Llama-31-8B_task-1_120-samples_config-2_full_auto

This model is a fine-tuned version of meta-llama/Meta-Llama-3.1-8B-Instruct on the GaetanMichelet/chat-60_ft_task-1_auto and the GaetanMichelet/chat-120_ft_task-1_auto datasets. It achieves the following results on the evaluation set:

• Loss: 0.7873

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
2.1572	0.9091	5	2.1149
2.0048	2.0	11	1.8150
1.651	2.9091	16	1.5524
1.2397	4.0	22	1.1141
0.9602	4.9091	27	0.9207
0.8646	6.0	33	0.8729
0.7895	6.9091	38	0.8490
0.7762	8.0	44	0.8273
0.7412	8.9091	49	0.8120
0.6669	10.0	55	0.7971
0.6184	10.9091	60	0.7873
0.5857	12.0	66	0.7943
0.5374	12.9091	71	0.8102
0.4629	14.0	77	0.8345
0.396	14.9091	82	0.8902
0.336	16.0	88	0.9196
0.2438	16.9091	93	1.0492
0.1943	18.0	99	1.1073

Framework versions

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