Upload config.json
Browse files- config.json +32 -0
config.json
ADDED
@@ -0,0 +1,32 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"architectures": [
|
3 |
+
"TokenFormerForCausalLM"
|
4 |
+
],
|
5 |
+
"num_layers": 40,
|
6 |
+
"hidden_size": 1536,
|
7 |
+
"num_attention_heads": 16,
|
8 |
+
"qkv_slot_num": 1536,
|
9 |
+
"proj_slot_num": 1536,
|
10 |
+
"ffn_slot_num": 6144,
|
11 |
+
"seq_length": 2048,
|
12 |
+
"max_position_embeddings": 2048,
|
13 |
+
"pos_emb": "rotary",
|
14 |
+
"rotary_pct": 0.25,
|
15 |
+
"no_weight_tying": false,
|
16 |
+
"norm": "layernorm_nonparam",
|
17 |
+
"final_norm": "layernorm",
|
18 |
+
"gpt_j_residual": false,
|
19 |
+
"output_layer_parallelism": "column",
|
20 |
+
"use_bias_in_attn_linear": false,
|
21 |
+
"attention_config": [[["tokenformer"], 40]],
|
22 |
+
"norm_activation_type": "l2_norm_gelu",
|
23 |
+
"scaled_upper_triang_masked_softmax_fusion": false,
|
24 |
+
"bias_gelu_fusion": false,
|
25 |
+
"rope_fusion": false,
|
26 |
+
"layernorm_fusion": false,
|
27 |
+
"init_method": "normal",
|
28 |
+
"output_layer_init_method": "wang_init",
|
29 |
+
"use_cache": true,
|
30 |
+
"torch_dtype": "float16",
|
31 |
+
"transformers_version": "4.36.0"
|
32 |
+
}
|