sae-lens 6.12.1__py3-none-any.whl → 6.21.0__py3-none-any.whl

sae_lens/training/sae_trainer.py

@@ -1,4 +1,5 @@
 import contextlib
+import math
 from pathlib import Path
 from typing import Any, Callable, Generic, Protocol
 
@@ -10,7 +11,11 @@ from tqdm.auto import tqdm
 
 from sae_lens import __version__
 from sae_lens.config import SAETrainerConfig
-from sae_lens.constants import ACTIVATION_SCALER_CFG_FILENAME, SPARSITY_FILENAME
+from sae_lens.constants import (
+    ACTIVATION_SCALER_CFG_FILENAME,
+    SPARSITY_FILENAME,
+    TRAINER_STATE_FILENAME,
+)
 from sae_lens.saes.sae import (
     T_TRAINING_SAE,
     T_TRAINING_SAE_CONFIG,
@@ -56,6 +61,7 @@ class SAETrainer(Generic[T_TRAINING_SAE, T_TRAINING_SAE_CONFIG]):
     data_provider: DataProvider
     activation_scaler: ActivationScaler
     evaluator: Evaluator[T_TRAINING_SAE] | None
+    coefficient_schedulers: dict[str, CoefficientScheduler]
 
     def __init__(
         self,
@@ -84,7 +90,9 @@ class SAETrainer(Generic[T_TRAINING_SAE, T_TRAINING_SAE_CONFIG]):
                 range(
                     0,
                     cfg.total_training_samples,
-                    cfg.total_training_samples // self.cfg.n_checkpoints,
+                    math.ceil(
+                        cfg.total_training_samples / (self.cfg.n_checkpoints + 1)
+                    ),
                 )
             )[1:]
 
@@ -93,11 +101,6 @@ class SAETrainer(Generic[T_TRAINING_SAE, T_TRAINING_SAE_CONFIG]):
             sae.cfg.d_sae, device=cfg.device
         )
         self.n_frac_active_samples = 0
-        # we don't train the scaling factor (initially)
-        # set requires grad to false for the scaling factor
-        for name, param in self.sae.named_parameters():
-            if "scaling_factor" in name:
-                param.requires_grad = False
 
         self.optimizer = Adam(
             sae.parameters(),
@@ -210,10 +213,7 @@ class SAETrainer(Generic[T_TRAINING_SAE, T_TRAINING_SAE_CONFIG]):
                 sparsity_path = checkpoint_path / SPARSITY_FILENAME
                 save_file({"sparsity": self.log_feature_sparsity}, sparsity_path)
 
-                activation_scaler_path = (
-                    checkpoint_path / ACTIVATION_SCALER_CFG_FILENAME
-                )
-                self.activation_scaler.save(str(activation_scaler_path))
+                self.save_trainer_state(checkpoint_path)
 
                 if self.cfg.logger.log_to_wandb:
                     self.cfg.logger.log(
@@ -227,6 +227,44 @@ class SAETrainer(Generic[T_TRAINING_SAE, T_TRAINING_SAE_CONFIG]):
         if self.save_checkpoint_fn is not None:
             self.save_checkpoint_fn(checkpoint_path=checkpoint_path)
 
+    def save_trainer_state(self, checkpoint_path: Path) -> None:
+        checkpoint_path.mkdir(exist_ok=True, parents=True)
+        scheduler_state_dicts = {
+            name: scheduler.state_dict()
+            for name, scheduler in self.coefficient_schedulers.items()
+        }
+        torch.save(
+            {
+                "optimizer": self.optimizer.state_dict(),
+                "lr_scheduler": self.lr_scheduler.state_dict(),
+                "n_training_samples": self.n_training_samples,
+                "n_training_steps": self.n_training_steps,
+                "act_freq_scores": self.act_freq_scores,
+                "n_forward_passes_since_fired": self.n_forward_passes_since_fired,
+                "n_frac_active_samples": self.n_frac_active_samples,
+                "started_fine_tuning": self.started_fine_tuning,
+                "coefficient_schedulers": scheduler_state_dicts,
+            },
+            str(checkpoint_path / TRAINER_STATE_FILENAME),
+        )
+        activation_scaler_path = checkpoint_path / ACTIVATION_SCALER_CFG_FILENAME
+        self.activation_scaler.save(str(activation_scaler_path))
+
+    def load_trainer_state(self, checkpoint_path: Path | str) -> None:
+        checkpoint_path = Path(checkpoint_path)
+        self.activation_scaler.load(checkpoint_path / ACTIVATION_SCALER_CFG_FILENAME)
+        state_dict = torch.load(checkpoint_path / TRAINER_STATE_FILENAME)
+        self.optimizer.load_state_dict(state_dict["optimizer"])
+        self.lr_scheduler.load_state_dict(state_dict["lr_scheduler"])
+        self.n_training_samples = state_dict["n_training_samples"]
+        self.n_training_steps = state_dict["n_training_steps"]
+        self.act_freq_scores = state_dict["act_freq_scores"]
+        self.n_forward_passes_since_fired = state_dict["n_forward_passes_since_fired"]
+        self.n_frac_active_samples = state_dict["n_frac_active_samples"]
+        self.started_fine_tuning = state_dict["started_fine_tuning"]
+        for name, scheduler_state_dict in state_dict["coefficient_schedulers"].items():
+            self.coefficient_schedulers[name].load_state_dict(scheduler_state_dict)
+
     def _train_step(
         self,
         sae: T_TRAINING_SAE,
@@ -249,16 +287,19 @@ class SAETrainer(Generic[T_TRAINING_SAE, T_TRAINING_SAE_CONFIG]):
                     sae_in=sae_in,
                     dead_neuron_mask=self.dead_neurons,
                     coefficients=self.get_coefficients(),
+                    n_training_steps=self.n_training_steps,
                 ),
             )
 
             with torch.no_grad():
-                did_fire = (train_step_output.feature_acts > 0).float().sum(-2) > 0
+                # calling .bool() should be equivalent to .abs() > 0, and work with coo tensors
+                firing_feats = train_step_output.feature_acts.bool().float()
+                did_fire = firing_feats.sum(-2).bool()
+                if did_fire.is_sparse:
+                    did_fire = did_fire.to_dense()
                 self.n_forward_passes_since_fired += 1
                 self.n_forward_passes_since_fired[did_fire] = 0
-                self.act_freq_scores += (
-                    (train_step_output.feature_acts.abs() > 0).float().sum(0)
-                )
+                self.act_freq_scores += firing_feats.sum(0)
                 self.n_frac_active_samples += self.cfg.train_batch_size_samples
 
         # Grad scaler will rescale gradients if autocast is enabled
@@ -310,7 +351,7 @@ class SAETrainer(Generic[T_TRAINING_SAE, T_TRAINING_SAE_CONFIG]):
         loss = output.loss.item()
 
         # metrics for currents acts
-        l0 = (feature_acts > 0).float().sum(-1).mean()
+        l0 = feature_acts.bool().float().sum(-1).to_dense().mean()
         current_learning_rate = self.optimizer.param_groups[0]["lr"]
 
         per_token_l2_loss = (sae_out - sae_in).pow(2).sum(dim=-1).squeeze()

sae_lens/training/types.py

@@ -1,4 +1,4 @@
-from typing import Iterator
+from collections.abc import Iterator
 
 import torch
 

sae_lens/util.py

@@ -5,6 +5,8 @@ from dataclasses import asdict, fields, is_dataclass
 from pathlib import Path
 from typing import Sequence, TypeVar
 
+from transformers import PreTrainedTokenizerBase
+
 K = TypeVar("K")
 V = TypeVar("V")
 
@@ -63,3 +65,28 @@ def path_or_tmp_dir(path: str | Path | None):
             yield Path(td)
     else:
         yield Path(path)
+
+
+def get_special_token_ids(tokenizer: PreTrainedTokenizerBase) -> list[int]:
+    """Get all special token IDs from a tokenizer."""
+    special_tokens = set()
+
+    # Get special tokens from tokenizer attributes
+    for attr in dir(tokenizer):
+        if attr.endswith("_token_id"):
+            token_id = getattr(tokenizer, attr)
+            if token_id is not None:
+                special_tokens.add(token_id)
+
+    # Get any additional special tokens from the tokenizer's special tokens map
+    if hasattr(tokenizer, "special_tokens_map"):
+        for token in tokenizer.special_tokens_map.values():
+            if isinstance(token, str):
+                token_id = tokenizer.convert_tokens_to_ids(token)  # type: ignore
+                special_tokens.add(token_id)
+            elif isinstance(token, list):
+                for t in token:
+                    token_id = tokenizer.convert_tokens_to_ids(t)  # type: ignore
+                    special_tokens.add(token_id)
+
+    return list(special_tokens)

{sae_lens-6.12.1.dist-info → sae_lens-6.21.0.dist-info}/METADATA

@@ -1,8 +1,9 @@
-Metadata-Version: 2.3
+Metadata-Version: 2.4
 Name: sae-lens
-Version: 6.12.1
+Version: 6.21.0
 Summary: Training and Analyzing Sparse Autoencoders (SAEs)
 License: MIT
+License-File: LICENSE
 Keywords: deep-learning,sparse-autoencoders,mechanistic-interpretability,PyTorch
 Author: Joseph Bloom
 Requires-Python: >=3.10,<4.0
@@ -12,6 +13,7 @@ Classifier: Programming Language :: Python :: 3.10
 Classifier: Programming Language :: Python :: 3.11
 Classifier: Programming Language :: Python :: 3.12
 Classifier: Programming Language :: Python :: 3.13
+Classifier: Programming Language :: Python :: 3.14
 Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
 Provides-Extra: mamba
 Requires-Dist: babe (>=0.0.7,<0.0.8)
@@ -28,19 +30,19 @@ Requires-Dist: tenacity (>=9.0.0)
 Requires-Dist: transformer-lens (>=2.16.1,<3.0.0)
 Requires-Dist: transformers (>=4.38.1,<5.0.0)
 Requires-Dist: typing-extensions (>=4.10.0,<5.0.0)
-Project-URL: Homepage, https://jbloomaus.github.io/SAELens
-Project-URL: Repository, https://github.com/jbloomAus/SAELens
+Project-URL: Homepage, https://decoderesearch.github.io/SAELens
+Project-URL: Repository, https://github.com/decoderesearch/SAELens
 Description-Content-Type: text/markdown
 
-<img width="1308" alt="Screenshot 2024-03-21 at 3 08 28 pm" src="https://github.com/jbloomAus/mats_sae_training/assets/69127271/209012ec-a779-4036-b4be-7b7739ea87f6">
+<img width="1308" height="532" alt="saes_pic" src="https://github.com/user-attachments/assets/2a5d752f-b261-4ee4-ad5d-ebf282321371" />
 
 # SAE Lens
 
 [![PyPI](https://img.shields.io/pypi/v/sae-lens?color=blue)](https://pypi.org/project/sae-lens/)
 [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
-[![build](https://github.com/jbloomAus/SAELens/actions/workflows/build.yml/badge.svg)](https://github.com/jbloomAus/SAELens/actions/workflows/build.yml)
-[![Deploy Docs](https://github.com/jbloomAus/SAELens/actions/workflows/deploy_docs.yml/badge.svg)](https://github.com/jbloomAus/SAELens/actions/workflows/deploy_docs.yml)
-[![codecov](https://codecov.io/gh/jbloomAus/SAELens/graph/badge.svg?token=N83NGH8CGE)](https://codecov.io/gh/jbloomAus/SAELens)
+[![build](https://github.com/decoderesearch/SAELens/actions/workflows/build.yml/badge.svg)](https://github.com/decoderesearch/SAELens/actions/workflows/build.yml)
+[![Deploy Docs](https://github.com/decoderesearch/SAELens/actions/workflows/deploy_docs.yml/badge.svg)](https://github.com/decoderesearch/SAELens/actions/workflows/deploy_docs.yml)
+[![codecov](https://codecov.io/gh/decoderesearch/SAELens/graph/badge.svg?token=N83NGH8CGE)](https://codecov.io/gh/decoderesearch/SAELens)
 
 SAELens exists to help researchers:
 
@@ -48,7 +50,7 @@ SAELens exists to help researchers:
 - Analyse sparse autoencoders / research mechanistic interpretability.
 - Generate insights which make it easier to create safe and aligned AI systems.
 
