Canstralian
commited on
Update model/config.json
Browse files- model/config.json +14 -0
model/config.json
CHANGED
@@ -1,8 +1,22 @@
|
|
1 |
{
|
|
|
2 |
"model_name": "RabbitRedux",
|
3 |
"version": "1.0",
|
4 |
"author": "Canstralian",
|
5 |
"description": "This configuration file defines the model settings for RabbitRedux.",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
6 |
"training_params": {
|
7 |
"batch_size": 32,
|
8 |
"learning_rate": 0.001,
|
|
|
1 |
{
|
2 |
+
"model_type": "gpt2", // Change to the appropriate model type
|
3 |
"model_name": "RabbitRedux",
|
4 |
"version": "1.0",
|
5 |
"author": "Canstralian",
|
6 |
"description": "This configuration file defines the model settings for RabbitRedux.",
|
7 |
+
"vocab_size": 50257, // Adjust to your model's vocabulary size
|
8 |
+
"hidden_size": 768, // Adjust according to your architecture
|
9 |
+
"num_hidden_layers": 12,// Adjust according to your architecture
|
10 |
+
"num_attention_heads": 12, // Adjust according to your architecture
|
11 |
+
"intermediate_size": 3072, // Adjust according to your architecture
|
12 |
+
"hidden_act": "gelu", // Activation function used in the model
|
13 |
+
"layer_norm_epsilon": 1e-5, // Epsilon for layer normalization
|
14 |
+
"initializer_range": 0.02, // Weight initialization range
|
15 |
+
"dropout": 0.1, // Dropout rate for layers
|
16 |
+
"attention_probs_dropout_prob": 0.1, // Dropout for attention probabilities
|
17 |
+
"pad_token_id": 50256, // Token ID for padding
|
18 |
+
"eos_token_id": 50256, // Token ID for end-of-sequence
|
19 |
+
"bos_token_id": 50256, // Token ID for beginning-of-sequence
|
20 |
"training_params": {
|
21 |
"batch_size": 32,
|
22 |
"learning_rate": 0.001,
|