Upload folder using huggingface_hub

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

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7f19a696764868d008a6f43c233f66ac92d68da61d245615623fdcbd3489ef30
+size 302197782

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

@@ -0,0 +1,28 @@
+{
+    "trainer": {
+        "dict_class": "GatedAutoEncoder",
+        "trainer_class": "GatedSAETrainer",
+        "activation_dim": 2304,
+        "dict_size": 16384,
+        "lr": 0.0003,
+        "l1_penalty": 0.012,
+        "warmup_steps": 1000,
+        "sparsity_warmup_steps": 5000,
+        "decay_start": 195312,
+        "seed": 3407,
+        "device": "cuda:0",
+        "layer": 12,
+        "lm_name": "google/gemma-2-2b",
+        "wandb_name": "GatedTrainer-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:0"
+    }
+}

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

@@ -0,0 +1 @@
+{"l2_loss": 29.6775, "l1_loss": 2589.2, "l0": 1354.48447265625, "frac_variance_explained": 0.953515625, "cossim": 0.9765625, "l2_ratio": 0.984375, "relative_reconstruction_bias": 1.007734375, "frac_alive": 0.99969482421875, "hyperparameters": {"n_inputs": 200, "context_length": 2048}}

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

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9a3ab5a5de97b3f16874991e2b435844dee10fa4b57449605562e362d7fecc8e
+size 302197782

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

@@ -0,0 +1,28 @@
+{
+    "trainer": {
+        "dict_class": "GatedAutoEncoder",
+        "trainer_class": "GatedSAETrainer",
+        "activation_dim": 2304,
+        "dict_size": 16384,
+        "lr": 0.0003,
+        "l1_penalty": 0.018,
+        "warmup_steps": 1000,
+        "sparsity_warmup_steps": 5000,
+        "decay_start": 195312,
+        "seed": 3407,
+        "device": "cuda:0",
+        "layer": 12,
+        "lm_name": "google/gemma-2-2b",
+        "wandb_name": "GatedTrainer-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:0"
+    }
+}

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

@@ -0,0 +1 @@
+{"l2_loss": 44.1975, "l1_loss": 1712.76, "l0": 826.72779296875, "frac_variance_explained": 0.89998046875, "cossim": 0.94916015625, "l2_ratio": 0.9608984375, "relative_reconstruction_bias": 1.010390625, "frac_alive": 0.9998779296875, "hyperparameters": {"n_inputs": 200, "context_length": 2048}}

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

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:72375f2e644fda363dcf4b597f8b61a133e7f8b5c0795f38d07b21399fb1ab57
+size 302197782

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

@@ -0,0 +1,28 @@
+{
+    "trainer": {
+        "dict_class": "GatedAutoEncoder",
+        "trainer_class": "GatedSAETrainer",
+        "activation_dim": 2304,
+        "dict_size": 16384,
+        "lr": 0.0003,
+        "l1_penalty": 0.024,
+        "warmup_steps": 1000,
+        "sparsity_warmup_steps": 5000,
+        "decay_start": 195312,
+        "seed": 3407,
+        "device": "cuda:0",
+        "layer": 12,
+        "lm_name": "google/gemma-2-2b",
+        "wandb_name": "GatedTrainer-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:0"
+    }
+}

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

@@ -0,0 +1 @@
+{"l2_loss": 51.69, "l1_loss": 1324.48, "l0": 525.338974609375, "frac_variance_explained": 0.8630859375, "cossim": 0.92958984375, "l2_ratio": 0.943828125, "relative_reconstruction_bias": 1.0133984375, "frac_alive": 0.9998779296875, "hyperparameters": {"n_inputs": 200, "context_length": 2048}}

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

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3a71ba9b72dc2fc93a77941dcaddd2b4e4b0a6807df4bd48e6306a2f7a4d1a75
+size 302197782

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

@@ -0,0 +1,28 @@
+{
+    "trainer": {
+        "dict_class": "GatedAutoEncoder",
+        "trainer_class": "GatedSAETrainer",
+        "activation_dim": 2304,
+        "dict_size": 16384,
+        "lr": 0.0003,
+        "l1_penalty": 0.04,
+        "warmup_steps": 1000,
+        "sparsity_warmup_steps": 5000,
+        "decay_start": 195312,
+        "seed": 3407,
+        "device": "cuda:0",
+        "layer": 12,
+        "lm_name": "google/gemma-2-2b",
+        "wandb_name": "GatedTrainer-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:0"
+    }
+}

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

@@ -0,0 +1 @@
+{"l2_loss": 62.25375, "l1_loss": 844.72, "l0": 226.5963427734375, "frac_variance_explained": 0.7987890625, "cossim": 0.89490234375, "l2_ratio": 0.91029296875, "relative_reconstruction_bias": 1.0123828125, "frac_alive": 0.99810791015625, "hyperparameters": {"n_inputs": 200, "context_length": 2048}}

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

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fa3b7c26b3e5f56e627e1ff9ceb299bd0321901518ab081a235b9b261d2c6c0b
+size 302197782

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

@@ -0,0 +1,28 @@
+{
+    "trainer": {
+        "dict_class": "GatedAutoEncoder",
+        "trainer_class": "GatedSAETrainer",
+        "activation_dim": 2304,
+        "dict_size": 16384,
+        "lr": 0.0003,
+        "l1_penalty": 0.06,
+        "warmup_steps": 1000,
+        "sparsity_warmup_steps": 5000,
+        "decay_start": 195312,
+        "seed": 3407,
+        "device": "cuda:0",
+        "layer": 12,
+        "lm_name": "google/gemma-2-2b",
+        "wandb_name": "GatedTrainer-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:0"
+    }
+}

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

@@ -0,0 +1 @@
+{"l2_loss": 69.3725, "l1_loss": 586.44, "l0": 112.71653076171874, "frac_variance_explained": 0.75146484375, "cossim": 0.867578125, "l2_ratio": 0.88220703125, "relative_reconstruction_bias": 1.013046875, "frac_alive": 0.9769287109375, "hyperparameters": {"n_inputs": 200, "context_length": 2048}}

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

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:570f87b4d33198961a055d2d5500083a748cbf702923209d6939fbadde293b16
+size 302197782

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

@@ -0,0 +1,28 @@
+{
+    "trainer": {
+        "dict_class": "GatedAutoEncoder",
+        "trainer_class": "GatedSAETrainer",
+        "activation_dim": 2304,
+        "dict_size": 16384,
+        "lr": 0.0003,
+        "l1_penalty": 0.08,
+        "warmup_steps": 1000,
+        "sparsity_warmup_steps": 5000,
+        "decay_start": 195312,
+        "seed": 3407,
+        "device": "cuda:0",
+        "layer": 12,
+        "lm_name": "google/gemma-2-2b",
+        "wandb_name": "GatedTrainer-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:0"
+    }
+}

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

@@ -0,0 +1 @@
+{"l2_loss": 74.2625, "l1_loss": 448.28, "l0": 67.0883837890625, "frac_variance_explained": 0.714453125, "cossim": 0.8471875, "l2_ratio": 0.8592578125, "relative_reconstruction_bias": 1.01453125, "frac_alive": 0.89825439453125, "hyperparameters": {"n_inputs": 200, "context_length": 2048}}

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

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a5a55ae262a22da69f33d6cac4392d200e6526822dc7780f90b356083108b4ca
+size 302066710

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

@@ -0,0 +1,31 @@
+{
+    "trainer": {
+        "trainer_class": "TopKTrainer",
+        "dict_class": "AutoEncoderTopK",
+        "lr": 0.0003,
+        "steps": 244140,
+        "auxk_alpha": 0.03125,
+        "warmup_steps": 1000,
+        "decay_start": 195312,
+        "threshold_beta": 0.999,
+        "threshold_start_step": 1000,
+        "seed": 3407,
+        "activation_dim": 2304,
+        "dict_size": 16384,
+        "k": 80,
+        "device": "cuda:0",
+        "layer": 12,
+        "lm_name": "google/gemma-2-2b",
+        "wandb_name": "TopKTrainer-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:0"
+    }
+}

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

@@ -0,0 +1 @@
+{"l2_loss": 70.7575, "l1_loss": 536.3, "l0": 79.93999267578126, "frac_variance_explained": 0.7394140625, "cossim": 0.8617578125, "l2_ratio": 0.8646875, "relative_reconstruction_bias": 1.0025, "frac_alive": 0.98187255859375, "hyperparameters": {"n_inputs": 200, "context_length": 2048}}

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

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:71e6ef677dbce70b528ff57caaaf8512e9d15b57c0515f045eba3d46f7e5c132
+size 302066710

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

@@ -0,0 +1,31 @@
+{
+    "trainer": {
+        "trainer_class": "TopKTrainer",
+        "dict_class": "AutoEncoderTopK",
+        "lr": 0.0003,
+        "steps": 244140,
+        "auxk_alpha": 0.03125,
+        "warmup_steps": 1000,
+        "decay_start": 195312,
+        "threshold_beta": 0.999,
+        "threshold_start_step": 1000,
+        "seed": 3407,
+        "activation_dim": 2304,
+        "dict_size": 16384,
+        "k": 160,
+        "device": "cuda:0",
+        "layer": 12,
+        "lm_name": "google/gemma-2-2b",
+        "wandb_name": "TopKTrainer-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:0"
+    }
+}

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

@@ -0,0 +1 @@
+{"l2_loss": 64.27, "l1_loss": 870.32, "l0": 159.8549658203125, "frac_variance_explained": 0.78310546875, "cossim": 0.88716796875, "l2_ratio": 0.888515625, "relative_reconstruction_bias": 0.9996875, "frac_alive": 0.990478515625, "hyperparameters": {"n_inputs": 200, "context_length": 2048}}