Upload folder using huggingface_hub

._saes_Qwen_Qwen2.5-Coder-32B-Instruct_batch_top_k/resid_post_layer_32/trainer_0/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7910a7080d6bb0f930eca90971cabd0ac7f3c8ff07019841d793be400706aef3
+size 671176726

._saes_Qwen_Qwen2.5-Coder-32B-Instruct_batch_top_k/resid_post_layer_32/trainer_0/config.json

@@ -0,0 +1,32 @@
+{
+    "trainer": {
+        "trainer_class": "BatchTopKTrainer",
+        "dict_class": "BatchTopKSAE",
+        "lr": 0.0003,
+        "steps": 122070,
+        "auxk_alpha": 0.03125,
+        "warmup_steps": 1000,
+        "decay_start": 97656,
+        "threshold_beta": 0.999,
+        "threshold_start_step": 1000,
+        "top_k_aux": 2560,
+        "seed": 0,
+        "activation_dim": 5120,
+        "dict_size": 16384,
+        "k": 80,
+        "device": "cuda:0",
+        "layer": 32,
+        "lm_name": "Qwen/Qwen2.5-Coder-32B-Instruct",
+        "wandb_name": "BatchTopKTrainer-Qwen/Qwen2.5-Coder-32B-Instruct-resid_post_layer_32_trainer_0",
+        "submodule_name": "resid_post_layer_32"
+    },
+    "buffer": {
+        "d_submodule": 5120,
+        "io": "out",
+        "n_ctxs": 122,
+        "ctx_len": 2048,
+        "refresh_batch_size": 4,
+        "out_batch_size": 4096,
+        "device": "cuda:0"
+    }
+}

._saes_Qwen_Qwen2.5-Coder-32B-Instruct_batch_top_k/resid_post_layer_32/trainer_0/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 146.145, "l1_loss": 1198.44, "l0": 78.4512841796875, "frac_variance_explained": 0.91130859375, "cossim": 0.84859375, "l2_ratio": 0.8502734375, "relative_reconstruction_bias": 0.99208984375, "frac_alive": 0.99041748046875, "hyperparameters": {"n_inputs": 200, "context_length": 2048}}

._saes_Qwen_Qwen2.5-Coder-32B-Instruct_batch_top_k/resid_post_layer_32/trainer_1/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5bb4b45b0edf9f2e75af3d3c67aa4dbe4cd4d92c3d47d55e2128ea66ac0fa484
+size 671176726

._saes_Qwen_Qwen2.5-Coder-32B-Instruct_batch_top_k/resid_post_layer_32/trainer_1/config.json

@@ -0,0 +1,32 @@
+{
+    "trainer": {
+        "trainer_class": "BatchTopKTrainer",
+        "dict_class": "BatchTopKSAE",
+        "lr": 0.0003,
+        "steps": 122070,
+        "auxk_alpha": 0.03125,
+        "warmup_steps": 1000,
+        "decay_start": 97656,
+        "threshold_beta": 0.999,
+        "threshold_start_step": 1000,
+        "top_k_aux": 2560,
+        "seed": 0,
+        "activation_dim": 5120,
+        "dict_size": 16384,
+        "k": 160,
+        "device": "cuda:0",
+        "layer": 32,
+        "lm_name": "Qwen/Qwen2.5-Coder-32B-Instruct",
+        "wandb_name": "BatchTopKTrainer-Qwen/Qwen2.5-Coder-32B-Instruct-resid_post_layer_32_trainer_1",
+        "submodule_name": "resid_post_layer_32"
+    },
+    "buffer": {
+        "d_submodule": 5120,
+        "io": "out",
+        "n_ctxs": 122,
+        "ctx_len": 2048,
+        "refresh_batch_size": 4,
+        "out_batch_size": 4096,
+        "device": "cuda:0"
+    }
+}

._saes_Qwen_Qwen2.5-Coder-32B-Instruct_batch_top_k/resid_post_layer_32/trainer_1/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 135.115, "l1_loss": 1809.16, "l0": 160.7628955078125, "frac_variance_explained": 0.932578125, "cossim": 0.871875, "l2_ratio": 0.8730859375, "relative_reconstruction_bias": 0.99796875, "frac_alive": 0.994873046875, "hyperparameters": {"n_inputs": 200, "context_length": 2048}}

