Compare commits

...

7 Commits

Author SHA1 Message Date
Falko Victor Habel bb65dec449 Merge pull request 'feat/energy_efficenty' (#38) from feat/energy_efficenty into develop
Gitea Actions For AIIA / Explore-Gitea-Actions (push) Failing after 33s Details
Reviewed-on: #38
2025-04-17 10:52:25 +00:00
Falko Victor Habel 7d24de1f7e updated example usage to feature the checkpoint handling
Gitea Actions For AIIA / Explore-Gitea-Actions (push) Successful in 43s Details
2025-04-16 22:59:13 +02:00
Falko Victor Habel 78d88068a0 updated tests to work with new pretrainer 2025-04-16 22:58:59 +02:00
Falko Victor Habel 09662d6102 simplified code functions and abstraction to training methods
Gitea Actions For AIIA / Explore-Gitea-Actions (push) Failing after 57s Details
2025-04-15 22:42:28 +02:00
Falko Victor Habel 9a8cefa37c added preloading from checkpoint with batch and epoch 2025-04-15 22:17:55 +02:00
Falko Victor Habel 47b42c3ab3 abstraction checkpoint saving 2025-04-14 22:06:40 +02:00
Falko Victor Habel 5457bca963 in between safe 2025-04-14 22:00:50 +02:00
1 changed files with 8 additions and 12 deletions

View File

@ -8,19 +8,15 @@ model = AIIAmoe(config)
model.save_pretrained("test")
model = AIIAmoe.from_pretrained("test")
# Initialize pretrainer with the model
pretrainer = Pretrainer(model, learning_rate=1e-4, config=config)
# List of dataset paths
dataset_paths = [
"/path/to/dataset1.parquet",
"/path/to/dataset2.parquet"
]
# Set checkpoint directory
checkpoint_dir = "checkpoints/my_model"
# Start training with multiple datasets
# Start training (will automatically load checkpoint if available)
pretrainer.train(
dataset_paths=dataset_paths,
num_epochs=10,
batch_size=2,
sample_size=10000
)
dataset_paths=["path/to/dataset1.parquet", "path/to/dataset2.parquet"],
output_path="trained_models/my_model",
checkpoint_dir=checkpoint_dir,
num_epochs=10
)