Upload folder using huggingface_hub

EleutherAI/pythia-70m-deduped/top_k_emb_jump_relu_batch_top_k_top_k/resid_post_layer_3/trainer_0/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:77f298c009f298f19efa2c33371ea87aef4d33138166042946b87ef92fef903d
+size 206250518

EleutherAI/pythia-70m-deduped/top_k_emb_jump_relu_batch_top_k_top_k/resid_post_layer_3/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": 512,
+        "dict_size": 50304,
+        "k": 32,
+        "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": 1953,
+        "ctx_len": 1024,
+        "refresh_batch_size": 64,
+        "out_batch_size": 2048,
+        "device": "cuda:0"
+    }
+}

EleutherAI/pythia-70m-deduped/top_k_emb_jump_relu_batch_top_k_top_k/resid_post_layer_3/trainer_0/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 3.6702676713466644, "l1_loss": 40.3086051940918, "l0": 31.997340440750122, "frac_variance_explained": 0.9403489008545876, "cossim": 0.9629507549107075, "l2_ratio": 0.9629187695682049, "relative_reconstruction_bias": 1.0013323947787285, "loss_original": 2.988881543278694, "loss_reconstructed": 3.1971689760684967, "loss_zero": 12.706698060035706, "frac_recovered": 0.9785529784858227, "frac_alive": 0.8209287524223328, "hyperparameters": {"n_inputs": 200, "context_length": 1024}}

EleutherAI/pythia-70m-deduped/top_k_emb_jump_relu_batch_top_k_top_k/resid_post_layer_3/trainer_1/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9e37845b8d6673ef4aab419c3593d8f093c9db41497378211ff8b4bedf75ae60
+size 206250518

EleutherAI/pythia-70m-deduped/top_k_emb_jump_relu_batch_top_k_top_k/resid_post_layer_3/trainer_1/config.json

@@ -0,0 +1,31 @@
+{
+    "trainer": {
+        "trainer_class": "TopKEmbTrainer",
+        "dict_class": "AutoEncoderTopKEmb",
+        "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": 512,
+        "dict_size": 50304,
+        "k": 32,
+        "device": "cuda:0",
+        "layer": 3,
+        "lm_name": "EleutherAI/pythia-70m-deduped",
+        "wandb_name": "TopKEmbTrainer-EleutherAI/pythia-70m-deduped-resid_post_layer_3_trainer_1",
+        "submodule_name": "resid_post_layer_3"
+    },
+    "buffer": {
+        "d_submodule": 512,
+        "io": "out",
+        "n_ctxs": 1953,
+        "ctx_len": 1024,
+        "refresh_batch_size": 64,
+        "out_batch_size": 2048,
+        "device": "cuda:0"
+    }
+}

EleutherAI/pythia-70m-deduped/top_k_emb_jump_relu_batch_top_k_top_k/resid_post_layer_3/trainer_1/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 9.35824728012085, "l1_loss": 40.792572021484375, "l0": 32.0, "frac_variance_explained": 0.4826584868133068, "cossim": 0.7637246735394001, "l2_ratio": 0.7740565240383148, "relative_reconstruction_bias": 0.9987811036407948, "loss_original": 2.988881543278694, "loss_reconstructed": 5.6814234256744385, "loss_zero": 12.706698060035706, "frac_recovered": 0.7233006656169891, "frac_alive": 0.7672352194786072, "hyperparameters": {"n_inputs": 200, "context_length": 1024}}

EleutherAI/pythia-70m-deduped/top_k_emb_jump_relu_batch_top_k_top_k/resid_post_layer_3/trainer_2/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:12b682c1231ea75b949636379fe75c746ee8f5dc081d5a04190a8cc86cde813f
+size 206250518

EleutherAI/pythia-70m-deduped/top_k_emb_jump_relu_batch_top_k_top_k/resid_post_layer_3/trainer_2/config.json

