Upload config.json with huggingface_hub
Browse files- config.json +21 -0
config.json
ADDED
|
@@ -0,0 +1,21 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"architectures": [
|
| 3 |
+
"Step1MoEForCausalLM"
|
| 4 |
+
],
|
| 5 |
+
"model_type": "step1",
|
| 6 |
+
"hidden_size": 576,
|
| 7 |
+
"intermediate_size": 5016,
|
| 8 |
+
"num_attention_heads": 9,
|
| 9 |
+
"num_attention_groups": 9,
|
| 10 |
+
"num_hidden_layers": 6,
|
| 11 |
+
"max_seq_len": 65536,
|
| 12 |
+
"vocab_size": 65536,
|
| 13 |
+
"rms_norm_eps": 1e-05,
|
| 14 |
+
"torch_dtype": "bfloat16",
|
| 15 |
+
"moe_every_n_layer": 64,
|
| 16 |
+
"moe_intermediate_size": 4096,
|
| 17 |
+
"moe_num_experts": 8,
|
| 18 |
+
"moe_top_k": 2,
|
| 19 |
+
"use_moe": true,
|
| 20 |
+
"moe_layer_offset": 1
|
| 21 |
+
}
|