Upload base_resnet_lstm/config.yaml with huggingface_hub
Browse files- base_resnet_lstm/config.yaml +38 -0
base_resnet_lstm/config.yaml
ADDED
|
@@ -0,0 +1,38 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
seed: 42
|
| 2 |
+
paths:
|
| 3 |
+
raw_images_dir: data/raw/images
|
| 4 |
+
captions_file: data/raw/captions.txt
|
| 5 |
+
processed_dir: data/processed
|
| 6 |
+
features_dir: data/features
|
| 7 |
+
models_dir: models
|
| 8 |
+
split:
|
| 9 |
+
train_ratio: 0.8
|
| 10 |
+
val_ratio: 0.1
|
| 11 |
+
test_ratio: 0.1
|
| 12 |
+
vocab:
|
| 13 |
+
min_freq: 5
|
| 14 |
+
max_len: 35
|
| 15 |
+
encoder:
|
| 16 |
+
type: resnet50
|
| 17 |
+
feature_dim: 2048
|
| 18 |
+
freeze: true
|
| 19 |
+
decoder:
|
| 20 |
+
type: lstm
|
| 21 |
+
embed_dim: 256
|
| 22 |
+
hidden_dim: 512
|
| 23 |
+
num_layers: 1
|
| 24 |
+
dropout: 0.0
|
| 25 |
+
training:
|
| 26 |
+
batch_size: 32
|
| 27 |
+
epochs: 30
|
| 28 |
+
lr: 0.001
|
| 29 |
+
optimizer: adam
|
| 30 |
+
lr_scheduler_patience: 2
|
| 31 |
+
lr_scheduler_factor: 0.5
|
| 32 |
+
early_stop_patience: 5
|
| 33 |
+
num_workers: 2
|
| 34 |
+
device: auto
|
| 35 |
+
inference:
|
| 36 |
+
decoding: greedy
|
| 37 |
+
beam_width: 3
|
| 38 |
+
run_name: base_resnet_lstm
|