HER Hack-Astron #5 is now open. This challenge invites developers to turn a real Spark X2.5 run into a reproducible, useful edge-agent case. Use Spark X2.5-4B or Spark X2.5-1.7B, publish the complete case in the matching Hugging Face Discussions, and reply to the event Issue with the direct link.
What can you build?
Choose one direction or combine several:
- Agent workflows with Codex, Claude Code, OpenClaw, Hermes, or another harness
- Real repository fixes, online-judge tasks, and end-to-end coding agents
- Native 1M-context document or codebase tasks
- Multilingual understanding, generation, and translation
- GGUF quantization, throughput, TTFT/TPOT, and edge memory use
- Instruction following, mathematics, reasoning, and safety boundaries
- Hardware adaptation across NVIDIA, Ascend, Hygon, HOUMO.AI, and others
- Licensed-data SFT/RL experiments with honest before-and-after comparisons
How to enter
- Run a real experiment and record the pinned model revision, runtime, hardware, precision/quantization, and parameters.
- Keep representative inputs, raw Spark X2.5 outputs, logs or screenshots, metrics, failures, and limitations.
- Publish the full case in the Discussions section for the model you used. The title only needs to include HER Hack-Astron #5.
- Reply to the event Issue with the direct Discussion link.
Discussion pages:
- 1.7B: https://huggingface.co/XHToken/Spark-X2.5-1.7B/discussions
- 4B: https://huggingface.co/XHToken/Spark-X2.5-4B/discussions
Awards and timeline
- Winner (1): RMB 500 and the title HER Hack-Astron #5 Champion
- Outstanding contribution awards: Astron merchandise, 5,000 Loomy points, or Luckin Coffee vouchers
- Submission deadline: September 6, 2026, 24:00 Beijing Time (UTC+8)
- Judging and awards: September 7–13, 2026
For team entries, list all contributors in the GitHub Issue reply. At least 50% of listed contributors must have profiles identifying them as women.
Read the full rules and join here:
