Add files using upload-large-folder tool

resid_post_layer_11/trainer_0/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8b21124c04f4ea14be614697239460f06149e0579ed5ddae1e6d1f79ddef79d3
+size 4295510533

resid_post_layer_11/trainer_0/config.json

@@ -0,0 +1,33 @@
+{
+    "trainer": {
+        "trainer_class": "BatchTopKTrainer",
+        "dict_class": "BatchTopKSAE",
+        "lr": 0.0001,
+        "steps": 244140,
+        "auxk_alpha": 0.03125,
+        "warmup_steps": 1000,
+        "decay_start": 195312,
+        "threshold_beta": 0.999,
+        "threshold_start_step": 1000,
+        "top_k_aux": 2048,
+        "seed": 0,
+        "activation_dim": 4096,
+        "dict_size": 131072,
+        "k": 32,
+        "device": "cuda",
+        "layer": 11,
+        "lm_name": "meta-llama/Llama-3.1-8B-Instruct",
+        "wandb_name": "3_l11-BatchTopKTrainer-meta-llama/Llama-3.1-8B-Instruct-resid_post_layer_11_trainer_0",
+        "submodule_name": "resid_post_layer_11"
+    },
+    "buffer": {
+        "d_submodule": 4096,
+        "io": "out",
+        "n_ctxs": 244,
+        "ctx_len": 1024,
+        "refresh_batch_size": 16,
+        "out_batch_size": 2048,
+        "device": "cuda",
+        "internal_device": "cuda"
+    }
+}

resid_post_layer_11/trainer_0/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 3.684375, "l1_loss": 18.024375, "l0": 30.123955078125, "frac_variance_explained": 0.688828125, "cossim": 0.86130859375, "l2_ratio": 0.86015625, "relative_reconstruction_bias": 1.00130859375, "loss_original": 2.1476953125, "loss_reconstructed": 2.2249609375, "loss_zero": 10.1440625, "frac_recovered": 0.99142578125, "frac_alive": 0.69793701171875, "hyperparameters": {"n_inputs": 200, "context_length": 1024}}

resid_post_layer_11/trainer_1/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:91ddda1674e2ebc04626de7ac2eb21346e2e2e31a8866f4519926de6836b8070
+size 4295510533

resid_post_layer_11/trainer_1/config.json

@@ -0,0 +1,33 @@
+{
+    "trainer": {
+        "trainer_class": "BatchTopKTrainer",
+        "dict_class": "BatchTopKSAE",
+        "lr": 0.0001,
+        "steps": 244140,
+        "auxk_alpha": 0.03125,
+        "warmup_steps": 1000,
+        "decay_start": 195312,
+        "threshold_beta": 0.999,
+        "threshold_start_step": 1000,
+        "top_k_aux": 2048,
+        "seed": 0,
+        "activation_dim": 4096,
+        "dict_size": 131072,
+        "k": 64,
+        "device": "cuda",
+        "layer": 11,
+        "lm_name": "meta-llama/Llama-3.1-8B-Instruct",
+        "wandb_name": "3_l11-BatchTopKTrainer-meta-llama/Llama-3.1-8B-Instruct-resid_post_layer_11_trainer_1",
+        "submodule_name": "resid_post_layer_11"
+    },
+    "buffer": {
+        "d_submodule": 4096,
+        "io": "out",
+        "n_ctxs": 244,
+        "ctx_len": 1024,
+        "refresh_batch_size": 16,
+        "out_batch_size": 2048,
+        "device": "cuda",
+        "internal_device": "cuda"
+    }
+}

resid_post_layer_11/trainer_1/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 3.36890625, "l1_loss": 24.319375, "l0": 59.1200146484375, "frac_variance_explained": 0.73994140625, "cossim": 0.88548828125, "l2_ratio": 0.8856640625, "relative_reconstruction_bias": 1.00236328125, "loss_original": 2.15236328125, "loss_reconstructed": 2.19287109375, "loss_zero": 10.14, "frac_recovered": 0.99552734375, "frac_alive": 0.809478759765625, "hyperparameters": {"n_inputs": 200, "context_length": 1024}}

resid_post_layer_11/trainer_2/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:aa8da7920664d083916f264cf403004e5335785b45c513a179782c338967a159
+size 4295510533

