Upload folder using huggingface_hub
Browse files
BT(F)/blocks_18_hook_resid_post_8X_2048_jumprelu/cfg.json
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
{"model_name": "Qwen/Qwen2.5-7B-Instruct", "model_class_name": "HookedTransformer", "hook_name": "blocks.18.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 18, "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": "jumprelu", "d_in": 3584, "d_sae": 28672, "b_dec_init_method": "geometric_median", "expansion_factor": 8, "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, "train_batch_size_tokens": 128, "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, "adam_beta1": 0.9, "adam_beta2": 0.999, "mse_loss_normalization": null, "l1_coefficient": 0.01, "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": null, "log_to_wandb": true, "log_activations_store_to_wandb": false, "log_optimizer_state_to_wandb": true, "wandb_project": "Qwen2.5-7B-Instruct_blocks_18_hook_resid_post_8X_2048_jumprelu_20000_B", "wandb_id": null, "run_name": "28672-L1-0.01-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_B/Qwen2.5-7B-Instruct/blocks_18_hook_resid_post_8X_2048_jumprelu/fxtjbsrl", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {"center_writing_weights": false}, "sae_lens_version": "5.3.3", "sae_lens_training_version": "5.3.3", "tokens_per_buffer": 134217728}
|
BT(F)/blocks_18_hook_resid_post_8X_2048_jumprelu/sae_weights.safetensors
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:6b46aeed4f2de8bdafb67edc3c55174de4ca13ce77391adb0a081095cddc8cca
|
3 |
+
size 822327696
|
BT(F)/blocks_18_hook_resid_post_8X_2048_jumprelu/sparsity.safetensors
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:2257e7ac650c949057cd20ba9ef3f28ae980cfa17da757ad015c1089c420fe8e
|
3 |
+
size 114768
|