@@ -0,0 +1,32 @@
+{
+    "trainer": {
+        "trainer_class": "BatchTopKTrainer",
+        "dict_class": "BatchTopKSAE",
+        "lr": 0.0003,
+        "steps": 48828,
+        "auxk_alpha": 0.03125,
+        "warmup_steps": 1000,
+        "decay_start": 39062,
+        "threshold_beta": 0.999,
+        "threshold_start_step": 1000,
+        "top_k_aux": 256,
+        "seed": 0,
+        "activation_dim": 512,
+        "dict_size": 50304,
+        "k": 32,
+        "device": "cuda:0",
+        "layer": 3,
+        "lm_name": "EleutherAI/pythia-70m-deduped",
+        "wandb_name": "BatchTopKTrainer-EleutherAI/pythia-70m-deduped-resid_post_layer_3_trainer_2",
+        "submodule_name": "resid_post_layer_3"
+    },
+    "buffer": {
+        "d_submodule": 512,
+        "io": "out",
+        "n_ctxs": 1953,
+        "ctx_len": 1024,
+        "refresh_batch_size": 64,
+        "out_batch_size": 2048,
+        "device": "cuda:0"
+    }
+}

EleutherAI/pythia-70m-deduped/top_k_emb_jump_relu_batch_top_k_top_k/resid_post_layer_3/trainer_2/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 3.8707719296216965, "l1_loss": 42.45883083343506, "l0": 32.06422531604767, "frac_variance_explained": 0.9344661235809326, "cossim": 0.9588854685425758, "l2_ratio": 0.9584895856678486, "relative_reconstruction_bias": 1.0008445084095001, "loss_original": 2.988881543278694, "loss_reconstructed": 3.2311274707317352, "loss_zero": 12.706698060035706, "frac_recovered": 0.9750224277377129, "frac_alive": 0.5356234312057495, "hyperparameters": {"n_inputs": 200, "context_length": 1024}}

EleutherAI/pythia-70m-deduped/top_k_emb_jump_relu_batch_top_k_top_k/resid_post_layer_3/trainer_3/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a077e3395b16e3ec132dc53a51dd8c1614d83e5f22f677974ef4456f47cd413d
+size 206451487

EleutherAI/pythia-70m-deduped/top_k_emb_jump_relu_batch_top_k_top_k/resid_post_layer_3/trainer_3/config.json

@@ -0,0 +1,29 @@
+{
+    "trainer": {
+        "trainer_class": "JumpReluTrainer",
+        "dict_class": "JumpReluAutoEncoder",
+        "lr": 0.0003,
+        "steps": 48828,
+        "seed": 0,
+        "activation_dim": 512,
+        "dict_size": 50304,
+        "device": "cuda:0",
+        "layer": 3,
+        "lm_name": "EleutherAI/pythia-70m-deduped",
+        "wandb_name": "JumpReluTrainer-EleutherAI/pythia-70m-deduped-resid_post_layer_3_trainer_3",
+        "submodule_name": "resid_post_layer_3",
+        "bandwidth": 0.001,
+        "sparsity_penalty": 1.0,
+        "sparsity_warmup_steps": 5000,
+        "target_l0": 32
+    },
+    "buffer": {
+        "d_submodule": 512,
+        "io": "out",
+        "n_ctxs": 1953,
+        "ctx_len": 1024,
+        "refresh_batch_size": 64,
+        "out_batch_size": 2048,
+        "device": "cuda:0"
+    }
+}

EleutherAI/pythia-70m-deduped/top_k_emb_jump_relu_batch_top_k_top_k/resid_post_layer_3/trainer_3/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 3.806805431842804, "l1_loss": 42.213606119155884, "l0": 32.728954553604126, "frac_variance_explained": 0.9359978176653385, "cossim": 0.9602095633745193, "l2_ratio": 0.9597240686416626, "relative_reconstruction_bias": 1.0008267238736153, "loss_original": 2.988881543278694, "loss_reconstructed": 3.2248510867357254, "loss_zero": 12.706698060035706, "frac_recovered": 0.9756911918520927, "frac_alive": 0.5014909505844116, "hyperparameters": {"n_inputs": 200, "context_length": 1024}}