Safetensors
llama3_SAE
custom_code
SCAR / config.json
RuHae's picture
Update config.json
82aba88 verified
{
"_name_or_path": "AIML-TUDA/SCAR",
"activation": "topk",
"activation_k": 2048,
"architectures": [
"LLama3_SAE"
],
"attention_bias": false,
"attention_dropout": 0.0,
"auto_map": {
"AutoConfig": "AIML-TUDA/SCAR--configuration_llama3_SAE.LLama3_SAE_Config",
"AutoModelForCausalLM": "modeling_llama3_SAE.LLama3_SAE"
},
"base_model_name": "meta-llama/Meta-Llama-3-8B",
"bos_token_id": 128000,
"eos_token_id": 128001,
"head_dim": 128,
"hidden_act": "silu",
"hidden_size": 4096,
"hook_block_num": 25,
"initializer_range": 0.02,
"intermediate_size": 14336,
"max_position_embeddings": 8192,
"mlp_bias": false,
"mod_features": null,
"mod_replacement": null,
"mod_scaling": null,
"mod_threshold": null,
"model_type": "llama3_SAE",
"n_inputs": 4096,
"n_latents": 24576,
"normalize": false,
"num_attention_heads": 32,
"num_hidden_layers": 32,
"num_key_value_heads": 8,
"pretraining_tp": 1,
"rms_norm_eps": 1e-05,
"rope_scaling": null,
"rope_theta": 500000.0,
"site": "mlp",
"tie_word_embeddings": false,
"tied": false,
"torch_dtype": "float32",
"transformers_version": "4.46.2",
"use_cache": true,
"vocab_size": 128256
}