rocker417/llama3.2-3B-added-tokens-wiki-cursor-backspace-left-right-cosine-loss-2-lora

llama3.2-3B-added-tokens-1.1-1-epoch

This model is a fine-tuned version of rocker417/llama3.2-3B-added-tokens on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 2.1410

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: 4
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• optimizer: Use adamw_8bit with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: reduce_lr_on_plateau
• lr_scheduler_warmup_ratio: 0.03
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss
3.1845	0.0988	1000	2.5983
2.5334	0.1976	2000	2.4032
2.4184	0.2965	3000	2.3327
2.336	0.3953	4000	2.2915
2.2717	0.4941	5000	2.2409
2.2903	0.5929	6000	2.2048
2.248	0.6918	7000	2.1751
2.2132	0.7906	8000	2.1653
2.1643	0.8894	9000	2.1480
2.1624	0.9882	10000	2.1410

Framework versions

• Transformers 4.55.4
• Pytorch 2.8.0+cu128
• Datasets 3.6.0
• Tokenizers 0.21.1