resid_post_layer_11/trainer_2/config.json

@@ -0,0 +1,33 @@
+{
+    "trainer": {
+        "trainer_class": "BatchTopKTrainer",
+        "dict_class": "BatchTopKSAE",
+        "lr": 0.0001,
+        "steps": 244140,
+        "auxk_alpha": 0.03125,
+        "warmup_steps": 1000,
+        "decay_start": 195312,
+        "threshold_beta": 0.999,
+        "threshold_start_step": 1000,
+        "top_k_aux": 2048,
+        "seed": 0,
+        "activation_dim": 4096,
+        "dict_size": 131072,
+        "k": 128,
+        "device": "cuda",
+        "layer": 11,
+        "lm_name": "meta-llama/Llama-3.1-8B-Instruct",
+        "wandb_name": "3_l11-BatchTopKTrainer-meta-llama/Llama-3.1-8B-Instruct-resid_post_layer_11_trainer_2",
+        "submodule_name": "resid_post_layer_11"
+    },
+    "buffer": {
+        "d_submodule": 4096,
+        "io": "out",
+        "n_ctxs": 244,
+        "ctx_len": 1024,
+        "refresh_batch_size": 16,
+        "out_batch_size": 2048,
+        "device": "cuda",
+        "internal_device": "cuda"
+    }
+}

resid_post_layer_11/trainer_2/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 3.079453125, "l1_loss": 34.06625, "l0": 116.1669091796875, "frac_variance_explained": 0.781953125, "cossim": 0.905390625, "l2_ratio": 0.9049609375, "relative_reconstruction_bias": 1.00125, "loss_original": 2.15947265625, "loss_reconstructed": 2.18658203125, "loss_zero": 10.151875, "frac_recovered": 0.996875, "frac_alive": 0.84259033203125, "hyperparameters": {"n_inputs": 200, "context_length": 1024}}

resid_post_layer_11/trainer_3/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2e7a847bdde36bb2a9247d777af37d671f7bd306c767a0bb5907598a7e45d637
+size 4295510533

resid_post_layer_11/trainer_3/config.json

@@ -0,0 +1,33 @@
+{
+    "trainer": {
+        "trainer_class": "BatchTopKTrainer",
+        "dict_class": "BatchTopKSAE",
+        "lr": 0.0001,
+        "steps": 244140,
+        "auxk_alpha": 0.03125,
+        "warmup_steps": 1000,
+        "decay_start": 195312,
+        "threshold_beta": 0.999,
+        "threshold_start_step": 1000,
+        "top_k_aux": 2048,
+        "seed": 0,
+        "activation_dim": 4096,
+        "dict_size": 131072,
+        "k": 256,
+        "device": "cuda",
+        "layer": 11,
+        "lm_name": "meta-llama/Llama-3.1-8B-Instruct",
+        "wandb_name": "3_l11-BatchTopKTrainer-meta-llama/Llama-3.1-8B-Instruct-resid_post_layer_11_trainer_3",
+        "submodule_name": "resid_post_layer_11"
+    },
+    "buffer": {
+        "d_submodule": 4096,
+        "io": "out",
+        "n_ctxs": 244,
+        "ctx_len": 1024,
+        "refresh_batch_size": 16,
+        "out_batch_size": 2048,
+        "device": "cuda",
+        "internal_device": "cuda"
+    }
+}

resid_post_layer_11/trainer_3/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 2.811328125, "l1_loss": 50.305, "l0": 232.714482421875, "frac_variance_explained": 0.8197265625, "cossim": 0.92201171875, "l2_ratio": 0.9220703125, "relative_reconstruction_bias": 1.00140625, "loss_original": 2.1376953125, "loss_reconstructed": 2.15634765625, "loss_zero": 10.1215625, "frac_recovered": 0.99814453125, "frac_alive": 0.8564682006835938, "hyperparameters": {"n_inputs": 200, "context_length": 1024}}

resid_post_layer_15/trainer_0/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b2460809919e23d636bd5db32621689c2fbd2b0c225e51e6db08bd3d6d220c0d
+size 4295510533

