jkazdan/llama3b-real-only-sftsd1

llama-real-and-synthetic-sftsd1

This model is a fine-tuned version of meta-llama/Llama-3.2-3B-Instruct on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.9714
• Num Input Tokens Seen: 1904984

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: 8e-06
• train_batch_size: 8
• eval_batch_size: 16
• seed: 1
• gradient_accumulation_steps: 16
• total_train_batch_size: 128
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: constant_with_warmup
• lr_scheduler_warmup_ratio: 0.05
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss	Input Tokens Seen
No log	0	0	1.5937	0
1.3209	0.0856	5	1.2827	161496
1.1224	0.1711	10	1.1693	322408
1.0851	0.2567	15	1.1131	489712
1.06	0.3422	20	1.0862	655168
1.0298	0.4278	25	1.0657	818584
0.9931	0.5134	30	1.0473	978832
0.9974	0.5989	35	1.0294	1143648
0.9779	0.6845	40	1.0105	1316224
0.9688	0.7701	45	0.9915	1476728
0.9191	0.8556	50	0.9793	1645280
0.9213	0.9412	55	0.9748	1807968

Framework versions

• Transformers 4.44.0
• Pytorch 2.4.0+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1