Upload folder using huggingface_hub

Llama3_1Base-L23TC-8x/checkpoints/final.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ecb673e95a277ac2b3cc2b1e5db3deee8929219cb36e13c7c4ee944c52b9a55e
+size 536945024

Llama3_1Base-L23TC-8x/hyperparams.json

@@ -0,0 +1,35 @@
+{
+    "device": "cuda:0",
+    "seed": 42,
+    "dtype": "torch.bfloat16",
+    "hook_point_in": "blocks.23.ln2.hook_normalized",
+    "hook_point_out": "blocks.23.hook_mlp_out",
+    "use_decoder_bias": true,
+    "apply_decoder_bias_to_pre_encoder": false,
+    "expansion_factor": 8,
+    "d_model": 4096,
+    "d_sae": 32768,
+    "bias_init_method": "all_zero",
+    "act_fn": "jumprelu",
+    "jump_relu_threshold": 0.177734375,
+    "norm_activation": "dataset-wise",
+    "dataset_average_activation_norm": {
+        "in": 64.0,
+        "out": 7.40625
+    },
+    "decoder_exactly_fixed_norm": false,
+    "sparsity_include_decoder_norm": true,
+    "use_glu_encoder": false,
+    "init_decoder_norm": 0.5,
+    "init_encoder_norm": 0.5,
+    "init_encoder_with_decoder_transpose": false,
+    "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-L23TC-8x/lm_config.json

@@ -0,0 +1,10 @@
+{
+    "device": "cuda",
+    "seed": 42,
+    "dtype": "torch.bfloat16",
+    "model_name": "meta-llama/Meta-Llama-3.1-8B",
+    "use_flash_attn": false,
+    "cache_dir": null,
+    "d_model": 4096,
+    "local_files_only": false
+}