Upload generation_config.json (max2-nano weights)
Browse files- generation_config.json +22 -0
generation_config.json
ADDED
|
@@ -0,0 +1,22 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"architectures": [
|
| 3 |
+
"Max2ForCausalLM"
|
| 4 |
+
],
|
| 5 |
+
"model_type": "max2",
|
| 6 |
+
"model_name": "max2-nano",
|
| 7 |
+
"hidden_size": 768,
|
| 8 |
+
"intermediate_size": 2048,
|
| 9 |
+
"num_hidden_layers": 12,
|
| 10 |
+
"num_attention_heads": 12,
|
| 11 |
+
"num_key_value_heads": 3,
|
| 12 |
+
"vocab_size": 32000,
|
| 13 |
+
"max_position_embeddings": 4096,
|
| 14 |
+
"rope_theta": 10000.0,
|
| 15 |
+
"use_moe": true,
|
| 16 |
+
"num_experts": 4,
|
| 17 |
+
"num_experts_per_tok": 1,
|
| 18 |
+
"expert_hidden_size": 512,
|
| 19 |
+
"rms_norm_eps": 1e-06,
|
| 20 |
+
"torch_dtype": "float16",
|
| 21 |
+
"transformers_version": "4.40.0"
|
| 22 |
+
}
|