Create config.json
Browse files- config.json +38 -0
config.json
ADDED
@@ -0,0 +1,38 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"model_type": "CustomModel",
|
3 |
+
"architecture": "Feedforward Neural Network with Ba-inspired Activation Function",
|
4 |
+
"input_size": 512,
|
5 |
+
"hidden_size": 128,
|
6 |
+
"output_size": 768,
|
7 |
+
"activation_function": {
|
8 |
+
"name": "Ba-inspired Activation",
|
9 |
+
"details": {
|
10 |
+
"weights_initialization": "Random Normal Distribution",
|
11 |
+
"a": 0.5,
|
12 |
+
"epsilon": 1e-6,
|
13 |
+
"function": "epsilon * torch.cos(np.pi * a * fractional_inspired * torch.log(torch.abs(fractional_inspired) + epsilon))"
|
14 |
+
}
|
15 |
+
},
|
16 |
+
"tokenizer": "bert-base-uncased",
|
17 |
+
"training_details": {
|
18 |
+
"optimizer": "AdamW",
|
19 |
+
"learning_rate": 5e-5,
|
20 |
+
"loss_function": "CrossEntropyLoss",
|
21 |
+
"batch_size": 8,
|
22 |
+
"epochs": 3,
|
23 |
+
"dataset": "Custom Dataset from JSON Lines File",
|
24 |
+
"dataset_preprocessing": {
|
25 |
+
"max_length": 512,
|
26 |
+
"padding": true,
|
27 |
+
"truncation": true
|
28 |
+
}
|
29 |
+
},
|
30 |
+
"performance": {
|
31 |
+
"final_accuracy": "Dependent on specific run and dataset",
|
32 |
+
"final_loss": "Dependent on specific run and dataset"
|
33 |
+
},
|
34 |
+
"usage": {
|
35 |
+
"inference": "Model can be used for tasks requiring sequence classification. Ensure input size matches model configuration.",
|
36 |
+
"additional_notes": "Model and tokenizer need to be loaded with Hugging Face's transformers library for usage."
|
37 |
+
}
|
38 |
+
}
|