Upload folder using huggingface_hub

trained_EleutherAI_pythia-1b_top_k/resid_post_layer_10/trainer_0/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:761c54b9fd02cf6c7b9f4766030024b3fe443801f10f2bd8a7230515d9dffa4d
+size 268511254

trained_EleutherAI_pythia-1b_top_k/resid_post_layer_10/trainer_0/config.json

@@ -0,0 +1,31 @@
+{
+    "trainer": {
+        "trainer_class": "TopKTrainer",
+        "dict_class": "AutoEncoderTopK",
+        "lr": 0.0003,
+        "steps": 48828,
+        "auxk_alpha": 0.03125,
+        "warmup_steps": 1000,
+        "decay_start": 39062,
+        "threshold_beta": 0.999,
+        "threshold_start_step": 1000,
+        "seed": 0,
+        "activation_dim": 2048,
+        "dict_size": 16384,
+        "k": 20,
+        "device": "cuda:0",
+        "layer": 10,
+        "lm_name": "EleutherAI/pythia-1b",
+        "wandb_name": "TopKTrainer-EleutherAI/pythia-1b-resid_post_layer_10_trainer_0",
+        "submodule_name": "resid_post_layer_10"
+    },
+    "buffer": {
+        "d_submodule": 2048,
+        "io": "out",
+        "n_ctxs": 244,
+        "ctx_len": 1024,
+        "refresh_batch_size": 4,
+        "out_batch_size": 2048,
+        "device": "cuda:0"
+    }
+}

trained_EleutherAI_pythia-1b_top_k/resid_post_layer_10/trainer_0/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 23.985, "l1_loss": 114.46, "l0": 19.9999853515625, "frac_variance_explained": 0.95412109375, "cossim": 0.90083984375, "l2_ratio": 0.9014453125, "relative_reconstruction_bias": 0.99986328125, "loss_original": 2.128095703125, "loss_reconstructed": 2.605498046875, "loss_zero": 11.00125, "frac_recovered": 0.9469921875, "frac_alive": 0.92596435546875, "hyperparameters": {"n_inputs": 200, "context_length": 1024}}

trained_EleutherAI_pythia-1b_top_k/resid_post_layer_10/trainer_1/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e2012f50587050666f5354e860d591b43c4308e5ea511f170fd072e09321e4ce
+size 268511254

trained_EleutherAI_pythia-1b_top_k/resid_post_layer_10/trainer_1/config.json

@@ -0,0 +1,31 @@
+{
+    "trainer": {
+        "trainer_class": "TopKTrainer",
+        "dict_class": "AutoEncoderTopK",
+        "lr": 0.0003,
+        "steps": 48828,
+        "auxk_alpha": 0.03125,
+        "warmup_steps": 1000,
+        "decay_start": 39062,
+        "threshold_beta": 0.999,
+        "threshold_start_step": 1000,
+        "seed": 0,
+        "activation_dim": 2048,
+        "dict_size": 16384,
+        "k": 40,
+        "device": "cuda:0",
+        "layer": 10,
+        "lm_name": "EleutherAI/pythia-1b",
+        "wandb_name": "TopKTrainer-EleutherAI/pythia-1b-resid_post_layer_10_trainer_1",
+        "submodule_name": "resid_post_layer_10"
+    },
+    "buffer": {
+        "d_submodule": 2048,
+        "io": "out",
+        "n_ctxs": 244,
+        "ctx_len": 1024,
+        "refresh_batch_size": 4,
+        "out_batch_size": 2048,
+        "device": "cuda:0"
+    }
+}

trained_EleutherAI_pythia-1b_top_k/resid_post_layer_10/trainer_1/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 22.178125, "l1_loss": 152.275, "l0": 39.9995068359375, "frac_variance_explained": 0.95654296875, "cossim": 0.91541015625, "l2_ratio": 0.9160546875, "relative_reconstruction_bias": 1.00005859375, "loss_original": 2.128095703125, "loss_reconstructed": 2.451884765625, "loss_zero": 11.00125, "frac_recovered": 0.964453125, "frac_alive": 0.9920654296875, "hyperparameters": {"n_inputs": 200, "context_length": 1024}}