-Please refer to the [documentation](https://jbloomaus.github.io/SAELens/) for information on how to:
+Please refer to the [documentation](https://decoderesearch.github.io/SAELens/) for information on how to:
 
 - Download and Analyse pre-trained sparse autoencoders.
 - Train your own sparse autoencoders.
@@ -56,25 +58,25 @@ Please refer to the [documentation](https://jbloomaus.github.io/SAELens/) for in
 
 SAE Lens is the result of many contributors working collectively to improve humanity's understanding of neural networks, many of whom are motivated by a desire to [safeguard humanity from risks posed by artificial intelligence](https://80000hours.org/problem-profiles/artificial-intelligence/).
 
-This library is maintained by [Joseph Bloom](https://www.jbloomaus.com/), [Curt Tigges](https://curttigges.com/), [Anthony Duong](https://github.com/anthonyduong9) and [David Chanin](https://github.com/chanind).
+This library is maintained by [Joseph Bloom](https://www.decoderesearch.com/), [Curt Tigges](https://curttigges.com/), [Anthony Duong](https://github.com/anthonyduong9) and [David Chanin](https://github.com/chanind).
 
 ## Loading Pre-trained SAEs.
 
-Pre-trained SAEs for various models can be imported via SAE Lens. See this [page](https://jbloomaus.github.io/SAELens/sae_table/) in the readme for a list of all SAEs.
+Pre-trained SAEs for various models can be imported via SAE Lens. See this [page](https://decoderesearch.github.io/SAELens/sae_table/) in the readme for a list of all SAEs.
 
 ## Migrating to SAELens v6
 
-The new v6 update is a major refactor to SAELens and changes the way training code is structured. Check out the [migration guide](https://jbloomaus.github.io/SAELens/latest/migrating/) for more details.
+The new v6 update is a major refactor to SAELens and changes the way training code is structured. Check out the [migration guide](https://decoderesearch.github.io/SAELens/latest/migrating/) for more details.
 
 ## Tutorials
 
-- [SAE Lens + Neuronpedia](tutorials/tutorial_2_0.ipynb)[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://githubtocolab.com/jbloomAus/SAELens/blob/main/tutorials/tutorial_2_0.ipynb)
+- [SAE Lens + Neuronpedia](tutorials/tutorial_2_0.ipynb)[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://githubtocolab.com/decoderesearch/SAELens/blob/main/tutorials/tutorial_2_0.ipynb)
 - [Loading and Analysing Pre-Trained Sparse Autoencoders](tutorials/basic_loading_and_analysing.ipynb)
-  [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://githubtocolab.com/jbloomAus/SAELens/blob/main/tutorials/basic_loading_and_analysing.ipynb)
+  [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://githubtocolab.com/decoderesearch/SAELens/blob/main/tutorials/basic_loading_and_analysing.ipynb)
 - [Understanding SAE Features with the Logit Lens](tutorials/logits_lens_with_features.ipynb)
-  [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://githubtocolab.com/jbloomAus/SAELens/blob/main/tutorials/logits_lens_with_features.ipynb)
+  [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://githubtocolab.com/decoderesearch/SAELens/blob/main/tutorials/logits_lens_with_features.ipynb)
 - [Training a Sparse Autoencoder](tutorials/training_a_sparse_autoencoder.ipynb)
-  [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://githubtocolab.com/jbloomAus/SAELens/blob/main/tutorials/training_a_sparse_autoencoder.ipynb)
+  [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://githubtocolab.com/decoderesearch/SAELens/blob/main/tutorials/training_a_sparse_autoencoder.ipynb)
 
