Upload config.json with huggingface_hub
Browse files- config.json +65 -0
config.json
ADDED
@@ -0,0 +1,65 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"_name_or_path": "/wangbenyou/xidong/VisionJamba/ckpts/visionjambaSFT40PoolConcat2D",
|
3 |
+
"architectures": [
|
4 |
+
"LlavaJambaForCausalLM"
|
5 |
+
],
|
6 |
+
"attention_dropout": 0.0,
|
7 |
+
"attn_layer_offset": 4,
|
8 |
+
"attn_layer_period": 8,
|
9 |
+
"auto_map": {
|
10 |
+
"AutoConfig": "configuration_jamba.JambaConfig",
|
11 |
+
"AutoModel": "modeling_jamba.JambaModel",
|
12 |
+
"AutoModelForCausalLM": "modeling_jamba.JambaForCausalLM",
|
13 |
+
"AutoModelForSequenceClassification": "model.JambaForSequenceClassification"
|
14 |
+
},
|
15 |
+
"bos_token_id": 1,
|
16 |
+
"calc_logits_for_entire_prompt": false,
|
17 |
+
"eos_token_id": 2,
|
18 |
+
"expert_layer_offset": 1,
|
19 |
+
"expert_layer_period": 2,
|
20 |
+
"freeze_mm_mlp_adapter": false,
|
21 |
+
"hidden_act": "silu",
|
22 |
+
"hidden_size": 4096,
|
23 |
+
"image_aspect_ratio": "pad",
|
24 |
+
"initializer_range": 0.02,
|
25 |
+
"intermediate_size": 14336,
|
26 |
+
"mamba_conv_bias": true,
|
27 |
+
"mamba_d_conv": 4,
|
28 |
+
"mamba_d_state": 16,
|
29 |
+
"mamba_dt_rank": 256,
|
30 |
+
"mamba_expand": 2,
|
31 |
+
"mamba_inner_layernorms": true,
|
32 |
+
"mamba_proj_bias": false,
|
33 |
+
"mm_hidden_size": 1024,
|
34 |
+
"mm_patch_merge_type": "flat",
|
35 |
+
"mm_projector_lr": null,
|
36 |
+
"mm_projector_type": "mlp2x_gelu",
|
37 |
+
"mm_use_im_patch_token": false,
|
38 |
+
"mm_use_im_start_end": false,
|
39 |
+
"mm_vision_select_feature": "patch",
|
40 |
+
"mm_vision_select_layer": -2,
|
41 |
+
"mm_vision_tower": "/wangbenyou/guimingchen/models/clip_vit_large_patch14_336",
|
42 |
+
"model_type": "llava_jamba",
|
43 |
+
"n_ctx": 262144,
|
44 |
+
"num_attention_heads": 32,
|
45 |
+
"num_experts": 16,
|
46 |
+
"num_experts_per_tok": 2,
|
47 |
+
"num_hidden_layers": 32,
|
48 |
+
"num_key_value_heads": 8,
|
49 |
+
"output_router_logits": false,
|
50 |
+
"pad_token_id": 0,
|
51 |
+
"resamplePooling": "2d",
|
52 |
+
"rms_norm_eps": 1e-06,
|
53 |
+
"router_aux_loss_coef": 0.001,
|
54 |
+
"sliding_window": null,
|
55 |
+
"tie_word_embeddings": false,
|
56 |
+
"tokenizer_model_max_length": 40960,
|
57 |
+
"tokenizer_padding_side": "right",
|
58 |
+
"torch_dtype": "bfloat16",
|
59 |
+
"transformers_version": "4.42.3",
|
60 |
+
"tune_mm_mlp_adapter": false,
|
61 |
+
"use_cache": true,
|
62 |
+
"use_mamba_kernels": true,
|
63 |
+
"use_mm_proj": true,
|
64 |
+
"vocab_size": 65536
|
65 |
+
}
|