Update PersadianNanoV4Config.json
Browse files- PersadianNanoV4Config.json +23 -34
PersadianNanoV4Config.json
CHANGED
|
@@ -1,36 +1,25 @@
|
|
| 1 |
{
|
| 2 |
-
|
| 3 |
-
|
| 4 |
-
|
| 5 |
-
|
| 6 |
-
|
| 7 |
-
|
| 8 |
-
|
| 9 |
-
|
| 10 |
-
|
| 11 |
-
|
| 12 |
-
|
| 13 |
-
|
| 14 |
-
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
|
| 21 |
-
|
| 22 |
-
|
| 23 |
-
|
| 24 |
-
|
| 25 |
-
"eos_token_id": 50256,
|
| 26 |
-
|
| 27 |
-
"rope_theta": 10000.0,
|
| 28 |
-
|
| 29 |
-
"dropout": 0.1,
|
| 30 |
-
"layer_norm_eps": 1e-5,
|
| 31 |
-
|
| 32 |
-
"attention_dropout": 0.0,
|
| 33 |
-
"use_flash_attention": true,
|
| 34 |
-
|
| 35 |
-
"torch_dtype": "float16"
|
| 36 |
}
|
|
|
|
| 1 |
{
|
| 2 |
+
"hidden_size": 512,
|
| 3 |
+
"intermediate_size": 1024,
|
| 4 |
+
"num_hidden_layers": 12,
|
| 5 |
+
"num_attention_heads": 8,
|
| 6 |
+
"num_key_value_heads": 4,
|
| 7 |
+
"num_experts": 4,
|
| 8 |
+
"num_experts_per_tok": 2,
|
| 9 |
+
"progressive_experts": true,
|
| 10 |
+
"min_experts": 1,
|
| 11 |
+
"max_experts": 2,
|
| 12 |
+
"use_hyper_connection": true,
|
| 13 |
+
"use_compressed_attention": true,
|
| 14 |
+
"compress_ratio": 4,
|
| 15 |
+
"max_position_embeddings": 2048,
|
| 16 |
+
"vocab_size": 50257,
|
| 17 |
+
"bos_token_id": 50256,
|
| 18 |
+
"eos_token_id": 50256,
|
| 19 |
+
"rope_theta": 10000.0,
|
| 20 |
+
"dropout": 0.1,
|
| 21 |
+
"layer_norm_eps": 1e-5,
|
| 22 |
+
"attention_dropout": 0.0,
|
| 23 |
+
"use_flash_attention": true,
|
| 24 |
+
"torch_dtype": "float16"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 25 |
}
|