config.yaml

# Model Configuration
model_name: microsoft/deberta-v3-small  # Hugging Face model
name: "large"
max_width: 12
hidden_size: 768
dropout: 0.4
fine_tune: true
subtoken_pooling: first
span_mode: markerV0

# Training Parameters
num_steps: 30000
train_batch_size: 3
eval_every: 5000
warmup_ratio: 0.1

# Learning Rate Configuration
lr_encoder: 1e-5
lr_others: 5e-5

# Directory Paths
root_dir: gliner_logs
train_data: "data.json" # see https://github.com/urchade/GLiNER/tree/main/data
val_data_dir: "none"
# "NER_datasets": val data from the paper can be obtained from "https://drive.google.com/file/d/1T-5IbocGka35I7X3CE6yKe5N_Xg2lVKT/view"

# Pretrained Model Path
# Use "none" if no pretrained model is being used
prev_path: "none"

# Advanced Training Settings
size_sup: -1
max_types: 25
shuffle_types: true
random_drop: true
max_neg_type_ratio: 1
max_len: 384
freeze_token_rep: false