Llama-31-8B_task-2_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-2_auto and the GaetanMichelet/chat-120_ft_task-2_auto datasets. It achieves the following results on the evaluation set:

• Loss: 0.9959

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
1.4871	1.0	11	1.4805
1.2753	2.0	22	1.3422
1.1856	3.0	33	1.1891
1.0973	4.0	44	1.0591
1.0204	5.0	55	1.0277
0.9819	6.0	66	1.0086
0.948	7.0	77	0.9993
0.9103	8.0	88	0.9959
0.8302	9.0	99	1.0063
0.7247	10.0	110	1.0327
0.7159	11.0	121	1.0794
0.5506	12.0	132	1.1349
0.5606	13.0	143	1.1961
0.3523	14.0	154	1.2591
0.2739	15.0	165	1.3140

Framework versions

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