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

• Loss: 0.8270

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
2.2096	0.6957	2	2.1129
2.167	1.7391	5	1.9558
1.8726	2.7826	8	1.7428
1.7678	3.8261	11	1.5017
1.3895	4.8696	14	1.2525
1.234	5.9130	17	1.0325
0.9378	6.9565	20	0.9271
0.8782	8.0	23	0.8920
0.8394	8.6957	25	0.8784
0.7845	9.7391	28	0.8647
0.7863	10.7826	31	0.8503
0.7261	11.8261	34	0.8417
0.7333	12.8696	37	0.8337
0.6709	13.9130	40	0.8289
0.6612	14.9565	43	0.8270
0.6253	16.0	46	0.8289
0.6012	16.6957	48	0.8323
0.5792	17.7391	51	0.8385
0.5162	18.7826	54	0.8561
0.5219	19.8261	57	0.8603
0.445	20.8696	60	0.8802
0.4396	21.9130	63	0.9046

Framework versions

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