Datasets:

Changyeli03
/

checkpoints-V

Modalities:

Formats:

Size:

Libraries:

Dataset card Data Studio Files Files and versions

Dataset Viewer

Auto-converted to Parquet Duplicate

Split (1)

train · 1 rows

model_name string	local_model_path string	model_class_name string	hook_name string	hook_eval string	hook_layer int64	hook_head_index null	dataset_path string	dataset_trust_remote_code bool	streaming bool	is_dataset_tokenized bool	context_size int64	use_cached_activations bool	cached_activations_path null	architecture string	d_in int64	d_sae int64	b_dec_init_method string	expansion_factor int64	activation_fn string	activation_fn_kwargs unknown	normalize_sae_decoder bool	noise_scale float64	from_pretrained_path null	apply_b_dec_to_input bool	decoder_orthogonal_init bool	decoder_heuristic_init bool	init_encoder_as_decoder_transpose bool	n_batches_in_buffer int64	training_tokens int64	finetuning_tokens int64	store_batch_size_prompts int64	train_batch_size_tokens int64	normalize_activations string	device string	act_store_device string	seed int64	dtype string	prepend_bos bool	autocast bool	autocast_lm bool	compile_llm bool	llm_compilation_mode null	compile_sae bool	sae_compilation_mode null	adam_beta1 float64	adam_beta2 float64	mse_loss_normalization null	l1_coefficient int64	lp_norm float64	scale_sparsity_penalty_by_decoder_norm bool	l1_warm_up_steps int64	lr float64	lr_scheduler_name string	lr_warm_up_steps int64	lr_end float64	lr_decay_steps int64	n_restart_cycles int64	finetuning_method null	use_ghost_grads bool	feature_sampling_window int64	dead_feature_window int64	dead_feature_threshold float64	n_eval_batches int64	eval_batch_size_prompts null	log_to_wandb bool	log_activations_store_to_wandb bool	log_optimizer_state_to_wandb bool	wandb_project string	wandb_id null	run_name string	wandb_entity null	wandb_log_frequency int64	eval_every_n_wandb_logs int64	resume bool	n_checkpoints int64	checkpoint_path string	verbose bool	model_kwargs unknown	model_from_pretrained_kwargs dict	sae_lens_version string	sae_lens_training_version string	vision bool	tokens_per_buffer int64
llava-hf/llava-v1.6-mistral-7b-hf	/home/saev/changye/model/llava	HookedLlava	blocks.16.hook_resid_post	NOT_IN_USE	16	null	/home/saev/changye/data/obelics100k-tokenized-llava4096_4image	true	true	true	4,096	false	null	standard	4,096	65,536	zeros	16	relu	{}	false	0	null	false	false	true	true	32	122,880,000	0	1	4,096	expected_average_only_in	cuda:3	cuda:3	42	float32	true	false	false	false	null	false	null	0.9	0.999	null	5	1	true	1,500	0.00005	constant	0	0.000005	6,000	1	null	false	1,000	1,000	0.0001	10	null	true	false	false	interp-V	null	65536-L1-5-LR-5e-05-Tokens-1.229e+08	null	30	20	false	20	checkpoints-V/kxpk98cr	true	{}	{ "n_devices": 3 }	3.20.0	3.20.0	false	536,870,912

No dataset card yet

Downloads last month: 8

Number of rows:

Total file size: