CSE599O Fall 2025 Assignment 0: Warmup

Assignment Overview

This warmup assignment introduces the course workflow using a tokenization implementation based on OpenAI's tiktoken. Your tasks are to:

1. Test - Run the provided test suite (adapted from Stanford CS336)
2. Package - Follow the submission workflow to create a deliverable archive for Gradescope
3. Transfer - Transfer and Run on Remote Machine

This assignment introduces you to the course’s development environment, testing framework, and deployment pipeline.

Directory Structure

cse599o-assignment0/
├── README.md
├── pyproject.toml
├── Dockerfile
├── test_and_make_submission.sh
├── cse599o_warmup/
│   └── tokenizer.py
└── tests/
    ├── __init__.py
    └── test_tokenizer.py

Environment

We manage our environments with uv to ensure reproducibility, portability, and ease of use. Install uv here (recommended). Alternatively, you can install via command line:

pip install uv

or

brew install uv

We recommend reading about managing projects in uv here.

✅ 1. Run Tests Locally

Ensure your solution passes all tests:

uv run pytest

• uv run automatically creates a virtual environment and installs dependencies from pyproject.toml
• pytest -q runs the test suite quietly, failing fast on errors

Goal: We expect to see output similar to:

==20 failed, 3 passed, 1 xfailed in 4.65s ==

This indicates some tests are passing while others are failing, which is expected for the initial state of the assignment.

🔧 2. Implement the Missing Functionality

Your task: Implement the missing functionality in the tokenizer to make all tests pass. Focus on the decode method in cse599o_warmup/tokenizer.py - typically one line of code is sufficient.

Goal: After implementation, running uv run pytest should show:

== 23 passed, 1 xfailed in 4.60s ==

This indicates all core functionality is working correctly.

📦 3. Prepare Submission

Generate the submission archive:

bash ./test_and_make_submission.sh 

This command:

• Collects your code, configs, and metadata
• Creates a submission ready for submitting to Gradescope

Goal: cse599o-fall2025-assignment-0-submission.zip should appear in the repo root.

🌐 4. Transfer & Run on Remote Machine

We have two machines available for this course: "tomago.cs.washington.edu" and "tempura.cs.washington.edu". You can login to either using your UW NetID.

First, test SSH access to one of the course machines:

ssh UWNetID@tomago.cs.washington.edu
# or
ssh UWNetID@tempura.cs.washington.edu

Transfer your submission archive from your local machine to the remote server:

scp ./cse599o-fall2025-assignment-0-submission.zip UWNetID@tomago.cs.washington.edu:/homes/iws/UWNetID
# or
scp ./cse599o-fall2025-assignment-0-submission.zip UWNetID@tempura.cs.washington.edu:/homes/iws/UWNetID

On the remote machine, extract and test your submission:

unzip ./cse599o-fall2025-assignment-0-submission.zip -d assignment0
cd assignment0
uv run pytest

Goal: We expect to see output similar to:

== 23 passed, 1 xfailed in 4.60s ==

🔍 5. CUDA Visibility Test

On the remote machine, verify GPU access using your assigned device ID (provided by course staff):

CUDA_VISIBLE_DEVICES=YourDeviceID uv run cuda_visible_test.py 

Replace YourDeviceID with your assigned CUDA device ID from course staff.

Goal: Confirm GPU availability and proper device assignment.

🧠 Complete Workflow

uv run pytest -q
make submission
scp zipfile
ssh to remote
check GPU visibility

Upload cse599o-fall2025-assignment-0-submission.zip to "Assignment 0: Warmup" on Gradescope. This assignment does not count toward your grade and only ensures the workflow works for everyone’s account.