Llama-31-8B_task-1_120-samples_config-4

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: 1.2635

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: 1e-05
• 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: 150

Training results

Training Loss	Epoch	Step	Validation Loss
2.121	0.9091	5	2.1020
2.0709	2.0	11	2.0931
2.0454	2.9091	16	2.0755
2.0502	4.0	22	2.0472
2.0511	4.9091	27	2.0100
1.9554	6.0	33	1.9472
1.8921	6.9091	38	1.8795
1.8104	8.0	44	1.7813
1.7636	8.9091	49	1.6937
1.6011	10.0	55	1.6142
1.5128	10.9091	60	1.5751
1.4277	12.0	66	1.5353
1.4998	12.9091	71	1.5001
1.4154	14.0	77	1.4583
1.4201	14.9091	82	1.4252
1.3364	16.0	88	1.3921
1.2762	16.9091	93	1.3691
1.2851	18.0	99	1.3437
1.2239	18.9091	104	1.3261
1.221	20.0	110	1.3084
1.2011	20.9091	115	1.2951
1.1433	22.0	121	1.2824
1.1579	22.9091	126	1.2746
1.0871	24.0	132	1.2680
1.0745	24.9091	137	1.2635
1.0006	26.0	143	1.2674
0.9628	26.9091	148	1.2689
0.9237	28.0	154	1.2717
0.8824	28.9091	159	1.2880
0.8706	30.0	165	1.2961
0.8328	30.9091	170	1.3266
0.7667	32.0	176	1.3447

Framework versions

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