train_wsc_1754652157

This model is a fine-tuned version of meta-llama/Meta-Llama-3-8B-Instruct on the wsc dataset. It achieves the following results on the evaluation set:

• Loss: 0.3849
• Num Input Tokens Seen: 490000

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-05
• train_batch_size: 4
• eval_batch_size: 4
• seed: 123
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 10.0

Training results

Training Loss	Epoch	Step	Validation Loss	Input Tokens Seen
13.9838	0.504	63	13.8711	25504
9.9251	1.008	126	9.7372	49696
6.0466	1.512	189	5.3090	74112
1.6788	2.016	252	1.9802	99136
1.0818	2.52	315	0.7968	123904
0.8394	3.024	378	0.5601	148736
0.5184	3.528	441	0.4782	174432
0.3853	4.032	504	0.4613	198656
0.4549	4.536	567	0.4388	224032
0.4193	5.04	630	0.4215	247424
0.3691	5.5440	693	0.4073	271232
0.3746	6.048	756	0.4005	295728
0.427	6.552	819	0.4005	320464
0.3347	7.056	882	0.4107	345856
0.331	7.5600	945	0.4089	371040
0.4144	8.064	1008	0.3849	395216
0.3779	8.568	1071	0.3869	419184
0.3714	9.072	1134	0.3899	444560
0.3858	9.576	1197	0.3862	469104

Framework versions

• PEFT 0.15.2
• Transformers 4.51.3
• Pytorch 2.8.0+cu128
• Datasets 3.6.0
• Tokenizers 0.21.1