Upload folder using huggingface_hub

trained_saes__google_gemma-2-2b_jump_relu_batch_top_k_standard_new/resid_post_layer_12/trainer_0/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:16c490d578682a47914c761e9b33177806667b97c200df037b64ddb652a7d65e
+size 302066344

trained_saes__google_gemma-2-2b_jump_relu_batch_top_k_standard_new/resid_post_layer_12/trainer_0/config.json

@@ -0,0 +1,29 @@
+{
+    "trainer": {
+        "dict_class": "AutoEncoder",
+        "trainer_class": "StandardTrainerAprilUpdate",
+        "activation_dim": 2304,
+        "dict_size": 16384,
+        "lr": 0.0003,
+        "l1_penalty": 0.012,
+        "warmup_steps": 1000,
+        "sparsity_warmup_steps": 5000,
+        "steps": 244140,
+        "decay_start": 195312,
+        "seed": 3407,
+        "device": "cuda:1",
+        "layer": 12,
+        "lm_name": "google/gemma-2-2b",
+        "wandb_name": "StandardTrainerNew-google/gemma-2-2b-resid_post_layer_12_trainer_0",
+        "submodule_name": "resid_post_layer_12"
+    },
+    "buffer": {
+        "d_submodule": 2304,
+        "io": "out",
+        "n_ctxs": 122,
+        "ctx_len": 2048,
+        "refresh_batch_size": 4,
+        "out_batch_size": 2048,
+        "device": "cuda:1"
+    }
+}

trained_saes__google_gemma-2-2b_jump_relu_batch_top_k_standard_new/resid_post_layer_12/trainer_0/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 53.96, "l1_loss": 1025.8, "l0": 1096.11625, "frac_variance_explained": 0.84216796875, "cossim": 0.9234375, "l2_ratio": 0.893203125, "relative_reconstruction_bias": 0.999296875, "frac_alive": 0.801025390625, "hyperparameters": {"n_inputs": 200, "context_length": 2048}}

trained_saes__google_gemma-2-2b_jump_relu_batch_top_k_standard_new/resid_post_layer_12/trainer_1/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ab48e7fd50bd88ed534bec2bb0a3e377e6f1ffbcea4c56e696474b29acf71a96
+size 302066344

trained_saes__google_gemma-2-2b_jump_relu_batch_top_k_standard_new/resid_post_layer_12/trainer_1/config.json

@@ -0,0 +1,29 @@
+{
+    "trainer": {
+        "dict_class": "AutoEncoder",
+        "trainer_class": "StandardTrainerAprilUpdate",
+        "activation_dim": 2304,
+        "dict_size": 16384,
+        "lr": 0.0003,
+        "l1_penalty": 0.015,
+        "warmup_steps": 1000,
+        "sparsity_warmup_steps": 5000,
+        "steps": 244140,
+        "decay_start": 195312,
+        "seed": 3407,
+        "device": "cuda:1",
+        "layer": 12,
+        "lm_name": "google/gemma-2-2b",
+        "wandb_name": "StandardTrainerNew-google/gemma-2-2b-resid_post_layer_12_trainer_1",
+        "submodule_name": "resid_post_layer_12"
+    },
+    "buffer": {
+        "d_submodule": 2304,
+        "io": "out",
+        "n_ctxs": 122,
+        "ctx_len": 2048,
+        "refresh_batch_size": 4,
+        "out_batch_size": 2048,
+        "device": "cuda:1"
+    }
+}

trained_saes__google_gemma-2-2b_jump_relu_batch_top_k_standard_new/resid_post_layer_12/trainer_1/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 59.43375, "l1_loss": 816.34, "l0": 776.0542138671875, "frac_variance_explained": 0.80630859375, "cossim": 0.90646484375, "l2_ratio": 0.87634765625, "relative_reconstruction_bias": 1.0092578125, "frac_alive": 0.7996826171875, "hyperparameters": {"n_inputs": 200, "context_length": 2048}}

trained_saes__google_gemma-2-2b_jump_relu_batch_top_k_standard_new/resid_post_layer_12/trainer_2/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:091e077b90ee0d0651f420ee302473fdfa5e51e40a6831f2a889bdf90a13e1f9
+size 302066344

trained_saes__google_gemma-2-2b_jump_relu_batch_top_k_standard_new/resid_post_layer_12/trainer_2/config.json

@@ -0,0 +1,29 @@
+{
+    "trainer": {
+        "dict_class": "AutoEncoder",
+        "trainer_class": "StandardTrainerAprilUpdate",
+        "activation_dim": 2304,
+        "dict_size": 16384,
+        "lr": 0.0003,
+        "l1_penalty": 0.02,
+        "warmup_steps": 1000,
+        "sparsity_warmup_steps": 5000,
+        "steps": 244140,
+        "decay_start": 195312,
+        "seed": 3407,
+        "device": "cuda:1",
+        "layer": 12,
+        "lm_name": "google/gemma-2-2b",
+        "wandb_name": "StandardTrainerNew-google/gemma-2-2b-resid_post_layer_12_trainer_2",
+        "submodule_name": "resid_post_layer_12"
+    },
+    "buffer": {
+        "d_submodule": 2304,
+        "io": "out",
+        "n_ctxs": 122,
+        "ctx_len": 2048,
+        "refresh_batch_size": 4,
+        "out_batch_size": 2048,
+        "device": "cuda:1"
+    }
+}

trained_saes__google_gemma-2-2b_jump_relu_batch_top_k_standard_new/resid_post_layer_12/trainer_2/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 65.825, "l1_loss": 605.46, "l0": 475.75117431640626, "frac_variance_explained": 0.7558203125, "cossim": 0.884453125, "l2_ratio": 0.8563671875, "relative_reconstruction_bias": 1.0234375, "frac_alive": 0.79644775390625, "hyperparameters": {"n_inputs": 200, "context_length": 2048}}

trained_saes__google_gemma-2-2b_jump_relu_batch_top_k_standard_new/resid_post_layer_12/trainer_3/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b099cf0478630f138aa8112610cc9a26117523dfd9980452d15d9b6df8b7b0b0
+size 302066344

trained_saes__google_gemma-2-2b_jump_relu_batch_top_k_standard_new/resid_post_layer_12/trainer_3/config.json

@@ -0,0 +1,29 @@
+{
+    "trainer": {
+        "dict_class": "AutoEncoder",
+        "trainer_class": "StandardTrainerAprilUpdate",
+        "activation_dim": 2304,
+        "dict_size": 16384,
+        "lr": 0.0003,
+        "l1_penalty": 0.03,
+        "warmup_steps": 1000,
+        "sparsity_warmup_steps": 5000,
+        "steps": 244140,
+        "decay_start": 195312,
+        "seed": 3407,
+        "device": "cuda:1",
+        "layer": 12,
+        "lm_name": "google/gemma-2-2b",
+        "wandb_name": "StandardTrainerNew-google/gemma-2-2b-resid_post_layer_12_trainer_3",
+        "submodule_name": "resid_post_layer_12"
+    },
+    "buffer": {
+        "d_submodule": 2304,
+        "io": "out",
+        "n_ctxs": 122,
+        "ctx_len": 2048,
+        "refresh_batch_size": 4,
+        "out_batch_size": 2048,
+        "device": "cuda:1"
+    }
+}

trained_saes__google_gemma-2-2b_jump_relu_batch_top_k_standard_new/resid_post_layer_12/trainer_3/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 73.47, "l1_loss": 405.03, "l0": 230.87955810546876, "frac_variance_explained": 0.6711328125, "cossim": 0.85509765625, "l2_ratio": 0.82607421875, "relative_reconstruction_bias": 1.06302734375, "frac_alive": 0.79315185546875, "hyperparameters": {"n_inputs": 200, "context_length": 2048}}

trained_saes__google_gemma-2-2b_jump_relu_batch_top_k_standard_new/resid_post_layer_12/trainer_4/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a1680f889330350bd9a3b897052e2fc67f7da58d0bdbd7b99a4a4fb6b8f08ae8
+size 302066344

trained_saes__google_gemma-2-2b_jump_relu_batch_top_k_standard_new/resid_post_layer_12/trainer_4/config.json

@@ -0,0 +1,29 @@
+{
+    "trainer": {
+        "dict_class": "AutoEncoder",
+        "trainer_class": "StandardTrainerAprilUpdate",
+        "activation_dim": 2304,
+        "dict_size": 16384,
+        "lr": 0.0003,
+        "l1_penalty": 0.04,
+        "warmup_steps": 1000,
+        "sparsity_warmup_steps": 5000,
+        "steps": 244140,
+        "decay_start": 195312,
+        "seed": 3407,
+        "device": "cuda:1",
+        "layer": 12,
+        "lm_name": "google/gemma-2-2b",
+        "wandb_name": "StandardTrainerNew-google/gemma-2-2b-resid_post_layer_12_trainer_4",
+        "submodule_name": "resid_post_layer_12"
+    },
+    "buffer": {
+        "d_submodule": 2304,
+        "io": "out",
+        "n_ctxs": 122,
+        "ctx_len": 2048,
+        "refresh_batch_size": 4,
+        "out_batch_size": 2048,
+        "device": "cuda:1"
+    }
+}

trained_saes__google_gemma-2-2b_jump_relu_batch_top_k_standard_new/resid_post_layer_12/trainer_4/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 78.0325, "l1_loss": 311.06, "l0": 140.126806640625, "frac_variance_explained": 0.61591796875, "cossim": 0.83541015625, "l2_ratio": 0.80708984375, "relative_reconstruction_bias": 1.08521484375, "frac_alive": 0.7899169921875, "hyperparameters": {"n_inputs": 200, "context_length": 2048}}

trained_saes__google_gemma-2-2b_jump_relu_batch_top_k_standard_new/resid_post_layer_12/trainer_5/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:28dab4a789420a4cd6e0b77798a76d182ba9a28f6aaeb67da45a2d0a2144d7d1
+size 302066344

trained_saes__google_gemma-2-2b_jump_relu_batch_top_k_standard_new/resid_post_layer_12/trainer_5/config.json

@@ -0,0 +1,29 @@
+{
+    "trainer": {
+        "dict_class": "AutoEncoder",
+        "trainer_class": "StandardTrainerAprilUpdate",
+        "activation_dim": 2304,
+        "dict_size": 16384,
+        "lr": 0.0003,
+        "l1_penalty": 0.06,
+        "warmup_steps": 1000,
+        "sparsity_warmup_steps": 5000,
+        "steps": 244140,
+        "decay_start": 195312,
+        "seed": 3407,
+        "device": "cuda:1",
+        "layer": 12,
+        "lm_name": "google/gemma-2-2b",
+        "wandb_name": "StandardTrainerNew-google/gemma-2-2b-resid_post_layer_12_trainer_5",
+        "submodule_name": "resid_post_layer_12"
+    },
+    "buffer": {
+        "d_submodule": 2304,
+        "io": "out",
+        "n_ctxs": 122,
+        "ctx_len": 2048,
+        "refresh_batch_size": 4,
+        "out_batch_size": 2048,
+        "device": "cuda:1"
+    }
+}

trained_saes__google_gemma-2-2b_jump_relu_batch_top_k_standard_new/resid_post_layer_12/trainer_5/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 84.2, "l1_loss": 222.615, "l0": 72.55794921875, "frac_variance_explained": 0.51142578125, "cossim": 0.80783203125, "l2_ratio": 0.78095703125, "relative_reconstruction_bias": 1.15984375, "frac_alive": 0.78204345703125, "hyperparameters": {"n_inputs": 200, "context_length": 2048}}

trained_saes__google_gemma-2-2b_jump_relu_batch_top_k_standard_new/resid_post_layer_12/trainer_6/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b7d89e46005c4a0f9e07cf5e3efe1f406901c6f87c26e001eab32f7e940c2a53
+size 302066710

trained_saes__google_gemma-2-2b_jump_relu_batch_top_k_standard_new/resid_post_layer_12/trainer_6/config.json

