparser, more interpretable
)
# Or use TopK architecture
cfg = LanguageModelSAERunnerConfig(
architecture="topk",
activation_fn_kwargs={"k": 50}, # Exactly 50 active features
)
Issue: Memory errors during training
cfg = LanguageModelSAERunnerConfig(
train_batch_size_tokens=2048, # Reduce batch size
store_batch_size_prompts=4, # Fewer prompts in buffer
n_batches_in_buffer=8, # Smaller activation buffer
)
Integration with Neuronpedia
Browse pre-trained SAE features at neuronpedia.org:
# Features are indexed by SAE ID
# Example: gpt2-small layer 8 feature 1234
# → neuronpedia.org/gpt2-small/8-res-jb/1234
Key Classes Reference
| Class | Purpose |
|---|---|
SAE | Sparse Autoencoder model |
LanguageModelSAERunnerConfig | Training configuration |
SAETrainingRunner | Training loop manager |
ActivationsStore | Activation collection and batching |
HookedSAETransformer | TransformerLens + SAE integration |
Reference Documentation
For detailed API documentation, tutorials, and advanced usage, see the references/ folder:
| File | Contents |
|---|---|
| references/README.md | Overview and quick start guide |
| references/api.md | Complete API reference for SAE, TrainingSAE, configurations |
| references/tutorials.md | Step-by-step tutorials for training, analysis, steering |
External Resources
Tutorials
Papers
- Towards Monosemanticity - Anthropic (2023)
- Scaling Monosemanticity - Anthropic (2024)
- Sparse Autoencoders Find Highly Interpretable Features - Cunningham et al. (ICLR 2024)
Official Documentation
- SAELens Docs
- Neuronpedia - Feature browser
SAE Architectures
| Architecture | Description | Use Case |
|---|---|---|
| Standard | ReLU + L1 penalty | General purpose |
| Gated | Learned gating mechanism | Better sparsity control |
| TopK | Exactly K active features | Consistent sparsity |
# TopK SAE (exactly 50 features active)
cfg = LanguageModelSAERunnerConfig(
architecture="topk",
activation_fn="topk",
activation_fn_kwargs={"k": 50},
)
Frequently asked questions about Sparse Autoencoder Training
Similar skills
Spring Boot Testing
Master testing techniques for Spring Boot 4 applications.
GitHub Issues
Manage GitHub issues efficiently with MCP tools.
Geofeed Tuner
Optimize your IP geolocation feeds in CSV format.
Batch Files
Master Windows batch scripting for automation and task management.
Adobe Illustrator Scripting
Automate your Illustrator workflows with ExtendScript.
Plugin Structure
Create and organize Claude Code plugins effectively.
