MIDIGen_Llama3_8B-Instruct_finetuned_10epochs

This model is a fine-tuned version of meta-llama/Meta-Llama-3-8B-Instruct on the generator dataset. It achieves the following results on the evaluation set:

• Loss: 0.6254

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 8
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 2
• total_train_batch_size: 128
• total_eval_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: constant
• lr_scheduler_warmup_ratio: 0.03
• num_epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss
0.6577	0.9996	1386	0.6565
0.6373	2.0	2773	0.6403
0.6294	2.9996	4159	0.6344
0.6256	4.0	5546	0.6313
0.6257	4.9996	6932	0.6294
0.6255	6.0	8319	0.6280
0.6201	6.9996	9705	0.6271
0.6205	8.0	11092	0.6264
0.6205	8.9996	12478	0.6259
0.6252	9.9964	13860	0.6254

Framework versions

• PEFT 0.10.0
• Transformers 4.40.0
• Pytorch 2.1.0
• Datasets 2.18.0
• Tokenizers 0.19.1