resid_post_layer_15/trainer_0/config.json

@@ -0,0 +1,33 @@
+{
+    "trainer": {
+        "trainer_class": "BatchTopKTrainer",
+        "dict_class": "BatchTopKSAE",
+        "lr": 0.0001,
+        "steps": 244140,
+        "auxk_alpha": 0.03125,
+        "warmup_steps": 1000,
+        "decay_start": 195312,
+        "threshold_beta": 0.999,
+        "threshold_start_step": 1000,
+        "top_k_aux": 2048,
+        "seed": 0,
+        "activation_dim": 4096,
+        "dict_size": 131072,
+        "k": 32,
+        "device": "cuda",
+        "layer": 15,
+        "lm_name": "meta-llama/Llama-3.1-8B-Instruct",
+        "wandb_name": "3_l15-BatchTopKTrainer-meta-llama/Llama-3.1-8B-Instruct-resid_post_layer_15_trainer_0",
+        "submodule_name": "resid_post_layer_15"
+    },
+    "buffer": {
+        "d_submodule": 4096,
+        "io": "out",
+        "n_ctxs": 244,
+        "ctx_len": 1024,
+        "refresh_batch_size": 16,
+        "out_batch_size": 2048,
+        "device": "cuda",
+        "internal_device": "cuda"
+    }
+}

resid_post_layer_15/trainer_0/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 4.9178125, "l1_loss": 26.64375, "l0": 29.9983544921875, "frac_variance_explained": 0.71818359375, "cossim": 0.87537109375, "l2_ratio": 0.8741015625, "relative_reconstruction_bias": 1.00083984375, "loss_original": 2.1476953125, "loss_reconstructed": 2.222109375, "loss_zero": 11.509375, "frac_recovered": 0.99255859375, "frac_alive": 0.611968994140625, "hyperparameters": {"n_inputs": 200, "context_length": 1024}}

resid_post_layer_15/trainer_1/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6fb7704aa63e1a04a6bf94803f6b0abb6e21886bc305d945ea31ad12c368fdbe
+size 4295510533

resid_post_layer_15/trainer_1/config.json

@@ -0,0 +1,33 @@
+{
+    "trainer": {
+        "trainer_class": "BatchTopKTrainer",
+        "dict_class": "BatchTopKSAE",
+        "lr": 0.0001,
+        "steps": 244140,
+        "auxk_alpha": 0.03125,
+        "warmup_steps": 1000,
+        "decay_start": 195312,
+        "threshold_beta": 0.999,
+        "threshold_start_step": 1000,
+        "top_k_aux": 2048,
+        "seed": 0,
+        "activation_dim": 4096,
+        "dict_size": 131072,
+        "k": 64,
+        "device": "cuda",
+        "layer": 15,
+        "lm_name": "meta-llama/Llama-3.1-8B-Instruct",
+        "wandb_name": "3_l15-BatchTopKTrainer-meta-llama/Llama-3.1-8B-Instruct-resid_post_layer_15_trainer_1",
+        "submodule_name": "resid_post_layer_15"
+    },
+    "buffer": {
+        "d_submodule": 4096,
+        "io": "out",
+        "n_ctxs": 244,
+        "ctx_len": 1024,
+        "refresh_batch_size": 16,
+        "out_batch_size": 2048,
+        "device": "cuda",
+        "internal_device": "cuda"
+    }
+}

resid_post_layer_15/trainer_1/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 4.42296875, "l1_loss": 36.52375, "l0": 59.665791015625, "frac_variance_explained": 0.77166015625, "cossim": 0.9003515625, "l2_ratio": 0.89958984375, "relative_reconstruction_bias": 1.0005078125, "loss_original": 2.15236328125, "loss_reconstructed": 2.19091796875, "loss_zero": 11.5075, "frac_recovered": 0.99630859375, "frac_alive": 0.7205657958984375, "hyperparameters": {"n_inputs": 200, "context_length": 1024}}

resid_post_layer_15/trainer_2/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:17ef9cd6f293dbcb4670f80dace5223de385cfe9af0cb5147887ab79d277dfbb
+size 4295510533

