| { | |
| "model_type": "bert", # Change this based on your model type (e.g., gpt2, roberta, etc.) | |
| "num_labels": 2, # Number of output labels for classification (adjust for your task) | |
| "hidden_size": 768, # Hidden layer size (depends on your model) | |
| "intermediate_size": 3072, # Intermediate size for feed-forward layers | |
| "max_position_embeddings": 512, # Max token length | |
| "num_attention_heads": 12, # Number of attention heads | |
| "num_hidden_layers": 12, # Number of hidden layers in your transformer model | |
| "vocab_size": 30522, # Size of tokenizer vocabulary | |
| "hidden_act": "gelu", # Activation function in hidden layers | |
| "initializer_range": 0.02, # Initialization range for weights | |
| "layer_norm_eps": 1e-12, # Layer normalization epsilon | |
| "pad_token_id": 0, # Padding token ID (usually 0) | |
| "type_vocab_size": 2, # Type vocab size (typically 2 for sentence pairs) | |
| "attention_probs_dropout_prob": 0.1, # Dropout probability for attention layers | |
| "hidden_dropout_prob": 0.1, # Dropout probability for hidden layers | |
| "use_cache": true, # Whether to cache past keys/values | |
| "model_version": "1.0", # Your model version | |
| "tokenizer_class": "BertTokenizer", # Tokenizer class (adjust for your model type) | |
| "classifier_dropout": null, # Optional dropout for classification head | |
| "architectures": [ | |
| "BertForSequenceClassification" # Model architecture type | |
| ] | |
| } |