._saes_Qwen_Qwen2.5-Coder-32B-Instruct_batch_top_k/resid_post_layer_32/trainer_2/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4cdb20ddaa6baadc0b212b1a2117ed81f605fe4f1e96e2446fdf81808067f286
+size 2684639254

._saes_Qwen_Qwen2.5-Coder-32B-Instruct_batch_top_k/resid_post_layer_32/trainer_2/config.json

@@ -0,0 +1,32 @@
+{
+    "trainer": {
+        "trainer_class": "BatchTopKTrainer",
+        "dict_class": "BatchTopKSAE",
+        "lr": 0.0003,
+        "steps": 122070,
+        "auxk_alpha": 0.03125,
+        "warmup_steps": 1000,
+        "decay_start": 97656,
+        "threshold_beta": 0.999,
+        "threshold_start_step": 1000,
+        "top_k_aux": 2560,
+        "seed": 0,
+        "activation_dim": 5120,
+        "dict_size": 65536,
+        "k": 80,
+        "device": "cuda:0",
+        "layer": 32,
+        "lm_name": "Qwen/Qwen2.5-Coder-32B-Instruct",
+        "wandb_name": "BatchTopKTrainer-Qwen/Qwen2.5-Coder-32B-Instruct-resid_post_layer_32_trainer_2",
+        "submodule_name": "resid_post_layer_32"
+    },
+    "buffer": {
+        "d_submodule": 5120,
+        "io": "out",
+        "n_ctxs": 122,
+        "ctx_len": 2048,
+        "refresh_batch_size": 4,
+        "out_batch_size": 4096,
+        "device": "cuda:0"
+    }
+}

._saes_Qwen_Qwen2.5-Coder-32B-Instruct_batch_top_k/resid_post_layer_32/trainer_2/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 135.585, "l1_loss": 1153.0, "l0": 78.23333984375, "frac_variance_explained": 0.930390625, "cossim": 0.8705078125, "l2_ratio": 0.87232421875, "relative_reconstruction_bias": 0.99759765625, "frac_alive": 0.8715972900390625, "hyperparameters": {"n_inputs": 200, "context_length": 2048}}

._saes_Qwen_Qwen2.5-Coder-32B-Instruct_batch_top_k/resid_post_layer_32/trainer_3/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1c2225e66410c1ae9525ea75dd8cfad07431654f065ea5080315f5169a17748a
+size 2684639254

._saes_Qwen_Qwen2.5-Coder-32B-Instruct_batch_top_k/resid_post_layer_32/trainer_3/config.json

@@ -0,0 +1,32 @@
+{
+    "trainer": {
+        "trainer_class": "BatchTopKTrainer",
+        "dict_class": "BatchTopKSAE",
+        "lr": 0.0003,
+        "steps": 122070,
+        "auxk_alpha": 0.03125,
+        "warmup_steps": 1000,
+        "decay_start": 97656,
+        "threshold_beta": 0.999,
+        "threshold_start_step": 1000,
+        "top_k_aux": 2560,
+        "seed": 0,
+        "activation_dim": 5120,
+        "dict_size": 65536,
+        "k": 160,
+        "device": "cuda:0",
+        "layer": 32,
+        "lm_name": "Qwen/Qwen2.5-Coder-32B-Instruct",
+        "wandb_name": "BatchTopKTrainer-Qwen/Qwen2.5-Coder-32B-Instruct-resid_post_layer_32_trainer_3",
+        "submodule_name": "resid_post_layer_32"
+    },
+    "buffer": {
+        "d_submodule": 5120,
+        "io": "out",
+        "n_ctxs": 122,
+        "ctx_len": 2048,
+        "refresh_batch_size": 4,
+        "out_batch_size": 4096,
+        "device": "cuda:0"
+    }
+}

._saes_Qwen_Qwen2.5-Coder-32B-Instruct_batch_top_k/resid_post_layer_32/trainer_3/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 124.6925, "l1_loss": 1689.84, "l0": 158.69532470703126, "frac_variance_explained": 0.94259765625, "cossim": 0.89154296875, "l2_ratio": 0.892734375, "relative_reconstruction_bias": 0.99142578125, "frac_alive": 0.866241455078125, "hyperparameters": {"n_inputs": 200, "context_length": 2048}}