sae-lens 6.28.1__tar.gz → 6.29.1__tar.gz

{sae_lens-6.28.1 → sae_lens-6.29.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sae-lens
-Version: 6.28.1
+Version: 6.29.1
 Summary: Training and Analyzing Sparse Autoencoders (SAEs)
 License: MIT
 License-File: LICENSE
@@ -50,6 +50,8 @@ SAELens exists to help researchers:
 - Analyse sparse autoencoders / research mechanistic interpretability.
 - Generate insights which make it easier to create safe and aligned AI systems.
 
+SAELens inference works with any PyTorch-based model, not just TransformerLens. While we provide deep integration with TransformerLens via `HookedSAETransformer`, SAEs can be used with Hugging Face Transformers, NNsight, or any other framework by extracting activations and passing them to the SAE's `encode()` and `decode()` methods.
+
 Please refer to the [documentation](https://decoderesearch.github.io/SAELens/) for information on how to:
 
 - Download and Analyse pre-trained sparse autoencoders.
@@ -84,6 +86,14 @@ The new v6 update is a major refactor to SAELens and changes the way training co
 
 Feel free to join the [Open Source Mechanistic Interpretability Slack](https://join.slack.com/t/opensourcemechanistic/shared_invite/zt-375zalm04-GFd5tdBU1yLKlu_T_JSqZQ) for support!
 
+## Other SAE Projects
+
+- [dictionary-learning](https://github.com/saprmarks/dictionary_learning): An SAE training library that focuses on having hackable code.
+- [Sparsify](https://github.com/EleutherAI/sparsify): A lean SAE training library focused on TopK SAEs.
+- [Overcomplete](https://github.com/KempnerInstitute/overcomplete): SAE training library focused on vision models.
+- [SAE-Vis](https://github.com/callummcdougall/sae_vis): A library for visualizing SAE features, works with SAELens.
+- [SAEBench](https://github.com/adamkarvonen/SAEBench): A suite of LLM SAE benchmarks, works with SAELens.
+
 ## Citation
 
 Please cite the package as follows:

{sae_lens-6.28.1 → sae_lens-6.29.1}/README.md

@@ -14,6 +14,8 @@ SAELens exists to help researchers:
 - Analyse sparse autoencoders / research mechanistic interpretability.
 - Generate insights which make it easier to create safe and aligned AI systems.
 
+SAELens inference works with any PyTorch-based model, not just TransformerLens. While we provide deep integration with TransformerLens via `HookedSAETransformer`, SAEs can be used with Hugging Face Transformers, NNsight, or any other framework by extracting activations and passing them to the SAE's `encode()` and `decode()` methods.
+
 Please refer to the [documentation](https://decoderesearch.github.io/SAELens/) for information on how to:
 
 - Download and Analyse pre-trained sparse autoencoders.
@@ -48,6 +50,14 @@ The new v6 update is a major refactor to SAELens and changes the way training co
 
 Feel free to join the [Open Source Mechanistic Interpretability Slack](https://join.slack.com/t/opensourcemechanistic/shared_invite/zt-375zalm04-GFd5tdBU1yLKlu_T_JSqZQ) for support!
 
+## Other SAE Projects
+
+- [dictionary-learning](https://github.com/saprmarks/dictionary_learning): An SAE training library that focuses on having hackable code.
+- [Sparsify](https://github.com/EleutherAI/sparsify): A lean SAE training library focused on TopK SAEs.
+- [Overcomplete](https://github.com/KempnerInstitute/overcomplete): SAE training library focused on vision models.
+- [SAE-Vis](https://github.com/callummcdougall/sae_vis): A library for visualizing SAE features, works with SAELens.
+- [SAEBench](https://github.com/adamkarvonen/SAEBench): A suite of LLM SAE benchmarks, works with SAELens.
+
 ## Citation
 
 Please cite the package as follows:

{sae_lens-6.28.1 → sae_lens-6.29.1}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "sae-lens"
-version = "6.28.1"
+version = "6.29.1"
 description = "Training and Analyzing Sparse Autoencoders (SAEs)"
 authors = ["Joseph Bloom"]
 readme = "README.md"