 ## Join the Slack!
 
@@ -89,7 +91,7 @@ Please cite the package as follows:
    title = {SAELens},
    author = {Bloom, Joseph and Tigges, Curt and Duong, Anthony and Chanin, David},
    year = {2024},
-   howpublished = {\url{https://github.com/jbloomAus/SAELens}},
+   howpublished = {\url{https://github.com/decoderesearch/SAELens}},
 }
 ```
 

sae_lens-6.21.0.dist-info/RECORD

@@ -0,0 +1,41 @@
+sae_lens/__init__.py,sha256=P97iQ9YQGEa7PjR9_--pkEA-0WS9D_O7BkDNZ_qsxdA,4033
+sae_lens/analysis/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+sae_lens/analysis/hooked_sae_transformer.py,sha256=vRu6JseH1lZaEeILD5bEkQEQ1wYHHDcxD-f2olKmE9Y,14275
+sae_lens/analysis/neuronpedia_integration.py,sha256=Gx1W7hUBEuMoasNcnOnZ1wmqbXDd1pSZ1nqKEya1HQc,4962
+sae_lens/cache_activations_runner.py,sha256=KFR5SHZt_fy3iLBtHd_aBG-fskAqEtRE-7A-nqy1L-w,12588
+sae_lens/config.py,sha256=fxvpQxFfPOVUkryiHD19q9O1AJDSkIguWeYlbJuTxmY,30329
+sae_lens/constants.py,sha256=qX12uAE_xkha6hjss_0MGTbakI7gEkJzHABkZaHWQFU,683
+sae_lens/evals.py,sha256=P0NUsJeGzYxFBiVKhbPzd72IFKY4gH40HHlEZ3jEAmg,39598
+sae_lens/llm_sae_training_runner.py,sha256=M7BK55gSFYu2qFQKABHX3c8i46P1LfODCeyHFzGGuqU,15196
+sae_lens/load_model.py,sha256=C8AMykctj6H7tz_xRwB06-EXj6TfW64PtSJZR5Jxn1Y,8649
+sae_lens/loading/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+sae_lens/loading/pretrained_sae_loaders.py,sha256=X-gVZ4A74E85lSMFMsZ_rEQhHlR9AYFwhxvoA_vt2CQ,56051
+sae_lens/loading/pretrained_saes_directory.py,sha256=hejNfLUepYCSGPalRfQwxxCEUqMMUPsn1tufwvwct5k,3820
+sae_lens/pretokenize_runner.py,sha256=x-reJzVPFDS9iRFbZtrFYSzNguJYki9gd0pbHjYJ3r4,7085
+sae_lens/pretrained_saes.yaml,sha256=trDRsAS7Tlm5LPGVGi2pa2iidrghahtsjL_O2DfSSEw,604874
+sae_lens/registry.py,sha256=nhy7BPSudSATqW4lo9H_k3Na7sfGHmAf9v-3wpnLL_o,1490
+sae_lens/saes/__init__.py,sha256=nTNPnJ7edyfedo1MX96xwn9WOG8504yHbT9LFw9od_0,1778
+sae_lens/saes/batchtopk_sae.py,sha256=x4EbgZl0GUickRPcCmtKNGS2Ra3Uy1Z1OtF2FnrSabQ,5422
+sae_lens/saes/gated_sae.py,sha256=qcmM9JwBA8aZR8z_IRHV1_gQX-q_63tKewWXRnhdXuo,8986
+sae_lens/saes/jumprelu_sae.py,sha256=HHBF1sJ95lZvxwP5vwLSQFKdnJN2KKYK0WAEaLTrta0,13399
+sae_lens/saes/matryoshka_batchtopk_sae.py,sha256=4_1cVaxk6c6jgJEbxqebtG-cjQNIzaMAfjSPGfR7_VU,6062
+sae_lens/saes/sae.py,sha256=i6HwULvCrFQhRqKruCQo2aOY5a7c6FuUX7sD3TbnNTY,38084
+sae_lens/saes/standard_sae.py,sha256=9UqYyYtQuThYxXKNaDjYcyowpOx2-7cShG-TeUP6JCQ,5940
+sae_lens/saes/temporal_sae.py,sha256=jnHBqz3FvKMOw_lH2aGIJqgrMVJbpprronittCGGXPQ,13517
+sae_lens/saes/topk_sae.py,sha256=tzQM5eQFifMe--8_8NUBYWY7hpjQa6A_olNe6U71FE8,21275
+sae_lens/saes/transcoder.py,sha256=BfLSbTYVNZh-ruGxseZiZJ_acEL6_7QyTdfqUr0lDOg,12156
+sae_lens/tokenization_and_batching.py,sha256=D_o7cXvRqhT89H3wNzoRymNALNE6eHojBWLdXOUwUGE,5438
+sae_lens/training/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+sae_lens/training/activation_scaler.py,sha256=FzNfgBplLWmyiSlZ6TUvE-nur3lOiGTrlvC97ys8S24,1973
+sae_lens/training/activations_store.py,sha256=7fWUahMUqrvHhN5b_8Ma6kJX1NkRbIqNuNKzKvYmHFQ,33881
+sae_lens/training/mixing_buffer.py,sha256=vDpYG5ZE70szDvBsRKcNHEES3h_WTKJ16qDYk5jPOVA,2015
+sae_lens/training/optim.py,sha256=bJpqqcK4enkcPvQAJkeH4Ci1LUOlfjIMTv6-IlaAbRA,5588
+sae_lens/training/sae_trainer.py,sha256=zhkabyIKxI_tZTV3_kwz6zMrHZ95Ecr97krmwc-9ffs,17600
+sae_lens/training/types.py,sha256=1FpLx_Doda9vZpmfm-x1e8wGBYpyhe9Kpb_JuM5nIFM,90
+sae_lens/training/upload_saes_to_huggingface.py,sha256=r_WzI1zLtGZ5TzAxuG3xa_8T09j3zXJrWd_vzPsPGkQ,4469
+sae_lens/tutorial/tsea.py,sha256=fd1am_XXsf2KMbByDapJo-2qlxduKaa62Z2qcQZ3QKU,18145
+sae_lens/util.py,sha256=tCovQ-eZa1L7thPpNDL6PGOJrIMML2yLI5e0EHCOpS8,3309
+sae_lens-6.21.0.dist-info/METADATA,sha256=ZgrnVgZyTjJk-XmNZG1xHUJqdXoO0ldyFZgElVWVVno,5369
+sae_lens-6.21.0.dist-info/WHEEL,sha256=zp0Cn7JsFoX2ATtOhtaFYIiE2rmFAD4OcMhtUki8W3U,88
+sae_lens-6.21.0.dist-info/licenses/LICENSE,sha256=DW6e-hDosiu4CfW0-imI57sV1I5f9UEslpviNQcOAKs,1069
+sae_lens-6.21.0.dist-info/RECORD,,

{sae_lens-6.12.1.dist-info → sae_lens-6.21.0.dist-info}/WHEEL

@@ -1,4 +1,4 @@
 Wheel-Version: 1.0
-Generator: poetry-core 2.1.3
+Generator: poetry-core 2.2.1
 Root-Is-Purelib: true
 Tag: py3-none-any

sae_lens-6.12.1.dist-info/RECORD

@@ -1,39 +0,0 @@
-sae_lens/__init__.py,sha256=UbZevK2LHG9rpfVOKdJIGBHkF7lbrV2-dyYWF_x8-fw,3589
-sae_lens/analysis/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-sae_lens/analysis/hooked_sae_transformer.py,sha256=vRu6JseH1lZaEeILD5bEkQEQ1wYHHDcxD-f2olKmE9Y,14275
-sae_lens/analysis/neuronpedia_integration.py,sha256=Gx1W7hUBEuMoasNcnOnZ1wmqbXDd1pSZ1nqKEya1HQc,4962
-sae_lens/cache_activations_runner.py,sha256=cNeAtp2JQ_vKbeddZVM-tcPLYyyfTWL8NDna5KQpkLI,12583
-sae_lens/config.py,sha256=IdRXSKPfYY3hwUovj-u83eep8z52gkJHII0mY0KseYY,28739
-sae_lens/constants.py,sha256=CSjmiZ-bhjQeVLyRvWxAjBokCgkfM8mnvd7-vxLIWTY,639
-sae_lens/evals.py,sha256=4hanbyG8qZLItWqft94F4ZjUoytPVB7fw5s0P4Oi0VE,39504
-sae_lens/llm_sae_training_runner.py,sha256=sJTcDX1bUJJ_jZLUT88-8KUYIAPeUGoXktX68PsBqw0,15137
-sae_lens/load_model.py,sha256=C8AMykctj6H7tz_xRwB06-EXj6TfW64PtSJZR5Jxn1Y,8649