@@ -0,0 +1,32 @@
+{
+    "trainer": {
+        "trainer_class": "BatchTopKTrainer",
+        "dict_class": "BatchTopKSAE",
+        "lr": 0.0003,
+        "steps": 244140,
+        "auxk_alpha": 0.03125,
+        "warmup_steps": 1000,
+        "decay_start": 195312,
+        "threshold_beta": 0.999,
+        "threshold_start_step": 1000,
+        "top_k_aux": 1152,
+        "seed": 3407,
+        "activation_dim": 2304,
+        "dict_size": 16384,
+        "k": 80,
+        "device": "cuda:1",
+        "layer": 12,
+        "lm_name": "google/gemma-2-2b",
+        "wandb_name": "BatchTopKTrainer-google/gemma-2-2b-resid_post_layer_12_trainer_6",
+        "submodule_name": "resid_post_layer_12"
+    },
+    "buffer": {
+        "d_submodule": 2304,
+        "io": "out",
+        "n_ctxs": 122,
+        "ctx_len": 2048,
+        "refresh_batch_size": 4,
+        "out_batch_size": 2048,
+        "device": "cuda:1"
+    }
+}

trained_saes__google_gemma-2-2b_jump_relu_batch_top_k_standard_new/resid_post_layer_12/trainer_6/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 68.2225, "l1_loss": 653.98, "l0": 118.34955322265625, "frac_variance_explained": 0.75828125, "cossim": 0.8718359375, "l2_ratio": 0.88041015625, "relative_reconstruction_bias": 1.0090234375, "frac_alive": 0.957275390625, "hyperparameters": {"n_inputs": 200, "context_length": 2048}}

trained_saes__google_gemma-2-2b_jump_relu_batch_top_k_standard_new/resid_post_layer_12/trainer_7/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dcf1b6a6f2955b893b4f31e53adbd8ff813ef7d14da048c3f44bb2f50a5a7858
+size 302066710

trained_saes__google_gemma-2-2b_jump_relu_batch_top_k_standard_new/resid_post_layer_12/trainer_7/config.json

@@ -0,0 +1,32 @@
+{
+    "trainer": {
+        "trainer_class": "BatchTopKTrainer",
+        "dict_class": "BatchTopKSAE",
+        "lr": 0.0003,
+        "steps": 244140,
+        "auxk_alpha": 0.03125,
+        "warmup_steps": 1000,
+        "decay_start": 195312,
+        "threshold_beta": 0.999,
+        "threshold_start_step": 1000,
+        "top_k_aux": 1152,
+        "seed": 3407,
+        "activation_dim": 2304,
+        "dict_size": 16384,
+        "k": 160,
+        "device": "cuda:1",
+        "layer": 12,
+        "lm_name": "google/gemma-2-2b",
+        "wandb_name": "BatchTopKTrainer-google/gemma-2-2b-resid_post_layer_12_trainer_7",
+        "submodule_name": "resid_post_layer_12"
+    },
+    "buffer": {
+        "d_submodule": 2304,
+        "io": "out",
+        "n_ctxs": 122,
+        "ctx_len": 2048,
+        "refresh_batch_size": 4,
+        "out_batch_size": 2048,
+        "device": "cuda:1"
+    }
+}

trained_saes__google_gemma-2-2b_jump_relu_batch_top_k_standard_new/resid_post_layer_12/trainer_7/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 60.84, "l1_loss": 1058.6, "l0": 240.37475830078125, "frac_variance_explained": 0.80884765625, "cossim": 0.898828125, "l2_ratio": 0.904375, "relative_reconstruction_bias": 1.006015625, "frac_alive": 0.93707275390625, "hyperparameters": {"n_inputs": 200, "context_length": 2048}}

trained_saes__google_gemma-2-2b_jump_relu_batch_top_k_standard_new/resid_post_layer_12/trainer_8/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7ca96ae26c3395de8275063e4c47c1458ec1505e2c614ee97787c8ebd252a376
+size 302131999

trained_saes__google_gemma-2-2b_jump_relu_batch_top_k_standard_new/resid_post_layer_12/trainer_8/config.json

