Starter kits

We've provided a few starter kits for you to modify and copy paste. Additional info about each kit is available in their respective folders

We've provided all historical Hacker Cup problems in a single dataset on HuggingFace: https://huggingface.co/datasets/hackercupai/hackercup

Using open models

1. sample_data_solver, which evaluates CodeLlama, or any other causal language model on the HuggingFace model hub, with the sample input-output pairs only passed in the prompt. No other information (including problem statement) is used.
2. finetuning, which finetunes a pretrained causal language model, e.g., CodeLlama, on the Hackercup dataset, treated as a language modeling task. Evaluate off-the-shelf or finetuned model by generating n code solutions for each Hackercup problem, choosing the best (according to sample test case correctness), and evaluating on full test case input-output pairs.

Using closed models

1. autogen which is a programming framework for agentic AI https://microsoft.github.io/autogen/
2. SWE-agent starter kit which solves leetcode style problems https://princeton-nlp.github.io/SWE-agent/usage/coding_challenges/
3. langchain starter kit which is a general framework to create LLM applications
4. submit_first_solution A starter kit to submit the first solution generated by an AI model. Simple code to run and evaluate the solution generated by the AI model.

While the Hacker Cup competition is proceeding, you will download all problems in a round in a single compressed folder and it's up to you to uncompress it and feed it to your model. Once your model produces the answer which should include both the generated code and an output.txt, you'll submit that code just like a human would on the hacker cup website.

As we're getting ready to announce winners, we'll be going in order from the top ranking winners to the bottom ranking ones

1. For the open track until we find enough reproducible answers
2. For the closed track until we find enough models that we can query via an API

HackerCup Lectures

There is a series of lectures on YouTube that explain how to use the starter kits and showcase state of the art ideas to tackle the HackerCup competition.

Date	Time	Topic	Speaker	YouTube Link
August 12th	5pm CET/ 8am PST	Submit your first solution	Thomas Capelle, Weights & Biases	Watch
August 13th	5pm CET/ 8am PST	AutoGen: Solving Complex Coding Challenges Using a Multi-Agent Framework	David Titsworth & Andrzej Banburski-Fahey, Microsoft	Watch
August 14th	7pm CET/ 10am PST	Reinforcement learning	Vincent Moens, Meta	Watch
August 15th	5pm CET/ 8am PST	SWEAgent	Kilian Lieret, SWEAgent	Watch
August 20th	5pm CET/ 8am PST	RAG	Bharat Ramanathan, Weights & Biases	Watch
August 21st	5pm CET/ 8am PST	FineTuning	Joe Cummings, TorchTune	Watch
August 27th	5pm CET/ 8am PST	DsPy	Krista Opsahl-Ong & Omar Khattab, DsPy	Watch

and more to come...

For the full list check here