Downloads · 30 days
17
31% of all-time downloads
codeparrot-ds
Dohkim/codeparrot-ds
codeparrot-ds is a text generation model from Dohkim. Use it when you need the model to write or continue text. It is set up for transformers. The card lists the license as mit.
should probably proofread and complete it, then remove this comment. --
All-time downloads
54
Public
Parameters
124M
497 MB on disk
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors497 MB · 99%
At a glance
- Task
- Text Generation
- Library
- transformers
- License
- mit
- Model type
- gpt2
- Access
- Public
- Created
- Apr 8, 2024
- Updated
- Apr 8, 2024
- SHA
- 770e783f
Try a prompt
Base models
Model card
From the Hugging Face model README
<!-- 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. -->
codeparrot-ds
This model is a fine-tuned version of gpt2 on an unknown dataset. It achieves the following results on the evaluation set:
- eval_loss: 6.5630
- eval_runtime: 32.4403
- eval_samples_per_second: 432.826
- eval_steps_per_second: 13.533
- epoch: 0.6
- step: 30
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: 0.0005
- train_batch_size: 32
- eval_batch_size: 32
- seed: 42
- gradient_accumulation_steps: 8
- total_train_batch_size: 256
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_steps: 10
- num_epochs: 5
- mixed_precision_training: Native AMP
Framework versions
- Transformers 4.38.2
- Pytorch 2.2.1+cu121
- Datasets 2.18.0
- Tokenizers 0.15.2
- Task
- Text Generation
- Library
- transformers
- Type
- gpt2
- License
- mit
- Created
- Apr 8, 2024
- Updated
- Apr 8, 2024