4e-5 run test, lower batch size (bs4)
output_dir="qwen-audio-beat-trainer-t5",
num_train_epochs=1,
per_device_train_batch_size=4,
gradient_accumulation_steps=1,
learning_rate=4e-5,
max_grad_norm=0.2,
warmup_steps=20,
bf16=True,
remove_unused_columns=False,
gradient_checkpointing=True,
logging_steps=1,
save_strategy="steps",
save_steps=25,
- Downloads last month
- 3
Inference Providers
NEW
This model isn't deployed by any Inference Provider.
๐
Ask for provider support
HF Inference deployability: The model has no pipeline_tag.
Model tree for kalomaze/qwensaber-150-t5
Base model
Qwen/Qwen2-Audio-7B