resid_post_layer_15/trainer_2/config.json

@@ -0,0 +1,33 @@
+{
+    "trainer": {
+        "trainer_class": "BatchTopKTrainer",
+        "dict_class": "BatchTopKSAE",
+        "lr": 0.0001,
+        "steps": 244140,
+        "auxk_alpha": 0.03125,
+        "warmup_steps": 1000,
+        "decay_start": 195312,
+        "threshold_beta": 0.999,
+        "threshold_start_step": 1000,
+        "top_k_aux": 2048,
+        "seed": 0,
+        "activation_dim": 4096,
+        "dict_size": 131072,
+        "k": 128,
+        "device": "cuda",
+        "layer": 15,
+        "lm_name": "meta-llama/Llama-3.1-8B-Instruct",
+        "wandb_name": "3_l15-BatchTopKTrainer-meta-llama/Llama-3.1-8B-Instruct-resid_post_layer_15_trainer_2",
+        "submodule_name": "resid_post_layer_15"
+    },
+    "buffer": {
+        "d_submodule": 4096,
+        "io": "out",
+        "n_ctxs": 244,
+        "ctx_len": 1024,
+        "refresh_batch_size": 16,
+        "out_batch_size": 2048,
+        "device": "cuda",
+        "internal_device": "cuda"
+    }
+}

resid_post_layer_15/trainer_2/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 3.927109375, "l1_loss": 51.16125, "l0": 119.6800048828125, "frac_variance_explained": 0.8193359375, "cossim": 0.9218359375, "l2_ratio": 0.92201171875, "relative_reconstruction_bias": 1.001640625, "loss_original": 2.15947265625, "loss_reconstructed": 2.1763671875, "loss_zero": 11.5159375, "frac_recovered": 0.99890625, "frac_alive": 0.831817626953125, "hyperparameters": {"n_inputs": 200, "context_length": 1024}}

resid_post_layer_15/trainer_3/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:de1b705be57528950d148942db9c54c027dfdd513948abedafc7500d70270c1d
+size 4295510533

resid_post_layer_15/trainer_3/config.json

@@ -0,0 +1,33 @@
+{
+    "trainer": {
+        "trainer_class": "BatchTopKTrainer",
+        "dict_class": "BatchTopKSAE",
+        "lr": 0.0001,
+        "steps": 244140,
+        "auxk_alpha": 0.03125,
+        "warmup_steps": 1000,
+        "decay_start": 195312,
+        "threshold_beta": 0.999,
+        "threshold_start_step": 1000,
+        "top_k_aux": 2048,
+        "seed": 0,
+        "activation_dim": 4096,
+        "dict_size": 131072,
+        "k": 256,
+        "device": "cuda",
+        "layer": 15,
+        "lm_name": "meta-llama/Llama-3.1-8B-Instruct",
+        "wandb_name": "3_l15-BatchTopKTrainer-meta-llama/Llama-3.1-8B-Instruct-resid_post_layer_15_trainer_3",
+        "submodule_name": "resid_post_layer_15"
+    },
+    "buffer": {
+        "d_submodule": 4096,
+        "io": "out",
+        "n_ctxs": 244,
+        "ctx_len": 1024,
+        "refresh_batch_size": 16,
+        "out_batch_size": 2048,
+        "device": "cuda",
+        "internal_device": "cuda"
+    }
+}

resid_post_layer_15/trainer_3/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 3.508046875, "l1_loss": 86.9775, "l0": 241.1074365234375, "frac_variance_explained": 0.85701171875, "cossim": 0.93802734375, "l2_ratio": 0.9388671875, "relative_reconstruction_bias": 1.0024609375, "loss_original": 2.1376953125, "loss_reconstructed": 2.150859375, "loss_zero": 11.4753125, "frac_recovered": 0.99865234375, "frac_alive": 0.849334716796875, "hyperparameters": {"n_inputs": 200, "context_length": 1024}}

resid_post_layer_19/trainer_0/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c1491078389d074fd67bdaa5f2b98fd4f33154abc7d9be32e4633da109ca8975
+size 4295510533

