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

• Loss: 0.8656

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
2.3371	1.0	11	2.2386
1.7745	2.0	22	1.7577
1.2294	3.0	33	1.1298
0.9781	4.0	44	0.9825
0.8705	5.0	55	0.9252
0.8269	6.0	66	0.8844
0.6627	7.0	77	0.8656
0.6053	8.0	88	0.8689
0.5258	9.0	99	0.9307
0.3668	10.0	110	1.0742
0.248	11.0	121	1.2238
0.2039	12.0	132	1.3416
0.1357	13.0	143	1.4583
0.1016	14.0	154	1.5564

Framework versions

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