Upload folder using huggingface_hub

._saes_Qwen_Qwen2.5-Coder-32B-Instruct_batch_top_k/resid_post_layer_50/trainer_0/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:651bc22061b540e3fbe97f330460663104ff69c61d5600c6664fd5b0a7798a80
+size 671176726

._saes_Qwen_Qwen2.5-Coder-32B-Instruct_batch_top_k/resid_post_layer_50/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": 50,
+        "lm_name": "Qwen/Qwen2.5-Coder-32B-Instruct",
+        "wandb_name": "BatchTopKTrainer-Qwen/Qwen2.5-Coder-32B-Instruct-resid_post_layer_50_trainer_0",
+        "submodule_name": "resid_post_layer_50"
+    },
+    "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_50/trainer_0/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 232.7, "l1_loss": 2156.88, "l0": 78.24118896484374, "frac_variance_explained": 0.88583984375, "cossim": 0.871328125, "l2_ratio": 0.8742578125, "relative_reconstruction_bias": 0.9945703125, "frac_alive": 0.94476318359375, "hyperparameters": {"n_inputs": 200, "context_length": 2048}}

._saes_Qwen_Qwen2.5-Coder-32B-Instruct_batch_top_k/resid_post_layer_50/trainer_1/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5d169b1d707ff009a1dca853096135308df0edf299ceafa16147d9c98a2e64bf
+size 671176726

._saes_Qwen_Qwen2.5-Coder-32B-Instruct_batch_top_k/resid_post_layer_50/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": 50,
+        "lm_name": "Qwen/Qwen2.5-Coder-32B-Instruct",
+        "wandb_name": "BatchTopKTrainer-Qwen/Qwen2.5-Coder-32B-Instruct-resid_post_layer_50_trainer_1",
+        "submodule_name": "resid_post_layer_50"
+    },
+    "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_50/trainer_1/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 212.12, "l1_loss": 3819.52, "l0": 161.6133349609375, "frac_variance_explained": 0.9108984375, "cossim": 0.894375, "l2_ratio": 0.896171875, "relative_reconstruction_bias": 0.991875, "frac_alive": 0.9765625, "hyperparameters": {"n_inputs": 200, "context_length": 2048}}

._saes_Qwen_Qwen2.5-Coder-32B-Instruct_batch_top_k/resid_post_layer_50/trainer_2/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c666e63996352b26b4064ae89b9e48fed55001e63af13094ee2b12c8940b5eb4
+size 2684639254

._saes_Qwen_Qwen2.5-Coder-32B-Instruct_batch_top_k/resid_post_layer_50/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": 50,
+        "lm_name": "Qwen/Qwen2.5-Coder-32B-Instruct",
+        "wandb_name": "BatchTopKTrainer-Qwen/Qwen2.5-Coder-32B-Instruct-resid_post_layer_50_trainer_2",
+        "submodule_name": "resid_post_layer_50"
+    },
+    "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_50/trainer_2/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 219.34, "l1_loss": 2060.84, "l0": 78.80491455078125, "frac_variance_explained": 0.90408203125, "cossim": 0.88654296875, "l2_ratio": 0.8886328125, "relative_reconstruction_bias": 0.99619140625, "frac_alive": 0.7082977294921875, "hyperparameters": {"n_inputs": 200, "context_length": 2048}}

._saes_Qwen_Qwen2.5-Coder-32B-Instruct_batch_top_k/resid_post_layer_50/trainer_3/ae.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5abdeebc0775abe86d2522fcf15cc533b44ee960f0c6f7ae367cdbcd19bc10a2
+size 2684639254

._saes_Qwen_Qwen2.5-Coder-32B-Instruct_batch_top_k/resid_post_layer_50/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": 50,
+        "lm_name": "Qwen/Qwen2.5-Coder-32B-Instruct",
+        "wandb_name": "BatchTopKTrainer-Qwen/Qwen2.5-Coder-32B-Instruct-resid_post_layer_50_trainer_3",
+        "submodule_name": "resid_post_layer_50"
+    },
+    "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_50/trainer_3/eval_results.json

@@ -0,0 +1 @@
+{"l2_loss": 198.33, "l1_loss": 3952.96, "l0": 161.882958984375, "frac_variance_explained": 0.9212109375, "cossim": 0.90740234375, "l2_ratio": 0.90974609375, "relative_reconstruction_bias": 0.995078125, "frac_alive": 0.734619140625, "hyperparameters": {"n_inputs": 200, "context_length": 2048}}