| name | deep-learning-nlp |
| description | Paddle-based deep learning workflows from the course materials, including DNN/RNN text-style baselines and the CNN/LeNet image classification case using folder-labeled digit images. Use when a user asks Codex to train, evaluate, save, load, or predict with course DNN, CNN/LeNet, or RNN/GRU examples. |
Deep Learning NLP
Operating Rule
Use this skill as a fixed command-line workflow. Do not write custom Paddle training code unless the user explicitly asks for a new model architecture outside this skill.
- Decide whether the request is DNN, CNN/LeNet image classification, or RNN text-style modeling.
- If unsupported, state that this skill cannot solve it and name the missing model or task type.
- If supported, write a plan first. Include the model type, reference file, script, required data paths, training parameters, model output directory, and prediction output path when relevant.
- Ask for missing columns, image folders, model directory, or output path.
- Execute the referenced script with command-line arguments.
Capability Map
- DNN text classification: read
references/dnn.md, then call scripts/paddle_text_classifier.py --model-type dnn.
- CNN/LeNet image classification: read
references/cnn.md, then call scripts/paddle_image_cnn.py.
- RNN/GRU text classification: read
references/rnn.md, then call scripts/paddle_text_classifier.py --model-type rnn.
Input Expectations
- DNN/RNN text training input:
.csv or .xlsx with a text column and label column.
- CNN training input: image folders such as
data/CNN/train/0, data/CNN/train/1, where folder names are labels.
- CNN prediction input: one image file or a folder of labeled/unlabeled images.
Course Context
This skill follows the deep learning course notebooks:
- DNN and CNN concepts with Paddle.
- RNN/GRU concepts, training loop, evaluation, and save/load workflow.
The scripts adapt these course patterns to command-line workflows, so downstream AI only selects parameters and calls the CLI.
