Llama-31-8B_task-1_120-samples_config-1_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 and the GaetanMichelet/chat-120_ft_task-1_auto datasets. It achieves the following results on the evaluation set:

• Loss: 0.7838

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
2.1444	1.0	11	1.9951
1.5597	2.0	22	1.5854
1.1135	3.0	33	1.0142
0.8595	4.0	44	0.8794
0.7701	5.0	55	0.8356
0.7434	6.0	66	0.8024
0.6039	7.0	77	0.7895
0.5441	8.0	88	0.7838
0.4965	9.0	99	0.8283
0.353	10.0	110	0.9092
0.2505	11.0	121	1.0033
0.2204	12.0	132	1.1738
0.1355	13.0	143	1.3070
0.1041	14.0	154	1.3560
0.0759	15.0	165	1.3713

Framework versions

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