saelens – skill Dữ liệu & Khoa học cho agent AI

SAELens: 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](https://github.com/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")
# In sae-lens v6, SAE.from_pretrained() returns JUST the SAE (not a tuple).
sae = SAE.from_pretrained(
    release="gpt2-small-res-jb",
    sae_id="blocks.8.hook_resid_pre",
    device="cuda"
)
# If you also need the cfg dict and feature sparsity, use:
# sae, cfg_dict, sparsity = SAE.from_pretrained_with_cfg_and_sparsity(...)

# 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 (
    LanguageModelSAETrainingRunner,
    LanguageModelSAERunnerConfig,
    StandardTrainingSAEConfig,
    LoggingConfig,
)

# 1. Configure training (v6 uses a NESTED config: SAE-specific options live in a
#    `sae=` sub-config, and logging options live in a `logger=` sub-config).
#    Note: `architecture`, `d_sae`, `l1_coefficient` etc. are now on the SAE sub-config,
#    and legacy flat options like `hook_layer`, `activation_fn`, `log_to_wandb` were removed.
cfg = LanguageModelSAERunnerConfig(
    # SAE architecture + sparsity (nested)
    sae=StandardTrainingSAEConfig(
        d_in=768,          # Model dimension
        d_sae=768 * 8,     # Expansion factor of 8
        l1_coefficient=8e-5,  # Sparsity penalty
        apply_b_dec_to_input=True,
        normalize_activations="expected_average_only_in",
    ),

    # Data-generating function (model + hook point)
    model_name="gpt2-small",
    hook_name="blocks.8.hook_resid_pre",  # layer is inferred from hook_name (no hook_layer)

    # Training
    lr=4e-4,
    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 (nested)
    logger=LoggingConfig(
        log_to_wandb=True,
        wandb_project="sae-training",
    ),

    # Checkpointing
    checkpoint_path="checkpoints",
    n_checkpoints=5,
)

# 2. Train
trainer = LanguageModelSAETrainingRunner(cfg)  # SAETrainingRunner still works as an alias
sae = trainer.run()

# 3. Evaluate
print(f"L0 (avg active features): {trainer.metrics['l0']}")
print(f"CE Loss Recovered: {trainer.metrics['ce_loss_score']}")

> v6 migration note: For other SAE types swap the sae= sub-config —

> GatedTrainingSAEConfig, TopKTrainingSAEConfig (set k directly), or

> JumpReLUTrainingSAEConfig (uses l0_coefficient). Legacy flat options

> (architecture, expansion_factor, hook_layer, activation_fn/activation_fn_kwargs,

> use_ghost_grads, ghost grads, b_dec/decoder init options) were removed in v6.

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(  # v6 returns just the SAE
    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 aff

… (nội dung đầy đủ có trong gói tải về)