llama-3.2-3b-sft-2

This model is a fine-tuned version of tanliboy/llama-3.2-3b on the tanliboy/OpenHermes-2.5-reformat dataset. It achieves the following results on the evaluation set:

• Loss: 0.6744

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: 5e-06
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 2
• total_train_batch_size: 128
• total_eval_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 1000
• training_steps: 10000

Training results

Training Loss	Epoch	Step	Validation Loss
0.7792	0.0673	500	0.7726
0.7496	0.1345	1000	0.7444
0.7243	0.2018	1500	0.7296
0.7178	0.2691	2000	0.7197
0.7077	0.3363	2500	0.7127
0.6992	0.4036	3000	0.7066
0.6992	0.4708	3500	0.7012
0.6945	0.5381	4000	0.6965
0.6879	0.6054	4500	0.6920
0.6901	0.6726	5000	0.6879
0.6759	0.7399	5500	0.6844
0.6752	0.8072	6000	0.6812
0.6826	0.8744	6500	0.6783
0.6804	0.9417	7000	0.6758
0.6131	1.0089	7500	0.6764
0.6012	1.0762	8000	0.6758
0.6136	1.1435	8500	0.6751
0.6127	1.2107	9000	0.6747
0.6076	1.2780	9500	0.6745
0.6033	1.3453	10000	0.6744

Framework versions

• Transformers 4.44.2
• Pytorch 2.4.0+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1