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

• Loss: 0.7848

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
2.1079	0.9412	8	2.0439
1.7735	2.0	17	1.6643
1.4106	2.9412	25	1.2322
0.901	4.0	34	0.9259
0.8657	4.9412	42	0.8770
0.846	6.0	51	0.8422
0.7582	6.9412	59	0.8153
0.7203	8.0	68	0.7906
0.6598	8.9412	76	0.7848
0.5979	10.0	85	0.7954
0.5547	10.9412	93	0.8095
0.4007	12.0	102	0.8623
0.3489	12.9412	110	0.9627
0.2894	14.0	119	1.0531
0.1972	14.9412	127	1.1217
0.1682	16.0	136	1.2316

Framework versions

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