| {"sae": {"expansion_factor": 32, "normalize_decoder": true, "num_latents": 131072, "k": 32, "multi_topk": false, "skip_connection": true}, "batch_size": 4, "grad_acc_steps": 2, "micro_acc_steps": 1, "lr": null, "lr_warmup_steps": 1000, "auxk_alpha": 0.0, "dead_feature_threshold": 10000000, "hookpoints": ["layers.0.mlp", "layers.1.mlp", "layers.2.mlp", "layers.3.mlp", "layers.4.mlp", "layers.5.mlp", "layers.6.mlp", "layers.7.mlp", "layers.8.mlp", "layers.9.mlp", "layers.10.mlp", "layers.11.mlp", "layers.12.mlp", "layers.13.mlp", "layers.14.mlp", "layers.15.mlp"], "init_seeds": [0], "layers": [], "layer_stride": 1, "transcode": true, "distribute_modules": true, "save_every": 1000, "log_to_wandb": true, "run_name": "llama-1b-transcode", "wandb_log_frequency": 1, "model": "meta-llama/Llama-3.2-1B", "dataset": "EleutherAI/rpj-v2-sample", "split": "train", "ctx_len": 2048, "revision": null, "load_in_8bit": false, "max_examples": null, "resume": false, "text_column": "raw_content", "finetune": null, "shuffle_seed": 42, "data_preprocessing_num_proc": 112} | |