resid_post_layer_19/trainer_0/config.json

@@ -0,0 +1,33 @@
+{
+    "trainer": {
+        "trainer_class": "BatchTopKTrainer",
+        "dict_class": "BatchTopKSAE",
+        "lr": 0.0001,
+        "steps": 244140,
+        "auxk_alpha": 0.03125,
+        "warmup_steps": 1000,
+        "decay_start": 195312,
+        "threshold_beta": 0.999,
+        "threshold_start_step": 1000,
+        "top_k_aux": 2048,
+        "seed": 0,
+        "activation_dim": 4096,
+        "dict_size": 131072,
+        "k": 32,
+        "device": "cuda",
+        "layer": 19,
+        "lm_name": "meta-llama/Llama-3.1-8B-Instruct",
+        "wandb_name": "3_l19-BatchTopKTrainer-meta-llama/Llama-3.1-8B-Instruct-resid_post_layer_19_trainer_0",
+        "submodule_name": "resid_post_layer_19"
+    },
+    "buffer": {
+        "d_submodule": 4096,
+        "io": "out",
+        "n_ctxs": 244,
+        "ctx_len": 1024,
+        "refresh_batch_size": 16,
+        "out_batch_size": 2048,
+        "device": "cuda",
+        "internal_device": "cuda"
+    }
+}

resid_post_layer_19/trainer_0/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 6.89234375, "l1_loss": 41.00625, "l0": 29.6808642578125, "frac_variance_explained": 0.75359375, "cossim": 0.89318359375, "l2_ratio": 0.89205078125, "relative_reconstruction_bias": 1.0003515625, "loss_original": 2.1476953125, "loss_reconstructed": 2.2572265625, "loss_zero": 11.365625, "frac_recovered": 0.988671875, "frac_alive": 0.629974365234375, "hyperparameters": {"n_inputs": 200, "context_length": 1024}}

resid_post_layer_19/trainer_1/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ba1c33d6be319db1fa2b11e10e75d89f0e63ea0ca303bd87fafa41578599ab01
+size 4295510533

resid_post_layer_19/trainer_1/config.json

@@ -0,0 +1,33 @@
+{
+    "trainer": {
+        "trainer_class": "BatchTopKTrainer",
+        "dict_class": "BatchTopKSAE",
+        "lr": 0.0001,
+        "steps": 244140,
+        "auxk_alpha": 0.03125,
+        "warmup_steps": 1000,
+        "decay_start": 195312,
+        "threshold_beta": 0.999,
+        "threshold_start_step": 1000,
+        "top_k_aux": 2048,
+        "seed": 0,
+        "activation_dim": 4096,
+        "dict_size": 131072,
+        "k": 64,
+        "device": "cuda",
+        "layer": 19,
+        "lm_name": "meta-llama/Llama-3.1-8B-Instruct",
+        "wandb_name": "3_l19-BatchTopKTrainer-meta-llama/Llama-3.1-8B-Instruct-resid_post_layer_19_trainer_1",
+        "submodule_name": "resid_post_layer_19"
+    },
+    "buffer": {
+        "d_submodule": 4096,
+        "io": "out",
+        "n_ctxs": 244,
+        "ctx_len": 1024,
+        "refresh_batch_size": 16,
+        "out_batch_size": 2048,
+        "device": "cuda",
+        "internal_device": "cuda"
+    }
+}

resid_post_layer_19/trainer_1/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 6.15609375, "l1_loss": 55.89125, "l0": 59.8660205078125, "frac_variance_explained": 0.80375, "cossim": 0.915625, "l2_ratio": 0.915703125, "relative_reconstruction_bias": 1.00119140625, "loss_original": 2.15236328125, "loss_reconstructed": 2.2250390625, "loss_zero": 11.3728125, "frac_recovered": 0.99181640625, "frac_alive": 0.74993896484375, "hyperparameters": {"n_inputs": 200, "context_length": 1024}}

resid_post_layer_19/trainer_2/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:de3f94ce3c11c7fbae1b5920a7d2c9c60d215cd09a58d0ab577e4ebf300273d9
+size 4295510533

resid_post_layer_19/trainer_2/config.json

@@ -0,0 +1,33 @@
+{
+    "trainer": {
+        "trainer_class": "BatchTopKTrainer",
+        "dict_class": "BatchTopKSAE",
+        "lr": 0.0001,
+        "steps": 244140,
+        "auxk_alpha": 0.03125,
+        "warmup_steps": 1000,
+        "decay_start": 195312,
+        "threshold_beta": 0.999,
+        "threshold_start_step": 1000,
+        "top_k_aux": 2048,
+        "seed": 0,
+        "activation_dim": 4096,
+        "dict_size": 131072,
+        "k": 128,
+        "device": "cuda",
+        "layer": 19,
+        "lm_name": "meta-llama/Llama-3.1-8B-Instruct",
+        "wandb_name": "3_l19-BatchTopKTrainer-meta-llama/Llama-3.1-8B-Instruct-resid_post_layer_19_trainer_2",
+        "submodule_name": "resid_post_layer_19"
+    },
+    "buffer": {
+        "d_submodule": 4096,
+        "io": "out",
+        "n_ctxs": 244,
+        "ctx_len": 1024,
+        "refresh_batch_size": 16,
+        "out_batch_size": 2048,
+        "device": "cuda",
+        "internal_device": "cuda"
+    }
+}