{sae_lens-6.28.1 → sae_lens-6.29.1}/sae_lens/__init__.py

@@ -1,5 +1,5 @@
 # ruff: noqa: E402
-__version__ = "6.28.1"
+__version__ = "6.29.1"
 
 import logging
 

{sae_lens-6.28.1 → sae_lens-6.29.1}/sae_lens/pretrained_saes.yaml

@@ -40631,7 +40631,7 @@ gemma-3-1b-res-matryoshka-dc:
   conversion_func: null
   links:
     model: https://huggingface.co/google/gemma-3-1b-pt
-  model: gemma-3-1b
+  model: google/gemma-3-1b-pt
   repo_id: chanind/gemma-3-1b-batch-topk-matryoshka-saes-w-32k-l0-40
   saes:
   - id: blocks.0.hook_resid_post

{sae_lens-6.28.1 → sae_lens-6.29.1}/sae_lens/synthetic/__init__.py

@@ -17,11 +17,14 @@ from sae_lens.synthetic.activation_generator import (
     ActivationGenerator,
     ActivationsModifier,
     ActivationsModifierInput,
+    CorrelationMatrixInput,
 )
 from sae_lens.synthetic.correlation import (
+    LowRankCorrelationMatrix,
     create_correlation_matrix_from_correlations,
     generate_random_correlation_matrix,
     generate_random_correlations,
+    generate_random_low_rank_correlation_matrix,
 )
 from sae_lens.synthetic.evals import (
     SyntheticDataEvalResult,
@@ -66,6 +69,9 @@ __all__ = [
     "create_correlation_matrix_from_correlations",
     "generate_random_correlations",
     "generate_random_correlation_matrix",
+    "generate_random_low_rank_correlation_matrix",
+    "LowRankCorrelationMatrix",
+    "CorrelationMatrixInput",
     # Feature modifiers
     "ActivationsModifier",
     "ActivationsModifierInput",

