Br-T-GPT-1
This model is a "Fast-Transformer", A normal Text Generation Model. It can "talks" like ChatGPT, LLama, Falcon etc. Model is: "Open Source, Free Model"
Model Details
Model Description
Parameters:
Vocab Size: 67304
- Developed by: Bertug Gunel
- Model type: Decoder only Transformer
- Language(s) (NLP): TR
- License: CC-BY-NC-ND-4.0
Uses
Model can used with; .safetensors file Web GUI is coming soon!
Direct Use
Web GUI is coming soon!
Out-of-Scope Use
-Model can only use with turkish, because %95+ of dataset is Turkish! -Model generates bad (low quality sentences) answers!
Bias, Risks, and Limitations
Risks: May generates Political answers! No NSFW sentences used in training!
Recommendations
Risks: May generates Political answers! No NSFW sentences used in training!
How to Get Started with the Model
Please install .safetensors file first, code is coming soon!
Training Details
Training Data
Model trained on: %90+ Turkish (Türkce) data, but it's contains a lot of Japanese, Engilish and Arabic words, names, places etc.
Train details: Number of epochs: 1 Number of iterations: 2167 Training time: 3 Minutes 41 seconds (221 Seconds) Training devices: 1x T4 GPU (Google Collab)
Training Procedure
1x T-4 GPU used for 3 Minutes and 50+ Seconds (About 4 Minutes)
Training Hyperparameters
Training Parameters: Learning Rate: 1e-4 Epochs: 1 Batch Size: 16 (Pairs (QA)) Time: 3 Min(s) 41 Second(s)