gpt2-medium-vericava-posts-v3

This is a model trained from scratch using the shape of gpt2-medium on a dataset of my posts on the Internet.

It achieves the following results on the evaluation set:

• Loss: 6.4732

Model description

It generates text resembling what I post on the Internet.

Intended uses & limitations

CAUTION: It may produce something I'd never say. I do not impose any restriction(s) on the use of this model.

Training and evaluation data

Twitter/X: https://x.com/vericava

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0005
• train_batch_size: 128
• eval_batch_size: 128
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 1024
• optimizer: Use OptimizerNames.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_steps: 1000
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
2.1117	11.1176	100	6.5645
1.461	22.2353	200	5.6692
1.3005	33.3529	300	5.3980
1.1776	44.4706	400	5.2793
1.0325	55.5882	500	5.3445
0.8629	66.7059	600	5.5766
0.6811	77.8235	700	5.8608
0.4943	88.9412	800	6.1404
0.3243	100.0	900	6.4732

Framework versions

• Transformers 4.52.4
• Pytorch 2.6.0+cu124
• Datasets 3.6.0
• Tokenizers 0.21.1