Llama-31-8B_task-1_60-samples_config-1_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 dataset. It achieves the following results on the evaluation set:

• Loss: 0.8973

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: 8
• total_train_batch_size: 8
• 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
2.5176	0.8696	5	2.3282
2.1942	1.9130	11	1.9932
1.8164	2.9565	17	1.6236
1.3441	4.0	23	1.1448
0.987	4.8696	28	1.0040
0.9101	5.9130	34	0.9508
0.8517	6.9565	40	0.9197
0.7732	8.0	46	0.8986
0.7365	8.8696	51	0.8973
0.6133	9.9130	57	0.9109
0.5483	10.9565	63	0.9300
0.4109	12.0	69	0.9910
0.285	12.8696	74	1.0815
0.2088	13.9130	80	1.2331
0.1666	14.9565	86	1.4608
0.1074	16.0	92	1.5691

Framework versions

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