sae_lens-6.29.1/sae_lens/synthetic/activation_generator.py

@@ -0,0 +1,388 @@
+"""
+Functions for generating synthetic feature activations.
+"""
+
+import math
+from collections.abc import Callable, Sequence
+
+import torch
+from torch import nn
+from torch.distributions import MultivariateNormal
+
+from sae_lens.synthetic.correlation import LowRankCorrelationMatrix
+from sae_lens.util import str_to_dtype
+
+ActivationsModifier = Callable[[torch.Tensor], torch.Tensor]
+ActivationsModifierInput = ActivationsModifier | Sequence[ActivationsModifier] | None
+CorrelationMatrixInput = (
+    torch.Tensor | LowRankCorrelationMatrix | tuple[torch.Tensor, torch.Tensor]
+)
+
+
+class ActivationGenerator(nn.Module):
+    """
+    Generator for synthetic feature activations.
+
+    This module provides a generator for synthetic feature activations with controlled properties.
+    """
+
+    num_features: int
+    firing_probabilities: torch.Tensor
+    std_firing_magnitudes: torch.Tensor
+    mean_firing_magnitudes: torch.Tensor
+    modify_activations: ActivationsModifier | None
+    correlation_matrix: torch.Tensor | None
+    low_rank_correlation: tuple[torch.Tensor, torch.Tensor] | None
+    correlation_thresholds: torch.Tensor | None
+    use_sparse_tensors: bool
+
+    def __init__(
+        self,
+        num_features: int,
+        firing_probabilities: torch.Tensor | float,
+        std_firing_magnitudes: torch.Tensor | float = 0.0,
+        mean_firing_magnitudes: torch.Tensor | float = 1.0,
+        modify_activations: ActivationsModifierInput = None,
+        correlation_matrix: CorrelationMatrixInput | None = None,
+        device: torch.device | str = "cpu",
+        dtype: torch.dtype | str = "float32",
+        use_sparse_tensors: bool = False,
+    ):
+        """
+        Create a new ActivationGenerator.
+
+        Args:
+            num_features: Number of features to generate activations for.
+            firing_probabilities: Probability of each feature firing. Can be a single
+                float (applied to all features) or a tensor of shape (num_features,).
+            std_firing_magnitudes: Standard deviation of firing magnitudes. Can be a
+                single float or a tensor of shape (num_features,). Defaults to 0.0
+                (deterministic magnitudes).
+            mean_firing_magnitudes: Mean firing magnitude when a feature fires. Can be
+                a single float or a tensor of shape (num_features,). Defaults to 1.0.
+            modify_activations: Optional function(s) to modify activations after
+                generation. Can be a single callable, a sequence of callables (applied
+                in order), or None. Useful for applying hierarchy constraints.
+            correlation_matrix: Optional correlation structure between features. Can be:
+
+                - A full correlation matrix tensor of shape (num_features, num_features)
+                - A LowRankCorrelationMatrix for memory-efficient large-scale correlations
+                - A tuple of (factor, diag) tensors representing low-rank structure
+
+            device: Device to place tensors on. Defaults to "cpu".
+            dtype: Data type for tensors. Defaults to "float32".
+            use_sparse_tensors: If True, return sparse COO tensors from sample().
+                Only recommended when using massive numbers of features. Defaults to False.
+        """
+        super().__init__()
+        self.num_features = num_features
+        self.firing_probabilities = _to_tensor(
+            firing_probabilities, num_features, device, dtype
+        )
+        self.std_firing_magnitudes = _to_tensor(
+            std_firing_magnitudes, num_features, device, dtype
+        )
+        self.mean_firing_magnitudes = _to_tensor(
+            mean_firing_magnitudes, num_features, device, dtype
+        )
+        self.modify_activations = _normalize_modifiers(modify_activations)
+        self.correlation_thresholds = None
+        self.correlation_matrix = None
+        self.low_rank_correlation = None
+        self.use_sparse_tensors = use_sparse_tensors
+
+        if correlation_matrix is not None:
+            if isinstance(correlation_matrix, torch.Tensor):
+                # Full correlation matrix
+                _validate_correlation_matrix(correlation_matrix, num_features)
+                self.correlation_matrix = correlation_matrix
+            else:
+                # Low-rank correlation (tuple or LowRankCorrelationMatrix)
+                correlation_factor, correlation_diag = (
+                    correlation_matrix[0],
+                    correlation_matrix[1],
+                )
+                _validate_low_rank_correlation(
+                    correlation_factor, correlation_diag, num_features
+                )
+                # Pre-compute sqrt for efficiency (used every sample call)
+                self.low_rank_correlation = (
+                    correlation_factor,
+                    correlation_diag.sqrt(),
+                )
+
+            # Vectorized inverse normal CDF: norm.ppf(1-p) = sqrt(2) * erfinv(1 - 2*p)
+            self.correlation_thresholds = math.sqrt(2) * torch.erfinv(
