DipanjanSanyal/wikipedia_sample_tiny_gpt2_base

<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->

wikipediasampletinygpt2base

This model is a fine-tuned version of gpt2 on the None dataset. It achieves the following results on the evaluation set:

• —Loss: 5.7538

Model description

This model is a result of an educational attempt of pre-training.

• —Data: 30,000 sample wikipedia articles
• —Tokenier: bert-base-uncased
• —Context: chunks of exactly 64 tokens with an overlap of 16 tokens
• —Initialization: <code>GPT2LMHead()</code> i.e. GPT2 structure without trained weights, and much smaller size and smaller number of transformer layers
• —Training: trained for 3 epochs

There is a previous version of the model (which I pasued in between because of budget). Below are the details for that:

• —Data: Same
• —Tokenizer: Same
• —Context: chunks of exactly 64 tokens with an overlap of 48 tokens (so a much larger dataset)
• —Initialization: Same
• —Training: trained for 15 epochs

To call this version, please use <code>...frompretrained('DipanjanSanyal/wikipediasampletinygpt2_base', revision = 87d9aa1eb492a5c20db562f113f07b8f8522f5d2')</code>

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: 3e-05
• —trainbatchsize: 32
• —evalbatchsize: 32
• —seed: 42
• —optimizer: Use OptimizerNames.ADAMWTORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments
• —lrschedulertype: linear
• —num_epochs: 3
• —mixedprecisiontraining: Native AMP

Training results

Framework versions

• —Transformers 4.51.3
• —Pytorch 2.6.0+cu124
• —Datasets 3.6.0
• —Tokenizers 0.21.1