resid_post_layer_19/trainer_2/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 5.47484375, "l1_loss": 78.9575, "l0": 119.4428759765625, "frac_variance_explained": 0.84373046875, "cossim": 0.93353515625, "l2_ratio": 0.933515625, "relative_reconstruction_bias": 1.0006640625, "loss_original": 2.15947265625, "loss_reconstructed": 2.19603515625, "loss_zero": 11.3696875, "frac_recovered": 0.99634765625, "frac_alive": 0.8534011840820312, "hyperparameters": {"n_inputs": 200, "context_length": 1024}}

resid_post_layer_19/trainer_3/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e466a2ccfb881a4f17c90ddc472c07446d06d8e842549ecab7714e8e6d5e051e
+size 4295510533

resid_post_layer_19/trainer_3/config.json

@@ -0,0 +1,33 @@
+{
+    "trainer": {
+        "trainer_class": "BatchTopKTrainer",
+        "dict_class": "BatchTopKSAE",
+        "lr": 0.0001,
+        "steps": 244140,
+        "auxk_alpha": 0.03125,
+        "warmup_steps": 1000,
+        "decay_start": 195312,
+        "threshold_beta": 0.999,
+        "threshold_start_step": 1000,
+        "top_k_aux": 2048,
+        "seed": 0,
+        "activation_dim": 4096,
+        "dict_size": 131072,
+        "k": 256,
+        "device": "cuda",
+        "layer": 19,
+        "lm_name": "meta-llama/Llama-3.1-8B-Instruct",
+        "wandb_name": "3_l19-BatchTopKTrainer-meta-llama/Llama-3.1-8B-Instruct-resid_post_layer_19_trainer_3",
+        "submodule_name": "resid_post_layer_19"
+    },
+    "buffer": {
+        "d_submodule": 4096,
+        "io": "out",
+        "n_ctxs": 244,
+        "ctx_len": 1024,
+        "refresh_batch_size": 16,
+        "out_batch_size": 2048,
+        "device": "cuda",
+        "internal_device": "cuda"
+    }
+}

resid_post_layer_19/trainer_3/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 4.903125, "l1_loss": 139.09, "l0": 240.38546875, "frac_variance_explained": 0.87529296875, "cossim": 0.9471484375, "l2_ratio": 0.94732421875, "relative_reconstruction_bias": 1.00080078125, "loss_original": 2.1376953125, "loss_reconstructed": 2.15904296875, "loss_zero": 11.34875, "frac_recovered": 0.997734375, "frac_alive": 0.8963851928710938, "hyperparameters": {"n_inputs": 200, "context_length": 1024}}

resid_post_layer_23/trainer_0/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:785beab787da4503256ff21ff05f760d477eb0edabce8a0ab52b600345d3c764
+size 4295510533

resid_post_layer_23/trainer_0/config.json

@@ -0,0 +1,33 @@
+{
+    "trainer": {
+        "trainer_class": "BatchTopKTrainer",
+        "dict_class": "BatchTopKSAE",
+        "lr": 0.0001,
+        "steps": 244140,
+        "auxk_alpha": 0.03125,
+        "warmup_steps": 1000,
+        "decay_start": 195312,
+        "threshold_beta": 0.999,
+        "threshold_start_step": 1000,
+        "top_k_aux": 2048,
+        "seed": 0,
+        "activation_dim": 4096,
+        "dict_size": 131072,
+        "k": 32,
+        "device": "cuda",
+        "layer": 23,
+        "lm_name": "meta-llama/Llama-3.1-8B-Instruct",
+        "wandb_name": "3_l23-BatchTopKTrainer-meta-llama/Llama-3.1-8B-Instruct-resid_post_layer_23_trainer_0",
+        "submodule_name": "resid_post_layer_23"
+    },
+    "buffer": {
+        "d_submodule": 4096,
+        "io": "out",
+        "n_ctxs": 244,
+        "ctx_len": 1024,
+        "refresh_batch_size": 16,
+        "out_batch_size": 2048,
+        "device": "cuda",
+        "internal_device": "cuda"
+    }
+}

