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

• Loss: 0.8982

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
2.5391	0.6957	2	2.3916
2.4703	1.7391	5	2.2074
2.1328	2.7826	8	1.9537
1.9709	3.8261	11	1.6769
1.5704	4.8696	14	1.3962
1.3765	5.9130	17	1.1358
1.0594	6.9565	20	1.0275
0.9969	8.0	23	0.9877
0.9485	8.6957	25	0.9700
0.8932	9.7391	28	0.9503
0.8815	10.7826	31	0.9331
0.8229	11.8261	34	0.9216
0.8136	12.8696	37	0.9111
0.7507	13.9130	40	0.9021
0.7373	14.9565	43	0.8982
0.6959	16.0	46	0.9020
0.6651	16.6957	48	0.9060
0.6589	17.7391	51	0.9119
0.5782	18.7826	54	0.9264
0.585	19.8261	57	0.9372
0.511	20.8696	60	0.9604
0.4767	21.9130	63	0.9806

Framework versions

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