Upload SAE

saes/k-8_l-2/cfg.json

@@ -0,0 +1 @@
+{"model_name": "drc33", "model_class_name": "HookedTransformer", "hook_name": "features_extractor.cell_list.2.hook_h", "hook_eval": "NOT_IN_USE", "hook_layer": 2, "hook_head_index": null, "dataset_path": null, "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 128, "use_cached_activations": true, "cached_activations_path": ["/training/activations_dataset/train_unfiltered/0_think_step/", "/training/activations_dataset/train_medium/0_think_step/", "/training/activations_dataset/hard/0_think_step/"], "architecture": "standard", "d_in": 32, "d_sae": 512, "b_dec_init_method": "geometric_median", "expansion_factor": 16, "activation_fn": "topk", "activation_fn_kwargs": {"k": 8}, "normalize_sae_decoder": true, "noise_scale": 0.0, "from_pretrained_path": null, "apply_b_dec_to_input": true, "decoder_orthogonal_init": false, "decoder_heuristic_init": false, "init_encoder_as_decoder_transpose": true, "n_batches_in_buffer": 20, "training_tokens": 600000000, "finetuning_tokens": 0, "store_batch_size_prompts": 32, "train_batch_size_tokens": 4096, "normalize_activations": "layer_norm", "device": "cuda", "act_store_device": "cuda", "seed": 968766353, "dtype": "float32", "prepend_bos": true, "autocast": false, "autocast_lm": false, "compile_llm": false, "llm_compilation_mode": null, "compile_sae": false, "sae_compilation_mode": null, "adam_beta1": 0.9, "adam_beta2": 0.999, "mse_loss_normalization": null, "l1_coefficient": 1e-10, "lp_norm": 1.0, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 10000, "lr": 5e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 5e-06, "lr_decay_steps": 40000, "n_restart_cycles": 1, "finetuning_method": null, "use_ghost_grads": false, "feature_sampling_window": 2000, "dead_feature_window": 1000, "dead_feature_threshold": 1e-08, "n_eval_batches": 10, "eval_batch_size_prompts": null, "log_to_wandb": true, "log_activations_store_to_wandb": false, "log_optimizer_state_to_wandb": false, "wandb_project": "lp_sae", "wandb_id": null, "run_name": "512-L1-1e-10-LR-5e-05-Tokens-6.000e+08-Epochs-1-GridWise", "wandb_entity": null, "wandb_log_frequency": 100, "eval_every_n_wandb_logs": 500, "resume": false, "n_checkpoints": 5, "checkpoint_path": "/training/TrainSAEConfig/03-train-sae-k1/wandb/run-20240909_181726-ho6ob1tk/local-files/checkpoint", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.18.2", "sae_lens_training_version": "3.18.2", "grid_wise": true, "epochs": 1, "num_envs": 64, "envpool": true, "tokens_per_buffer": 10485760}

saes/k-8_l-2/sae_weights.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:78be1c6326af2a6a67e6d453780bd61b925435417e1378b401fcd1ed23c7c50d
+size 133536

saes/k-8_l-2/sparsity.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2ce730a6db3b587fdaa31e1bed58ba949a2cfef48f15b15bc133907ae097563d
+size 2128