Telco Troubleshooting and Optimization Agentic Challenge
IMPORTANT: Please help us protect the integrity of this benchmark by not publicly sharing, re-uploading, or distributing the dataset.
Telco Troubleshooting Agentic Challenge focuses on network operations and maintenance. Participants are required to build intelligent agents that complete network fault diagnosis and troubleshooting tasks in the contest of wireless and IP networks. More specifically, the goal will be to design agent capabilities (simulations, memory, RAG, skills) and processes (fine-tuning, CoT) enabling them to deal with telecom network understanding and optimization. The entire competition uses Qwen3.5-35B-A3B as the base model. Participants may fine-tune the model but are not allowed to replace it with a different architecture or a model of a different parameter scale.
The competition is organized around three phases and 2 tracks:
The first phase (Phase 1) will start the 3rd of April and finish on the 4th of May. During this phase the participants will use the available data and software to design their agent. Participants can continuously submit the answers generated by their agent and monitor the public leaderboard.
The second phase (Phase 2) will run from the 4th of May till the 18th of May. In this phase, the participants will test their solution on new data (from the same distribution of that provided during phase 1). The top 30 solutions of each track on the public leaderboard will be selected for the final phase. These top teams will need to submit their agent for the final evaluation.
The third phase (Phase 3) will run from the 18th of May till the 29th of May, when the winners and the private leaderboard with the scores of the top-30 participants will be revealed.
The 2 tracks (Track A and Track B) focus on wireless and IP network troubleshooting problems, respectively.
To learn more, please refer to the README.md of each track.
- Downloads last month
- 3