Tensor Craft is a collection of utilities and how-to for a typical AI/ML project.
The ImgProj stands for an Image Classification project with EfficientNet CNN Architecture:
- img_dataset_prep.ipynb prepares the dataset
- img_classifier.ipynb trains the model
- img_grad_cam.ipynb computes Gradient-weighted Class Activation Mapping to obtain a heat map visualization that highlights areas of an image that contribute to a neural network’s decision
- img_smoke_tester.ipynb provides a notebook to run a smoke test
The TxtProj stands for a Text Classification project with SVM and Tfidf dictionary:
- txt_dataset_prep.ipynb prepares the dataset
- txt_classifier.ipynb trains the model
- txt_smoke_tester.ipynb provides a notebook to run a smoke test
The MlpBertProj stands for a Text Classification project of Bert Embeddings by the Multi-Level Perceptron:
- mlpbert_dataset_prep.ipynb prepares the dataset
- mlpbert_classifier.ipynb trains the model
- mlpbert_smoke_tester.ipynb provides a notebook to run a smoke test
The XplainProj stands for a Shapley Values Explanation project based of the LGBMClassifier model:
- xplain_dataset_prep.ipynb prepares the dataset
- xplain_classifier.ipynb trains the model
- xplain_shap_values.ipynb provides an illustration on how to use Shapley Permutation and Kernel Shapley
- xplain_smoke_tester.ipynb provides a notebook to run a smoke test
The LoraProj stands for a LoRA fine-tuning of the BERT model:
- lora_dataset_prep.ipynb prepares the dataset
- lora_classifier.ipynb fine-tunes the model
The LlmAdapterProj stands for an Adapter-based fine-tuning of the BERT model:
- llmadapter_dataset_prep.ipynb prepares the dataset
- llmadapter_classifier.ipynb fine-tunes the model
ContrastiveBert illustrates contrastive learning of the BERT model, where "left" and "right" texts are embedded into the shared latent space:
- contrastivebert_classifier.ipynb tunes the model
- contrastivebert_indexer.ipynb HNSW Indexer
Tensorboard instructions
source ~/virtualenvs/tensor_craft/bin/activate
tensorboard --samples_per_plugin "images=100" --logdir ~/workspace/tensor_craft/tensorboard.run --bind_all --port 6006 --reuse_port True serve
# open browser at: http://localhost:6006