jkazdan/llama3b-real-only-sftsd0

llama-real-and-synthetic-sftsd0

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: 1.0839
• Num Input Tokens Seen: 1902768

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: 0
• 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	1.5725	0
1.3115	0.0856	5	1.2522	162144
1.1433	0.1711	10	1.1670	326328
1.0894	0.2567	15	1.1623	488416
1.1305	0.3422	20	1.1582	652112
1.1378	0.4278	25	1.1555	812832
1.1177	0.5134	30	1.1481	979368
1.112	0.5989	35	1.1360	1141064
1.0575	0.6845	40	1.1230	1309504
1.0593	0.7701	45	1.1114	1472200
1.0479	0.8556	50	1.1004	1636976
1.0367	0.9412	55	1.0896	1806992

Framework versions

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