Upload folder using huggingface_hub

._test_run_EleutherAI_pythia-70m-deduped_top_k_sampled_sae/resid_post_layer_3/SampledActivationTrainer-entropy-EleutherAI_pythia-70m-deduped-resid_post_layer_3_trainer_2/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7a663a086b7460f06df1f8560ac2df2489bdcd4a97b4260524b03fbece3238d5
+size 268964962

._test_run_EleutherAI_pythia-70m-deduped_top_k_sampled_sae/resid_post_layer_3/SampledActivationTrainer-entropy-EleutherAI_pythia-70m-deduped-resid_post_layer_3_trainer_2/config.json

@@ -0,0 +1,36 @@
+{
+    "trainer": {
+        "trainer_class": "SampledActivationTrainer",
+        "dict_class": "SampledActivationSAE",
+        "lr": 0.0003,
+        "steps": 48,
+        "auxk_alpha": 0.03125,
+        "warmup_steps": 1000,
+        "decay_start": 38,
+        "threshold_beta": 0.999,
+        "threshold_start_step": 1000,
+        "sampling_update_freq": 1,
+        "sampling_method": "entropy",
+        "ridge_lambda": 0.01,
+        "sketching_size": 100,
+        "top_k_aux": 256,
+        "seed": 0,
+        "activation_dim": 512,
+        "dict_size": 65536,
+        "k": 20,
+        "device": "cuda:0",
+        "layer": 3,
+        "lm_name": "EleutherAI/pythia-70m-deduped",
+        "wandb_name": "SampledActivationTrainer-entropy-EleutherAI/pythia-70m-deduped-resid_post_layer_3_trainer_2",
+        "submodule_name": "resid_post_layer_3"
+    },
+    "buffer": {
+        "d_submodule": 512,
+        "io": "out",
+        "n_ctxs": 244,
+        "ctx_len": 1024,
+        "refresh_batch_size": 64,
+        "out_batch_size": 2048,
+        "device": "cuda:0"
+    }
+}

._test_run_EleutherAI_pythia-70m-deduped_top_k_sampled_sae/resid_post_layer_3/SampledActivationTrainer-entropy-EleutherAI_pythia-70m-deduped-resid_post_layer_3_trainer_2/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 767.1420555114746, "l1_loss": 14069.71826171875, "l0": 32533.035400390625, "frac_variance_explained": -3754.4764709472656, "cossim": 0.8404335379600525, "l2_ratio": 53.196314573287964, "relative_reconstruction_bias": 62.36626052856445, "loss_original": 2.997953861951828, "loss_reconstructed": 18.35732865333557, "loss_zero": 12.705852329730988, "frac_recovered": -0.5863620117306709, "frac_alive": 1.0, "hyperparameters": {"n_inputs": 200, "context_length": 1024}}

._test_run_EleutherAI_pythia-70m-deduped_top_k_sampled_sae/resid_post_layer_3/SampledActivationTrainer-leverage-EleutherAI_pythia-70m-deduped-resid_post_layer_3_trainer_1/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7abec87c5a8d6e0c50e00652a056c7d172d1748ea2682e1a206d92db9b6a49ad
+size 268964962

._test_run_EleutherAI_pythia-70m-deduped_top_k_sampled_sae/resid_post_layer_3/SampledActivationTrainer-leverage-EleutherAI_pythia-70m-deduped-resid_post_layer_3_trainer_1/config.json

