Scholar Etude - Modernized minimal GPT implementation from scratch.

---

Scholar Dashboard

Just run python scholar/dashboard.py and open http://localhost:5002 in your browser.

---

Scholar Commands

Install dependencies

$ pip install -r requirements.txt

GPU device is also required.

Pre-training the model:

#  torchrun --nproc_per_node=1 scholar/main.py train

Use CUDA_VISIBLE_DEVICES=0 to specify the certain GPU device to use. Alternatively, you can use dashboard to start training. It's the recommended way.

Generating next few words:

$ python scholar/main.py predict "杨过的姑姑"

Inspecting model internals:

$ python scholar/main.py debug