Llama-31-8B_task-2_120-samples_config-2_full

This model is a fine-tuned version of meta-llama/Meta-Llama-3.1-8B-Instruct on the GaetanMichelet/chat-60_ft_task-2 and the GaetanMichelet/chat-120_ft_task-2 datasets. It achieves the following results on the evaluation set:

• Loss: 1.0340

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: 0.0001
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• gradient_accumulation_steps: 16
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 50

Training results

Training Loss	Epoch	Step	Validation Loss
1.6102	0.9091	5	1.5743
1.5449	2.0	11	1.4671
1.3801	2.9091	16	1.3715
1.272	4.0	22	1.2587
1.1581	4.9091	27	1.1534
1.0495	6.0	33	1.0936
1.0426	6.9091	38	1.0769
1.0149	8.0	44	1.0613
0.9765	8.9091	49	1.0516
0.9857	10.0	55	1.0432
0.9438	10.9091	60	1.0381
0.9103	12.0	66	1.0347
0.8842	12.9091	71	1.0340
0.872	14.0	77	1.0358
0.7743	14.9091	82	1.0479
0.7851	16.0	88	1.0549
0.7288	16.9091	93	1.0722
0.6647	18.0	99	1.0962
0.6477	18.9091	104	1.1323
0.6269	20.0	110	1.1614

Framework versions

• PEFT 0.12.0
• Transformers 4.44.0
• Pytorch 2.1.2+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1