config.json
Browse files- config.json +34 -0
config.json
ADDED
@@ -0,0 +1,34 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"model_type": "unknown", // Specify the base model type if known (e.g., "bert", "gpt-2")
|
3 |
+
"architectures": ["unknown"], // Specify the architecture if known (e.g., "BertForMaskedLM", "GPT2LMHeadModel")
|
4 |
+
"finetuned_from": "vilsonrodrigues/falcon-7b-instruct-sharded",
|
5 |
+
"tasks": "conversational", // Specify the task if it's a standard task in Hugging Face
|
6 |
+
|
7 |
+
// Include training hyperparameters
|
8 |
+
"learning_rate": 0.0002,
|
9 |
+
"train_batch_size": 8,
|
10 |
+
"eval_batch_size": 8,
|
11 |
+
"seed": 42,
|
12 |
+
"gradient_accumulation_steps": 8,
|
13 |
+
"total_train_batch_size": 64,
|
14 |
+
"optimizer": "Adam with betas=(0.9, 0.999) and epsilon=1e-08",
|
15 |
+
"lr_scheduler_type": "cosine",
|
16 |
+
"lr_scheduler_warmup_ratio": 0.03,
|
17 |
+
"training_steps": 5,
|
18 |
+
|
19 |
+
// Include library versions used for training
|
20 |
+
"framework": "Pytorch",
|
21 |
+
"pytorch_version": "2.1.0+cu118",
|
22 |
+
"transformers_version": "4.34.1",
|
23 |
+
"datasets_version": "2.14.6",
|
24 |
+
"tokenizers_version": "0.14.1",
|
25 |
+
|
26 |
+
// Other relevant model configuration parameters
|
27 |
+
"vocab_size": 0, // Specify the vocabulary size
|
28 |
+
"hidden_size": 0, // Specify the hidden layer size
|
29 |
+
"num_hidden_layers": 0, // Specify the number of hidden layers
|
30 |
+
"num_attention_heads": 0, // Specify the number of attention heads
|
31 |
+
// ... other model-specific parameters
|
32 |
+
|
33 |
+
"unknown_parameters": "This model has been fine-tuned on an unknown dataset with specific parameters and may not conform to the standard tasks or architectures. Please refer to the model documentation for more details."
|
34 |
+
}
|