Upload folder using huggingface_hub

Llama3_1Base-L23A-8x/checkpoints/final.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0d9de3f26b5cff70b025572a0024295d680b386dcc6ae63432482d593428843c
-size 537076272
+oid sha256:22bd81808401289d7165455edb84077ab4385da096b4305dedc95c046cf4751a
+size 536945024

Llama3_1Base-L23A-8x/hyperparams.json

@@ -1,4 +1,7 @@
 {
+    "device": "cuda:0",
+    "seed": 42,
+    "dtype": "torch.bfloat16",
     "hook_point_in": "blocks.23.hook_attn_out",
     "hook_point_out": "blocks.23.hook_attn_out",
     "use_decoder_bias": true,
@@ -6,17 +9,26 @@
     "expansion_factor": 8,
     "d_model": 4096,
     "d_sae": 32768,
-    "norm_activation": "token-wise",
-    "dataset_average_activation_norm": null,
+    "bias_init_method": "all_zero",
+    "act_fn": "jumprelu",
+    "jump_relu_threshold": 0.119140625,
+    "norm_activation": "dataset-wise",
+    "dataset_average_activation_norm": {
+        "in": 2.890625,
+        "out": 2.890625
+    },
     "decoder_exactly_fixed_norm": false,
     "sparsity_include_decoder_norm": true,
     "use_glu_encoder": false,
-    "init_decoder_norm": null,
+    "init_decoder_norm": 0.5,
     "init_encoder_norm": null,
     "init_encoder_with_decoder_transpose": true,
-    "l1_coefficient": 1.6e-05,
-    "l1_coefficient_warmup_steps": 14648,
     "lp": 1,
+    "l1_coefficient": 8e-05,
+    "l1_coefficient_warmup_steps": 39062,
+    "top_k": 50,
+    "k_warmup_steps": 39062,
+    "use_batch_norm_mse": true,
     "use_ghost_grads": false,
     "tp_size": 1,
     "ddp_size": 1

Llama3_1Base-L23A-8x/lm_config.json

@@ -1,4 +1,7 @@
 {
+    "device": "cuda",
+    "seed": 42,
+    "dtype": "torch.bfloat16",
     "model_name": "meta-llama/Meta-Llama-3.1-8B",
     "use_flash_attn": false,
     "cache_dir": null,