Llama-31-8B_task-1_180-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-1, the GaetanMichelet/chat-120_ft_task-1 and the GaetanMichelet/chat-180_ft_task-1 datasets. It achieves the following results on the evaluation set:

• Loss: 0.8726

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
2.375	0.9412	8	2.2872
1.9761	2.0	17	1.8512
1.5652	2.9412	25	1.3759
1.0274	4.0	34	1.0474
0.9811	4.9412	42	0.9874
0.9464	6.0	51	0.9454
0.8491	6.9412	59	0.9103
0.8043	8.0	68	0.8866
0.7279	8.9412	76	0.8726
0.6704	10.0	85	0.8894
0.6154	10.9412	93	0.9156
0.4469	12.0	102	0.9677
0.3873	12.9412	110	1.0959
0.3086	14.0	119	1.2353
0.1952	14.9412	127	1.3687
0.1665	16.0	136	1.4523

Framework versions

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