config.json
Browse files- config.json +45 -0
config.json
ADDED
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"producer": {
|
3 |
+
"name": "modelopt",
|
4 |
+
"version": "0.11.2"
|
5 |
+
},
|
6 |
+
"architecture": "LlamaForCausalLM",
|
7 |
+
"dtype": "float16",
|
8 |
+
"num_hidden_layers": 32,
|
9 |
+
"num_attention_heads": 32,
|
10 |
+
"num_key_value_heads": 8,
|
11 |
+
"hidden_size": 4096,
|
12 |
+
"norm_epsilon": 1e-05,
|
13 |
+
"vocab_size": 128256,
|
14 |
+
"max_position_embeddings": 8192,
|
15 |
+
"hidden_act": "silu",
|
16 |
+
"use_parallel_embedding": true,
|
17 |
+
"embedding_sharding_dim": 0,
|
18 |
+
"quantization": {
|
19 |
+
"quant_algo": "W4A16_AWQ",
|
20 |
+
"kv_cache_quant_algo": null,
|
21 |
+
"group_size": 128,
|
22 |
+
"has_zero_point": false,
|
23 |
+
"pre_quant_scale": true,
|
24 |
+
"exclude_modules": [
|
25 |
+
"lm_head"
|
26 |
+
]
|
27 |
+
},
|
28 |
+
"mapping": {
|
29 |
+
"world_size": 1,
|
30 |
+
"tp_size": 1,
|
31 |
+
"pp_size": 1
|
32 |
+
},
|
33 |
+
"head_size": 128,
|
34 |
+
"intermediate_size": 14336,
|
35 |
+
"position_embedding_type": "rope_gpt_neox",
|
36 |
+
"share_embedding_table": false,
|
37 |
+
"residual_mlp": false,
|
38 |
+
"bias": false,
|
39 |
+
"rotary_pct": 1.0,
|
40 |
+
"rank": 0,
|
41 |
+
"decoder": "llama",
|
42 |
+
"rmsnorm": true,
|
43 |
+
"lm_head_bias": false,
|
44 |
+
"rotary_base": 500000.0
|
45 |
+
}
|