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

• Loss: 0.8694

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
2.448	0.9091	5	2.3732
2.2361	2.0	11	2.0279
1.8253	2.9091	16	1.7251
1.3791	4.0	22	1.2330
1.0878	4.9091	27	1.0337
0.9771	6.0	33	0.9739
0.8967	6.9091	38	0.9426
0.8815	8.0	44	0.9129
0.816	8.9091	49	0.8952
0.748	10.0	55	0.8762
0.6939	10.9091	60	0.8727
0.6449	12.0	66	0.8694
0.5874	12.9091	71	0.8921
0.4934	14.0	77	0.9429
0.4382	14.9091	82	1.0083
0.347	16.0	88	1.0592
0.2565	16.9091	93	1.1458
0.1926	18.0	99	1.2523
0.1477	18.9091	104	1.4710

Framework versions

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