Tanuki-8x8B-vision-exp / config.json
kanhatakeyama's picture
Duplicate from team-hatakeyama-phase2/Tanuki-8x8B-vision-v0-stage2-ckpt900
a6dd058 verified
raw
history blame
1.42 kB
{
"_name_or_path": "/storage5/shared/Nishijima/Llama-3-8b-MoE/5th_tonyu_iter_20000",
"architectures": [
"LlavaTanukiForCausalLM"
],
"attention_bias": false,
"attention_dropout": 0.0,
"auto_map": {
"AutoConfig": "configuration_tanuki.TanukiConfig",
"AutoModelForCausalLM": "modeling_tanuki.TanukiForCausalLM"
},
"bos_token_id": 1,
"eos_token_id": 2,
"hidden_act": "silu",
"hidden_size": 4096,
"image_aspect_ratio": "square",
"initializer_range": 0.02,
"intermediate_size": 14336,
"max_position_embeddings": 2048,
"mm_hidden_size": 2304,
"mm_projector_type": "mlp2x_gelu",
"mm_vision_select_feature": "patch",
"mm_vision_select_layer": -2,
"mm_vision_tower": "google/siglip-so400m-patch14-384",
"model_type": "llava-jp",
"num_attention_heads": 32,
"num_experts_per_tok": 2,
"num_experts_per_topk": 2,
"num_hidden_layers": 32,
"num_key_value_heads": 8,
"num_local_experts": 8,
"output_router_logits": false,
"rms_norm_eps": 1e-05,
"rope_theta": 10000,
"router_aux_loss_coef": 0.001,
"router_jitter_noise": 0.0,
"scales": [
1.0,
0.5
],
"sliding_window": null,
"tie_word_embeddings": false,
"tokenizer_model_max_length": 8192,
"tokenizer_padding_side": "right",
"torch_dtype": "bfloat16",
"transformers_version": "4.43.3",
"tune_mm_mlp_adapter": false,
"use_cache": false,
"use_mm_proj": true,
"vocab_size": 65024
}