-sae_lens/loading/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-sae_lens/loading/pretrained_sae_loaders.py,sha256=CVzHntSUKR1X3_gAqn8K_Ajq8D85qBrmrgEgU93IV4A,49609
-sae_lens/loading/pretrained_saes_directory.py,sha256=4Vn-Jex6SveD7EbxcSOBv8cx1gkPfUMLU1QOP-ww1ZE,3752
-sae_lens/pretokenize_runner.py,sha256=w0f6SfZLAxbp5eAAKnet8RqUB_DKofZ9RGsoJwFnYbA,7058
-sae_lens/pretrained_saes.yaml,sha256=6ca3geEB6NyhULUrmdtPDK8ea0YdpLp8_au78vIFC5w,602553
-sae_lens/registry.py,sha256=nhy7BPSudSATqW4lo9H_k3Na7sfGHmAf9v-3wpnLL_o,1490
-sae_lens/saes/__init__.py,sha256=jVwazK8Q6dW5J6_zFXPoNAuBvSxgziQ8eMOjGM3t-X8,1475
-sae_lens/saes/batchtopk_sae.py,sha256=GX_J0vH4vzeLqYxl0mkfsZQpFEoCEHMR4dIG8fz8N8w,3449
-sae_lens/saes/gated_sae.py,sha256=qcmM9JwBA8aZR8z_IRHV1_gQX-q_63tKewWXRnhdXuo,8986
-sae_lens/saes/jumprelu_sae.py,sha256=HHBF1sJ95lZvxwP5vwLSQFKdnJN2KKYK0WAEaLTrta0,13399
-sae_lens/saes/sae.py,sha256=gdUZuLaOHQrPjbDj-nZI813B6-_mNAnV9i9z4qTnpHk,38255
-sae_lens/saes/standard_sae.py,sha256=9UqYyYtQuThYxXKNaDjYcyowpOx2-7cShG-TeUP6JCQ,5940
-sae_lens/saes/topk_sae.py,sha256=CXMBI6CFvI5829bOhoQ350VXR9d8uFHUDlULTIWHXoU,8686
-sae_lens/saes/transcoder.py,sha256=BfLSbTYVNZh-ruGxseZiZJ_acEL6_7QyTdfqUr0lDOg,12156
-sae_lens/tokenization_and_batching.py,sha256=now7caLbU3p-iGokNwmqZDyIvxYoXgnG1uklhgiLZN4,4656
-sae_lens/training/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-sae_lens/training/activation_scaler.py,sha256=seEE-2Qd2JMHxqgnsNWPt-DGtYGZxWPnOwCGuVNSOtI,1719
-sae_lens/training/activations_store.py,sha256=2EUY2abqpT5El3T95sypM_JRDgiKL3VeT73U9SQIFGY,32903
-sae_lens/training/mixing_buffer.py,sha256=vDpYG5ZE70szDvBsRKcNHEES3h_WTKJ16qDYk5jPOVA,2015
-sae_lens/training/optim.py,sha256=TiI9nbffzXNsI8WjcIsqa2uheW6suxqL_KDDmWXobWI,5312
-sae_lens/training/sae_trainer.py,sha256=Jh5AyBGtfZjnprv9H3k0p_luWWnM7YFjlmHuO1W_J6U,15465
-sae_lens/training/types.py,sha256=qSjmGzXf3MLalygG0psnVjmhX_mpLmL47MQtZfe7qxg,81
-sae_lens/training/upload_saes_to_huggingface.py,sha256=r_WzI1zLtGZ5TzAxuG3xa_8T09j3zXJrWd_vzPsPGkQ,4469
-sae_lens/tutorial/tsea.py,sha256=fd1am_XXsf2KMbByDapJo-2qlxduKaa62Z2qcQZ3QKU,18145
-sae_lens/util.py,sha256=lW7fBn_b8quvRYlen9PUmB7km60YhKyjmuelB1f6KzQ,2253
-sae_lens-6.12.1.dist-info/LICENSE,sha256=DW6e-hDosiu4CfW0-imI57sV1I5f9UEslpviNQcOAKs,1069
-sae_lens-6.12.1.dist-info/METADATA,sha256=_mH3B2Hw3gzE6m0y6YyW-3yYOY1U3Nrf8ElewET3wjg,5245
-sae_lens-6.12.1.dist-info/WHEEL,sha256=b4K_helf-jlQoXBBETfwnf4B04YC67LOev0jo4fX5m8,88
-sae_lens-6.12.1.dist-info/RECORD,,