jkazdan/llama3b-real-only-sftsd2

llama-real-and-synthetic-sftsd2

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: 0.9722
• Num Input Tokens Seen: 1907704

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: 2
• 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	1.5881	0
1.3233	0.0856	5	1.2853	161264
1.1841	0.1711	10	1.1799	323576
1.0739	0.2567	15	1.1151	485960
1.0499	0.3422	20	1.0835	648472
1.0249	0.4278	25	1.0623	813392
1.0022	0.5134	30	1.0431	978584
1.0058	0.5989	35	1.0266	1144056
0.9862	0.6845	40	1.0096	1304592
0.93	0.7701	45	0.9925	1474856
0.9372	0.8556	50	0.9802	1637096
0.914	0.9412	55	0.9751	1807632

Framework versions

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