resid_post_layer_23/trainer_0/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 10.0478125, "l1_loss": 60.30375, "l0": 29.7135107421875, "frac_variance_explained": 0.75935546875, "cossim": 0.89796875, "l2_ratio": 0.89712890625, "relative_reconstruction_bias": 1.00087890625, "loss_original": 2.1476953125, "loss_reconstructed": 2.2941796875, "loss_zero": 11.52, "frac_recovered": 0.98423828125, "frac_alive": 0.670257568359375, "hyperparameters": {"n_inputs": 200, "context_length": 1024}}

resid_post_layer_23/trainer_1/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:74b61c74e0e4ee33e0bc9c7126fd7c475f8ee08b04f721e844c235d4093c332b
+size 4295510533

resid_post_layer_23/trainer_1/config.json

@@ -0,0 +1,33 @@
+{
+    "trainer": {
+        "trainer_class": "BatchTopKTrainer",
+        "dict_class": "BatchTopKSAE",
+        "lr": 0.0001,
+        "steps": 244140,
+        "auxk_alpha": 0.03125,
+        "warmup_steps": 1000,
+        "decay_start": 195312,
+        "threshold_beta": 0.999,
+        "threshold_start_step": 1000,
+        "top_k_aux": 2048,
+        "seed": 0,
+        "activation_dim": 4096,
+        "dict_size": 131072,
+        "k": 64,
+        "device": "cuda",
+        "layer": 23,
+        "lm_name": "meta-llama/Llama-3.1-8B-Instruct",
+        "wandb_name": "3_l23-BatchTopKTrainer-meta-llama/Llama-3.1-8B-Instruct-resid_post_layer_23_trainer_1",
+        "submodule_name": "resid_post_layer_23"
+    },
+    "buffer": {
+        "d_submodule": 4096,
+        "io": "out",
+        "n_ctxs": 244,
+        "ctx_len": 1024,
+        "refresh_batch_size": 16,
+        "out_batch_size": 2048,
+        "device": "cuda",
+        "internal_device": "cuda"
+    }
+}

resid_post_layer_23/trainer_1/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 9.0259375, "l1_loss": 81.5975, "l0": 59.552333984375, "frac_variance_explained": 0.80583984375, "cossim": 0.91814453125, "l2_ratio": 0.9178515625, "relative_reconstruction_bias": 1.00099609375, "loss_original": 2.15236328125, "loss_reconstructed": 2.2246875, "loss_zero": 11.5165625, "frac_recovered": 0.99216796875, "frac_alive": 0.8050079345703125, "hyperparameters": {"n_inputs": 200, "context_length": 1024}}

resid_post_layer_23/trainer_2/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e4f842e9538c7f5f54a7604fd5ee20848d64380835a5520333a1e45bcff6baf9
+size 4295510533

resid_post_layer_23/trainer_2/config.json

@@ -0,0 +1,33 @@
+{
+    "trainer": {
+        "trainer_class": "BatchTopKTrainer",
+        "dict_class": "BatchTopKSAE",
+        "lr": 0.0001,
+        "steps": 244140,
+        "auxk_alpha": 0.03125,
+        "warmup_steps": 1000,
+        "decay_start": 195312,
+        "threshold_beta": 0.999,
+        "threshold_start_step": 1000,
+        "top_k_aux": 2048,
+        "seed": 0,
+        "activation_dim": 4096,
+        "dict_size": 131072,
+        "k": 128,
+        "device": "cuda",
+        "layer": 23,
+        "lm_name": "meta-llama/Llama-3.1-8B-Instruct",
+        "wandb_name": "3_l23-BatchTopKTrainer-meta-llama/Llama-3.1-8B-Instruct-resid_post_layer_23_trainer_2",
+        "submodule_name": "resid_post_layer_23"
+    },
+    "buffer": {
+        "d_submodule": 4096,
+        "io": "out",
+        "n_ctxs": 244,
+        "ctx_len": 1024,
+        "refresh_batch_size": 16,
+        "out_batch_size": 2048,
+        "device": "cuda",
+        "internal_device": "cuda"
+    }
+}

