Implement top-p filtering and enhance the generate method in UstaModel for improved token sampling. Adjust parameters for temperature, top_k, and top_p to refine output generation process.

Update execution counts in module_3_2.ipynb for consistent notebook flow and enhance model evaluation output with updated training metrics. Adjust model loading to include device specification for improved compatibility.

second version of the codes prepared

Implement automatic model download in app.py and update requirements.txt for package version compatibility. Replace local model path with a download method for u_model.pth and upgrade Gradio and Torch versions.

Update model path in app.py for consistency and adjust execution counts in demo.ipynb for accurate notebook flow. Revise readme.md to reflect course title change and expand course details, including learning outcomes and module descriptions.

Add Gradio interface for Usta Model, including model loading, chat functionality, and example prompts. Create README_HF.md for Hugging Face integration and update requirements.txt. Remove unused files and refactor tokenizer and model structure for improved performance.

Enhance module_3_2.ipynb by adding model evaluation code, updating training epochs to 1,000,000, and implementing model saving/loading functionality. Update readme.md to include a link for running the model in Colab. Refactor UstaModel to utilize UstaEmbedding for improved embedding management.

Update module_pytorch_train.ipynb to correct training data indexing, adjust model architecture by simplifying the neural network layers, and enhance output logging with updated accuracy and loss metrics for improved training feedback.

Update .gitignore to exclude 'data/' directory and enhance module_3_1.ipynb with additional code cells for model loading and tokenization examples, including execution count adjustments and output updates.

Refactor UstaModel to incorporate UstaDecoderBlock and update architecture to support multiple layers and linear output head

Refactor UstaCausalAttention to UstaMultiHeadAttention and update UstaModel to utilize multi-head attention mechanism with context length support

Add UstaCausalAttention class and update model to use causal attention mechanism

Refactor text_dataset.py to include DataLoader and remove tokenizer.py

module 0 tamamlandı

first commit