trained_EleutherAI_pythia-1b_top_k/resid_post_layer_10/trainer_2/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9a81ff8ec9284e32f56bad9f3e7b348885bbec65b77c0792c3ebd6fd2c7bff6a
+size 268511254

trained_EleutherAI_pythia-1b_top_k/resid_post_layer_10/trainer_2/config.json

@@ -0,0 +1,31 @@
+{
+    "trainer": {
+        "trainer_class": "TopKTrainer",
+        "dict_class": "AutoEncoderTopK",
+        "lr": 0.0003,
+        "steps": 48828,
+        "auxk_alpha": 0.03125,
+        "warmup_steps": 1000,
+        "decay_start": 39062,
+        "threshold_beta": 0.999,
+        "threshold_start_step": 1000,
+        "seed": 0,
+        "activation_dim": 2048,
+        "dict_size": 16384,
+        "k": 80,
+        "device": "cuda:0",
+        "layer": 10,
+        "lm_name": "EleutherAI/pythia-1b",
+        "wandb_name": "TopKTrainer-EleutherAI/pythia-1b-resid_post_layer_10_trainer_2",
+        "submodule_name": "resid_post_layer_10"
+    },
+    "buffer": {
+        "d_submodule": 2048,
+        "io": "out",
+        "n_ctxs": 244,
+        "ctx_len": 1024,
+        "refresh_batch_size": 4,
+        "out_batch_size": 2048,
+        "device": "cuda:0"
+    }
+}

trained_EleutherAI_pythia-1b_top_k/resid_post_layer_10/trainer_2/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 20.609375, "l1_loss": 215.105, "l0": 79.9976220703125, "frac_variance_explained": 0.96400390625, "cossim": 0.92716796875, "l2_ratio": 0.92767578125, "relative_reconstruction_bias": 0.99984375, "loss_original": 2.128095703125, "loss_reconstructed": 2.3584765625, "loss_zero": 11.00125, "frac_recovered": 0.9748046875, "frac_alive": 0.998291015625, "hyperparameters": {"n_inputs": 200, "context_length": 1024}}

trained_EleutherAI_pythia-1b_top_k/resid_post_layer_10/trainer_3/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a90091458467036c40075c762f26f3e7e5e7d45fedbdbbece3a3d13aa82d747f
+size 268511254

trained_EleutherAI_pythia-1b_top_k/resid_post_layer_10/trainer_3/config.json

@@ -0,0 +1,31 @@
+{
+    "trainer": {
+        "trainer_class": "TopKTrainer",
+        "dict_class": "AutoEncoderTopK",
+        "lr": 0.0003,
+        "steps": 48828,
+        "auxk_alpha": 0.03125,
+        "warmup_steps": 1000,
+        "decay_start": 39062,
+        "threshold_beta": 0.999,
+        "threshold_start_step": 1000,
+        "seed": 0,
+        "activation_dim": 2048,
+        "dict_size": 16384,
+        "k": 160,
+        "device": "cuda:0",
+        "layer": 10,
+        "lm_name": "EleutherAI/pythia-1b",
+        "wandb_name": "TopKTrainer-EleutherAI/pythia-1b-resid_post_layer_10_trainer_3",
+        "submodule_name": "resid_post_layer_10"
+    },
+    "buffer": {
+        "d_submodule": 2048,
+        "io": "out",
+        "n_ctxs": 244,
+        "ctx_len": 1024,
+        "refresh_batch_size": 4,
+        "out_batch_size": 2048,
+        "device": "cuda:0"
+    }
+}

trained_EleutherAI_pythia-1b_top_k/resid_post_layer_10/trainer_3/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 18.89, "l1_loss": 304.52, "l0": 159.897294921875, "frac_variance_explained": 0.97150390625, "cossim": 0.9391796875, "l2_ratio": 0.94046875, "relative_reconstruction_bias": 0.99935546875, "loss_original": 2.128095703125, "loss_reconstructed": 2.292900390625, "loss_zero": 11.00125, "frac_recovered": 0.98166015625, "frac_alive": 0.9993896484375, "hyperparameters": {"n_inputs": 200, "context_length": 1024}}