Upload folder using huggingface_hub
Browse files
FAST/blocks_4_hook_resid_post_16X_2048_standard/cfg.json
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
{"model_name": "Qwen/Qwen2.5-3B-Instruct", "model_class_name": "HookedTransformer", "hook_name": "blocks.4.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 4, "hook_head_index": null, "dataset_path": "", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 2048, "use_cached_activations": false, "cached_activations_path": null, "architecture": "standard", "d_in": 2048, "d_sae": 32768, "b_dec_init_method": "geometric_median", "expansion_factor": 16, "activation_fn": "relu", "activation_fn_kwargs": {}, "normalize_sae_decoder": true, "noise_scale": 0.0, "from_pretrained_path": null, "apply_b_dec_to_input": false, "decoder_orthogonal_init": false, "decoder_heuristic_init": false, "init_encoder_as_decoder_transpose": false, "n_batches_in_buffer": 512, "training_tokens": 40960000, "finetuning_tokens": 0, "store_batch_size_prompts": 1, "normalize_activations": "none", "seqpos_slice": [null], "device": "cuda:0", "act_store_device": "cpu", "seed": 42, "dtype": "float32", "prepend_bos": false, "jumprelu_init_threshold": 0.001, "jumprelu_bandwidth": 0.001, "autocast": false, "autocast_lm": false, "compile_llm": false, "llm_compilation_mode": null, "compile_sae": false, "sae_compilation_mode": null, "train_batch_size_tokens": 128, "adam_beta1": 0.9, "adam_beta2": 0.999, "mse_loss_normalization": null, "l1_coefficient": 5, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 10000, "lr": 7e-05, "lr_scheduler_name": "cosineannealing", "lr_warm_up_steps": 16000, "lr_end": 6.999999999999999e-06, "lr_decay_steps": 64000, "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": 1, "log_to_wandb": true, "log_activations_store_to_wandb": false, "log_optimizer_state_to_wandb": true, "wandb_project": "Qwen2.5-3B-Instruct_blocks_4_hook_resid_post_16X_2048_standard_20000_FST", "wandb_id": null, "run_name": "32768-L1-5-LR-7e-05-Tokens-4.096e+07", "wandb_entity": null, "wandb_log_frequency": 10, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 1, "checkpoint_path": "../outputs/sae_FST/Qwen2.5-3B-Instruct/blocks_4_hook_resid_post_16X_2048_standard/1zn7ct2q", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {"center_writing_weights": false}, "sae_lens_version": "5.4.2", "sae_lens_training_version": "5.4.2", "exclude_special_tokens": false, "tokens_per_buffer": 134217728}
|
FAST/blocks_4_hook_resid_post_16X_2048_standard/sae_weights.safetensors
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:749c270593b0237afc47ad29866c2b40501d4f44636b26fdb2d649c50a0ae79e
|
3 |
+
size 537010496
|
FAST/blocks_4_hook_resid_post_16X_2048_standard/sparsity.safetensors
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:175ea40fe903a5a570e2007e22955507e097fad7dea67cd8d7fb6be262738a93
|
3 |
+
size 131152
|