@@ -0,0 +1,36 @@
+{
+    "trainer": {
+        "trainer_class": "SampledActivationTrainer",
+        "dict_class": "SampledActivationSAE",
+        "lr": 0.0003,
+        "steps": 48,
+        "auxk_alpha": 0.03125,
+        "warmup_steps": 1000,
+        "decay_start": 38,
+        "threshold_beta": 0.999,
+        "threshold_start_step": 1000,
+        "sampling_update_freq": 1,
+        "sampling_method": "leverage",
+        "ridge_lambda": 0.01,
+        "sketching_size": 100,
+        "top_k_aux": 256,
+        "seed": 0,
+        "activation_dim": 512,
+        "dict_size": 65536,
+        "k": 20,
+        "device": "cuda:0",
+        "layer": 3,
+        "lm_name": "EleutherAI/pythia-70m-deduped",
+        "wandb_name": "SampledActivationTrainer-leverage-EleutherAI/pythia-70m-deduped-resid_post_layer_3_trainer_1",
+        "submodule_name": "resid_post_layer_3"
+    },
+    "buffer": {
+        "d_submodule": 512,
+        "io": "out",
+        "n_ctxs": 244,
+        "ctx_len": 1024,
+        "refresh_batch_size": 64,
+        "out_batch_size": 2048,
+        "device": "cuda:0"
+    }
+}

._test_run_EleutherAI_pythia-70m-deduped_top_k_sampled_sae/resid_post_layer_3/SampledActivationTrainer-leverage-EleutherAI_pythia-70m-deduped-resid_post_layer_3_trainer_1/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 769.8623352050781, "l1_loss": 14121.056091308594, "l0": 32569.235229492188, "frac_variance_explained": -3766.6612548828125, "cossim": 0.8414181843400002, "l2_ratio": 53.31753587722778, "relative_reconstruction_bias": 62.42244005203247, "loss_original": 2.997953861951828, "loss_reconstructed": 18.36150825023651, "loss_zero": 12.705852329730988, "frac_recovered": -0.5867955405265093, "frac_alive": 1.0, "hyperparameters": {"n_inputs": 200, "context_length": 1024}}

._test_run_EleutherAI_pythia-70m-deduped_top_k_sampled_sae/resid_post_layer_3/TopKTrainer-EleutherAI_pythia-70m-deduped-resid_post_layer_3_trainer_0/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f6a919db78dafaa29370b2e7bf06c9e8afe1f3d45cf37d3e51259613f706127e
+size 268701718

._test_run_EleutherAI_pythia-70m-deduped_top_k_sampled_sae/resid_post_layer_3/TopKTrainer-EleutherAI_pythia-70m-deduped-resid_post_layer_3_trainer_0/config.json

@@ -0,0 +1,31 @@
+{
+    "trainer": {
+        "trainer_class": "TopKTrainer",
+        "dict_class": "AutoEncoderTopK",
+        "lr": 0.0003,
+        "steps": 48,
+        "auxk_alpha": 0.03125,
+        "warmup_steps": 1000,
+        "decay_start": 38,
+        "threshold_beta": 0.999,
+        "threshold_start_step": 1000,
+        "seed": 0,
+        "activation_dim": 512,
+        "dict_size": 65536,
+        "k": 20,
+        "device": "cuda:0",
+        "layer": 3,
+        "lm_name": "EleutherAI/pythia-70m-deduped",
+        "wandb_name": "TopKTrainer-EleutherAI/pythia-70m-deduped-resid_post_layer_3_trainer_0",
+        "submodule_name": "resid_post_layer_3"
+    },
+    "buffer": {
+        "d_submodule": 512,
+        "io": "out",
+        "n_ctxs": 244,
+        "ctx_len": 1024,
+        "refresh_batch_size": 64,
+        "out_batch_size": 2048,
+        "device": "cuda:0"
+    }
+}

._test_run_EleutherAI_pythia-70m-deduped_top_k_sampled_sae/resid_post_layer_3/TopKTrainer-EleutherAI_pythia-70m-deduped-resid_post_layer_3_trainer_0/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 10.643135368824005, "l1_loss": 39.116480588912964, "l0": 20.0, "frac_variance_explained": 0.2406930886209011, "cossim": 0.7164953276515007, "l2_ratio": 0.9136376194655895, "relative_reconstruction_bias": 1.275299347937107, "loss_original": 2.997953861951828, "loss_reconstructed": 5.598592609167099, "loss_zero": 12.705852329730988, "frac_recovered": 0.732391856610775, "frac_alive": 0.99761962890625, "hyperparameters": {"n_inputs": 200, "context_length": 1024}}