🥳 Spark-X2.5-4B just reached #1 on Hugging Face Trending (last 7 days) — thank you to everyone downloading, testing, and starring it!
HER Hack-Astron #5 is still being judged — and to keep the momentum going, we're opening #6: the Spark-X2.5 Math Reasoning Challenge.
The challenge
Put Spark-X2.5's mathematical reasoning to a real, honest, reproducible test. Get Spark-X2.5-4B or 1.7B from Hugging Face (or the ModelScope mirror), run a real math-reasoning evaluation or case study, and publish the complete, reproducible result in the model's Hugging Face Discussions.
Pick a track (or combine several)
- Standard benchmarks — GSM8K, MATH / MATH-500, AIME 2024/2025, OlympiadBench, MinervaMath; report pass@1 (and pass@k) with a clear scoring script
- Competition & olympiad — AMC / AIME / IMO-style or national-olympiad problems; analyze where it succeeds or fails
- Step-by-step reasoning quality — catch right-answer-wrong-reasoning and arithmetic slips, not just final answers
- Tool-assisted math — pair with a code interpreter / Python / SymPy vs. a no-tool baseline
- Formal & proof-oriented — theorem proving, proof sketching, Lean/Coq-style formalization, honestly reported
- Multilingual math — e.g. MGSM, per-language accuracy
- Robustness & contamination checks — perturbed / paraphrased / number-swapped problems
- Decoding & prompting study — temperature, self-consistency, and more
How to enter
- Get Spark-X2.5-4B or 1.7B (Hugging Face or ModelScope mirror)
- Run a real math evaluation or case study
- Publish a complete, reproducible result in the model's HF Discussions
- Reply on the event GitHub Issue with your Discussion link
Schedule
- Submissions open: September 7, 2026
- Challenge period: September 7–13, 2026
- Deadline: September 13, 2026, 24:00 Beijing Time (UTC+8)
- Judging & awards: September 14–20, 2026
Links
- 🎯 Event / enter here (Issue #9): https://github.com/XHToken/Spark-X2.5/issues/9
- 🤗 Model collection: https://huggingface.co/collections/XHToken/spark-x25
- Spark-X2.5-4B: https://huggingface.co/XHToken/Spark-X2.5-4B · 1.7B: https://huggingface.co/XHToken/Spark-X2.5-1.7B
- 🇨🇳 ModelScope mirror: https://www.modelscope.cn/collections/XHToken/Spark-X25
Run it for real, document it honestly, and share what the next developer can reproduce.
Top comments (0)