daibeiya's picture
model upload
4d46bcb verified
metadata
library_name: transformers
base_model: /fs-computility/plm/linzhouhan/daibeiya/models/gpt2
tags:
  - generated_from_trainer
datasets:
  - openwebtext
model-index:
  - name: gpt2_base_contextlm_l0212_add_lnnorm_wodetach_v2_lr_bf16_lr1e-3
    results: []

gpt2_base_contextlm_l0212_add_lnnorm_wodetach_v2_lr_bf16_lr1e-3

This model is a fine-tuned version of /fs-computility/plm/linzhouhan/daibeiya/models/gpt2 on the openwebtext dataset. It achieves the following results on the evaluation set:

  • Loss: 3.0300

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: 0.001
  • train_batch_size: 16
  • eval_batch_size: 8
  • seed: 42
  • distributed_type: multi-GPU
  • num_devices: 16
  • gradient_accumulation_steps: 2
  • total_train_batch_size: 512
  • total_eval_batch_size: 128
  • optimizer: Use adamw_torch with betas=(0.9,0.95) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
  • lr_scheduler_type: cosine
  • lr_scheduler_warmup_ratio: 0.05
  • num_epochs: 1.0

Training results

Training Loss Epoch Step Validation Loss
3.9616 0.0580 1000 3.8911
3.5512 0.1160 2000 3.4861
3.4279 0.1741 3000 3.3560
3.3471 0.2321 4000 3.2811
3.2957 0.2901 5000 3.2321
3.2677 0.3481 6000 3.1945
3.225 0.4062 7000 3.1653
3.2051 0.4642 8000 3.1390
3.1816 0.5222 9000 3.1161
3.1583 0.5802 10000 3.0971
3.1464 0.6383 11000 3.0794
3.1365 0.6963 12000 3.0645
3.1256 0.7543 13000 3.0509
3.1073 0.8123 14000 3.0417
3.108 0.8703 15000 3.0349
3.098 0.9284 16000 3.0312
3.092 0.9864 17000 3.0301

Framework versions

  • Transformers 4.51.3
  • Pytorch 2.3.0+cu121
  • Datasets 4.0.0
  • Tokenizers 0.21.4