resid_post_layer_23/trainer_2/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 8.1321875, "l1_loss": 113.7975, "l0": 118.788955078125, "frac_variance_explained": 0.84171875, "cossim": 0.93353515625, "l2_ratio": 0.9331640625, "relative_reconstruction_bias": 1.00060546875, "loss_original": 2.15947265625, "loss_reconstructed": 2.20662109375, "loss_zero": 11.50375, "frac_recovered": 0.99517578125, "frac_alive": 0.8776473999023438, "hyperparameters": {"n_inputs": 200, "context_length": 1024}}

resid_post_layer_23/trainer_3/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:00964cbedeeb0ea91755027825407ab36f8566104a7878f4539196186ad4e5c1
+size 4295510533

resid_post_layer_23/trainer_3/config.json

@@ -0,0 +1,33 @@
+{
+    "trainer": {
+        "trainer_class": "BatchTopKTrainer",
+        "dict_class": "BatchTopKSAE",
+        "lr": 0.0001,
+        "steps": 244140,
+        "auxk_alpha": 0.03125,
+        "warmup_steps": 1000,
+        "decay_start": 195312,
+        "threshold_beta": 0.999,
+        "threshold_start_step": 1000,
+        "top_k_aux": 2048,
+        "seed": 0,
+        "activation_dim": 4096,
+        "dict_size": 131072,
+        "k": 256,
+        "device": "cuda",
+        "layer": 23,
+        "lm_name": "meta-llama/Llama-3.1-8B-Instruct",
+        "wandb_name": "3_l23-BatchTopKTrainer-meta-llama/Llama-3.1-8B-Instruct-resid_post_layer_23_trainer_3",
+        "submodule_name": "resid_post_layer_23"
+    },
+    "buffer": {
+        "d_submodule": 4096,
+        "io": "out",
+        "n_ctxs": 244,
+        "ctx_len": 1024,
+        "refresh_batch_size": 16,
+        "out_batch_size": 2048,
+        "device": "cuda",
+        "internal_device": "cuda"
+    }
+}

resid_post_layer_23/trainer_3/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 7.38578125, "l1_loss": 195.195, "l0": 237.4473291015625, "frac_variance_explained": 0.86919921875, "cossim": 0.9453515625, "l2_ratio": 0.94541015625, "relative_reconstruction_bias": 1.0005859375, "loss_original": 2.1376953125, "loss_reconstructed": 2.16935546875, "loss_zero": 11.450625, "frac_recovered": 0.99662109375, "frac_alive": 0.9188232421875, "hyperparameters": {"n_inputs": 200, "context_length": 1024}}

resid_post_layer_27/trainer_0/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8e1337be7e748dffa3411c8b5396a27d1bcb6cfecc7663996ff16ef3384836ad
+size 4295510533

resid_post_layer_27/trainer_0/config.json

@@ -0,0 +1,33 @@
+{
+    "trainer": {
+        "trainer_class": "BatchTopKTrainer",
+        "dict_class": "BatchTopKSAE",
+        "lr": 0.0001,
+        "steps": 244140,
+        "auxk_alpha": 0.03125,
+        "warmup_steps": 1000,
+        "decay_start": 195312,
+        "threshold_beta": 0.999,
+        "threshold_start_step": 1000,
+        "top_k_aux": 2048,
+        "seed": 0,
+        "activation_dim": 4096,
+        "dict_size": 131072,
+        "k": 32,
+        "device": "cuda",
+        "layer": 27,
+        "lm_name": "meta-llama/Llama-3.1-8B-Instruct",
+        "wandb_name": "3_l27-BatchTopKTrainer-meta-llama/Llama-3.1-8B-Instruct-resid_post_layer_27_trainer_0",
+        "submodule_name": "resid_post_layer_27"
+    },
+    "buffer": {
+        "d_submodule": 4096,
+        "io": "out",
+        "n_ctxs": 244,
+        "ctx_len": 1024,
+        "refresh_batch_size": 16,
+        "out_batch_size": 2048,
+        "device": "cuda",
+        "internal_device": "cuda"
+    }
+}