Weco: The AI Research Engineer

Weco systematically optimizes your code, guided directly by your evaluation metrics.

Example applications include:

• GPU Kernel Optimization: Reimplement PyTorch functions using CUDA, Triton or Metal, optimizing for latency, throughput, or memory_bandwidth.
• Model Development: Tune feature transformations or architectures, optimizing for validation_accuracy, AUC, or Sharpe Ratio.
• Prompt Engineering: Refine prompts for LLMs, optimizing for win_rate, relevance, or format_adherence

Overview

The weco CLI leverages a tree search approach guided by Large Language Models (LLMs) to iteratively explore and refine your code. It automatically applies changes, runs your evaluation script, parses the results, and proposes further improvements based on the specified goal.

Setup

1. Install the Package:

   pip install weco
   

2. Set Up LLM API Keys (Required):

   weco requires API keys for the Large Language Models (LLMs) it uses internally. You must provide these keys via environment variables:

   • OpenAI: export OPENAI_API_KEY="your_key_here"
   • Anthropic: export ANTHROPIC_API_KEY="your_key_here"
   • Google DeepMind: export GEMINI_API_KEY="your_key_here" (Google AI Studio has a free API usage quota. Create a key here to use weco for free.)

   The optimization process will fail if the necessary keys for the chosen model are not found in your environment.

3. Log In to Weco (Optional):

   To associate your optimization runs with your Weco account and view them on the Weco dashboard, you can log in. weco uses a device authentication flow:

   • When you first run weco run, you’ll be prompted if you want to log in or proceed anonymously.
   • If you choose to log in (by pressing l), you’ll be shown a URL and weco will attempt to open it in your default web browser.
   • You then authenticate in the browser. Once authenticated, the CLI will detect this and complete the login.
   • This saves a Weco-specific API key locally (typically at ~/.config/weco/credentials.json).

   If you choose to skip login (by pressing Enter or s), weco will still function using the environment variable LLM keys, but the run history will not be linked to a Weco account.

   To log out and remove your saved Weco API key, use the weco logout command.

Usage

The CLI has two main commands:

• weco run: Initiates the code optimization process.
• weco logout: Logs you out of your Weco account.

weco run Command

This command starts the optimization process.

Example: Optimizing Simple PyTorch Operations

This basic example shows how to optimize a simple PyTorch function for speedup.

For more advanced examples, including Metal/MLX, Triton, CUDA kernel optimization, and ML model optimization, please see the README.md files within the corresponding subdirectories under the examples/ folder.

# Navigate to the example directory
cd examples/hello-kernel-world

# Install dependencies
pip install torch

# Run Weco
weco run --source optimize.py \
     --eval-command "python evaluate.py --solution-path optimize.py --device cpu" \
     --metric speedup \
     --maximize true \
     --steps 15 \
     --model gemini-2.5-pro-exp-03-25 \
     --additional-instructions "Fuse operations in the forward method while ensuring the max float deviation remains small. Maintain the same format of the code."

Note: If you have an NVIDIA GPU, change the device in the --eval-command to cuda. If you are running this on Apple Silicon, set it to mps.

Arguments for weco run:

weco logout Command

This command logs you out by removing the locally stored Weco API key.

weco logout

Performance & Expectations

Weco, powered by the AIDE algorithm, optimizes code iteratively based on your evaluation results. Achieving significant improvements, especially on complex research-level tasks, often requires substantial exploration time.

The following plot from the independent Research Engineering Benchmark (RE-Bench) report shows the performance of AIDE (the algorithm behind Weco) on challenging ML research engineering tasks over different time budgets.

As shown, AIDE demonstrates strong performance gains over time, surpassing lower human expert percentiles within hours and continuing to improve. This highlights the potential of evaluation-driven optimization but also indicates that reaching high levels of performance comparable to human experts on difficult benchmarks can take considerable time (tens of hours in this specific benchmark, corresponding to many --steps in the Weco CLI). Factor this into your planning when setting the number of --steps for your optimization runs.

Important Note on Evaluation

The command specified by --eval-command is crucial. It’s responsible for executing the potentially modified code from --source and assessing its performance. This command MUST print the metric you specified with --metric along with its numerical value to the terminal (standard output or standard error). Weco reads this output to understand how well each code version performs and guide the optimization process.

For example, if you set --metric speedup, your evaluation script (eval.py in the examples) should output a line like:

speedup: 1.5

or

Final speedup value = 1.5

Weco will parse this output to extract the numerical value (1.5 in this case) associated with the metric name (‘speedup’).

Contributing

We welcome contributions! To get started:

1. Fork and Clone the Repository:

   git clone https://github.com/WecoAI/weco-cli.git
   cd weco-cli
   

2. Install Development Dependencies:

   pip install -e ".[dev]"
   

3. Create a Feature Branch:

   git checkout -b feature/your-feature-name
   

4. Make Your Changes: Ensure your code adheres to our style guidelines and includes relevant tests.

5. Commit and Push your changes, then open a pull request with a clear description of your enhancements.