+                1 - 2 * self.firing_probabilities
+            )
+
+    @torch.no_grad()
+    def sample(self, batch_size: int) -> torch.Tensor:
+        """
+        Generate a batch of feature activations with controlled properties.
+
+        This is the main function for generating synthetic training data for SAEs.
+        Features fire independently according to their firing probabilities unless
+        a correlation matrix is provided.
+
+        Args:
+            batch_size: Number of samples to generate
+
+        Returns:
+            Tensor of shape [batch_size, num_features] with non-negative activations
+        """
+        # All tensors (firing_probabilities, std_firing_magnitudes, mean_firing_magnitudes)
+        # are on the same device from __init__ via _to_tensor()
+        device = self.firing_probabilities.device
+
+        if self.correlation_matrix is not None:
+            assert self.correlation_thresholds is not None
+            firing_indices = _generate_correlated_features(
+                batch_size,
+                self.correlation_matrix,
+                self.correlation_thresholds,
+                device,
+            )
+        elif self.low_rank_correlation is not None:
+            assert self.correlation_thresholds is not None
+            firing_indices = _generate_low_rank_correlated_features(
+                batch_size,
+                self.low_rank_correlation[0],
+                self.low_rank_correlation[1],
+                self.correlation_thresholds,
+                device,
+            )
+        else:
+            firing_indices = torch.bernoulli(
+                self.firing_probabilities.unsqueeze(0).expand(batch_size, -1)
+            ).nonzero(as_tuple=True)
+
+        # Compute activations only at firing positions (sparse optimization)
+        feature_indices = firing_indices[1]
+        num_firing = feature_indices.shape[0]
+        mean_at_firing = self.mean_firing_magnitudes[feature_indices]
+        std_at_firing = self.std_firing_magnitudes[feature_indices]
+        random_deltas = (
+            torch.randn(
+                num_firing, device=device, dtype=self.mean_firing_magnitudes.dtype
+            )
+            * std_at_firing
+        )
+        activations_at_firing = (mean_at_firing + random_deltas).relu()
+
+        if self.use_sparse_tensors:
+            # Return sparse COO tensor
+            indices = torch.stack(firing_indices)  # [2, nnz]
+            feature_activations = torch.sparse_coo_tensor(
+                indices,
+                activations_at_firing,
+                size=(batch_size, self.num_features),
+                device=device,
+                dtype=self.mean_firing_magnitudes.dtype,
+            )
+        else:
+            # Dense tensor path
+            feature_activations = torch.zeros(
+                batch_size,
+                self.num_features,
+                device=device,
+                dtype=self.mean_firing_magnitudes.dtype,
+            )
+            feature_activations[firing_indices] = activations_at_firing
+
+        if self.modify_activations is not None:
+            feature_activations = self.modify_activations(feature_activations)
+            if feature_activations.is_sparse:
+                # Apply relu to sparse values
+                feature_activations = feature_activations.coalesce()
+                feature_activations = torch.sparse_coo_tensor(
+                    feature_activations.indices(),
+                    feature_activations.values().relu(),
+                    feature_activations.shape,
+                    device=feature_activations.device,
+                    dtype=feature_activations.dtype,
+                )
+            else:
+                feature_activations = feature_activations.relu()
+
+        return feature_activations
+
+    def forward(self, batch_size: int) -> torch.Tensor:
+        return self.sample(batch_size)
+
+
+def _generate_correlated_features(
+    batch_size: int,
+    correlation_matrix: torch.Tensor,
+    thresholds: torch.Tensor,
+    device: torch.device,
+) -> tuple[torch.Tensor, torch.Tensor]:
+    """
+    Generate correlated binary features using multivariate Gaussian sampling.
+
+    Uses the Gaussian copula approach: sample from a multivariate normal
+    distribution, then threshold to get binary features.
+
+    Args:
+        batch_size: Number of samples to generate
+        correlation_matrix: Correlation matrix between features
+        thresholds: Pre-computed thresholds for each feature (from inverse normal CDF)
+        device: Device to generate samples on
+
+    Returns:
+        Tuple of (row_indices, col_indices) for firing features
+    """
+    num_features = correlation_matrix.shape[0]
+
+    mvn = MultivariateNormal(
+        loc=torch.zeros(num_features, device=device, dtype=thresholds.dtype),
+        covariance_matrix=correlation_matrix.to(device=device, dtype=thresholds.dtype),
+    )
+
+    gaussian_samples = mvn.sample((batch_size,))
+    indices = (gaussian_samples > thresholds.unsqueeze(0)).nonzero(as_tuple=True)
+    return indices[0], indices[1]
+
+
+def _generate_low_rank_correlated_features(
+    batch_size: int,