@@ -0,0 +1,29 @@
+{
+    "trainer": {
+        "trainer_class": "JumpReluTrainer",
+        "dict_class": "JumpReluAutoEncoder",
+        "lr": 0.0003,
+        "steps": 244140,
+        "seed": 3407,
+        "activation_dim": 2304,
+        "dict_size": 16384,
+        "device": "cuda:1",
+        "layer": 12,
+        "lm_name": "google/gemma-2-2b",
+        "wandb_name": "JumpReluTrainer-google/gemma-2-2b-resid_post_layer_12_trainer_8",
+        "submodule_name": "resid_post_layer_12",
+        "bandwidth": 0.001,
+        "sparsity_penalty": 1.0,
+        "sparsity_warmup_steps": 5000,
+        "target_l0": 80
+    },
+    "buffer": {
+        "d_submodule": 2304,
+        "io": "out",
+        "n_ctxs": 122,
+        "ctx_len": 2048,
+        "refresh_batch_size": 4,
+        "out_batch_size": 2048,
+        "device": "cuda:1"
+    }
+}

trained_saes__google_gemma-2-2b_jump_relu_batch_top_k_standard_new/resid_post_layer_12/trainer_8/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 68.04, "l1_loss": 670.48, "l0": 122.69531494140625, "frac_variance_explained": 0.759765625, "cossim": 0.8728515625, "l2_ratio": 0.88236328125, "relative_reconstruction_bias": 1.0092578125, "frac_alive": 0.93853759765625, "hyperparameters": {"n_inputs": 200, "context_length": 2048}}

trained_saes__google_gemma-2-2b_jump_relu_batch_top_k_standard_new/resid_post_layer_12/trainer_9/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0f4199349113fc6417bb9e5569291888b7350d3a22aa0f5e3f38bb272d28f29d
+size 302131999

trained_saes__google_gemma-2-2b_jump_relu_batch_top_k_standard_new/resid_post_layer_12/trainer_9/config.json

@@ -0,0 +1,29 @@
+{
+    "trainer": {
+        "trainer_class": "JumpReluTrainer",
+        "dict_class": "JumpReluAutoEncoder",
+        "lr": 0.0003,
+        "steps": 244140,
+        "seed": 3407,
+        "activation_dim": 2304,
+        "dict_size": 16384,
+        "device": "cuda:1",
+        "layer": 12,
+        "lm_name": "google/gemma-2-2b",
+        "wandb_name": "JumpReluTrainer-google/gemma-2-2b-resid_post_layer_12_trainer_9",
+        "submodule_name": "resid_post_layer_12",
+        "bandwidth": 0.001,
+        "sparsity_penalty": 1.0,
+        "sparsity_warmup_steps": 5000,
+        "target_l0": 160
+    },
+    "buffer": {
+        "d_submodule": 2304,
+        "io": "out",
+        "n_ctxs": 122,
+        "ctx_len": 2048,
+        "refresh_batch_size": 4,
+        "out_batch_size": 2048,
+        "device": "cuda:1"
+    }
+}

trained_saes__google_gemma-2-2b_jump_relu_batch_top_k_standard_new/resid_post_layer_12/trainer_9/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 59.90375, "l1_loss": 987.2, "l0": 252.56128662109376, "frac_variance_explained": 0.81595703125, "cossim": 0.9024609375, "l2_ratio": 0.90943359375, "relative_reconstruction_bias": 1.0080859375, "frac_alive": 0.98199462890625, "hyperparameters": {"n_inputs": 200, "context_length": 2048}}