+    correlation_factor: torch.Tensor,
+    cov_diag_sqrt: torch.Tensor,
+    thresholds: torch.Tensor,
+    device: torch.device,
+) -> tuple[torch.Tensor, torch.Tensor]:
+    """
+    Generate correlated binary features using low-rank multivariate Gaussian sampling.
+
+    Uses the Gaussian copula approach with a low-rank covariance structure for scalability.
+    The covariance is represented as: cov = factor @ factor.T + diag(diag_term)
+
+    Args:
+        batch_size: Number of samples to generate
+        correlation_factor: Factor matrix of shape (num_features, rank)
+        cov_diag_sqrt: Pre-computed sqrt of diagonal term, shape (num_features,)
+        thresholds: Pre-computed thresholds for each feature (from inverse normal CDF)
+        device: Device to generate samples on
+
+    Returns:
+        Tuple of (row_indices, col_indices) for firing features
+    """
+    # Manual low-rank MVN sampling to enable autocast for the expensive matmul
+    # samples = eps @ cov_factor.T + eta * sqrt(cov_diag)
+    # where eps ~ N(0, I_rank) and eta ~ N(0, I_n)
+
+    num_features, rank = correlation_factor.shape
+
+    # Generate random samples in float32 for numerical stability
+    eps = torch.randn(batch_size, rank, device=device, dtype=correlation_factor.dtype)
+    eta = torch.randn(
+        batch_size, num_features, device=device, dtype=cov_diag_sqrt.dtype
+    )
+
+    gaussian_samples = eps @ correlation_factor.T + eta * cov_diag_sqrt
+
+    indices = (gaussian_samples > thresholds.unsqueeze(0)).nonzero(as_tuple=True)
+    return indices[0], indices[1]
+
+
+def _to_tensor(
+    value: torch.Tensor | float,
+    num_features: int,
+    device: torch.device | str,
+    dtype: torch.dtype | str,
+) -> torch.Tensor:
+    dtype = str_to_dtype(dtype)
+    device = torch.device(device)
+    if not isinstance(value, torch.Tensor):
+        value = value * torch.ones(num_features, device=device, dtype=dtype)
+    if value.shape != (num_features,):
+        raise ValueError(
+            f"Value must be a tensor of shape ({num_features},) or a float. Got {value.shape}"
+        )
+    return value.to(device, dtype)
+
+
+def _normalize_modifiers(
+    modify_activations: ActivationsModifierInput,
+) -> ActivationsModifier | None:
+    """Convert modifier input to a single modifier or None."""
+    if modify_activations is None:
+        return None
+    if callable(modify_activations):
+        return modify_activations
+    # It's a sequence of modifiers - chain them
+    modifiers = list(modify_activations)
+    if len(modifiers) == 0:
+        return None
+    if len(modifiers) == 1:
+        return modifiers[0]
+
+    def chained(activations: torch.Tensor) -> torch.Tensor:
+        result = activations
+        for modifier in modifiers:
+            result = modifier(result)
+        return result
+
+    return chained
+
+
+def _validate_correlation_matrix(
+    correlation_matrix: torch.Tensor, num_features: int
+) -> None:
+    """Validate that a correlation matrix has correct properties.
+
+    Args:
+        correlation_matrix: The matrix to validate
+        num_features: Expected number of features (matrix should be (num_features, num_features))
+
+    Raises:
+        ValueError: If the matrix has incorrect shape, non-unit diagonal, or is not positive definite
+    """
+    expected_shape = (num_features, num_features)
+    if correlation_matrix.shape != expected_shape:
+        raise ValueError(
+            f"Correlation matrix must have shape {expected_shape}, "
+            f"got {tuple(correlation_matrix.shape)}"
+        )
+
+    diagonal = torch.diag(correlation_matrix)
+    if not torch.allclose(diagonal, torch.ones_like(diagonal)):
+        raise ValueError("Correlation matrix diagonal must be all 1s")
+
+    try:
+        torch.linalg.cholesky(correlation_matrix)
+    except RuntimeError as e:
+        raise ValueError("Correlation matrix must be positive definite") from e
+
+
+def _validate_low_rank_correlation(
+    correlation_factor: torch.Tensor,
+    correlation_diag: torch.Tensor,
+    num_features: int,
+) -> None:
+    """Validate that low-rank correlation parameters have correct properties.
+
+    Args:
+        correlation_factor: Factor matrix of shape (num_features, rank)
+        correlation_diag: Diagonal term of shape (num_features,)
+        num_features: Expected number of features
+
+    Raises:
+        ValueError: If shapes are incorrect or diagonal terms are not positive
+    """
+    if correlation_factor.ndim != 2:
+        raise ValueError(
+            f"correlation_factor must be 2D, got {correlation_factor.ndim}D"
+        )
+    if correlation_factor.shape[0] != num_features:
+        raise ValueError(
+            f"correlation_factor must have shape ({num_features}, rank), "
+            f"got {tuple(correlation_factor.shape)}"
+        )
+    if correlation_diag.shape != (num_features,):
+        raise ValueError(
+            f"correlation_diag must have shape ({num_features},), "
+            f"got {tuple(correlation_diag.shape)}"
+        )
+    if torch.any(correlation_diag <= 0):
+        raise ValueError("correlation_diag must have all positive values")