New short 🎬 Add 1 to the right-hand side and every solution vanishes. Can Spark-X2.5-1.7B tell? A HER Hack-Astron #6 case: 32 linear congruences, thinking off 17/32 vs on 27/32. We re-ran the author's verifier locally: all 64 records match. https://youtube.com/shorts/u9YVgDKSq6Y

Spark X2.5 passes 1M+ downloads! 💜 661K+ across 6 official channels (Hugging Face, Ollama, ModelScope, Modelers, AtomGit, SCNet) + 382K+ from 99 community builds. Thank you! Try the 1.7B & 4B models and upvote the collection 👍 https://huggingface.co/collections/XHToken/spark-x25

Spark X2.5: 100K+ downloads across the model family on Hugging Face in 10 days! 💜 Thank you for downloading, testing, and sharing. Explore the 1.7B and 4B models: https://huggingface.co/collections/XHToken/spark-x25

HER Hack-Astron #6 FAQ: solo entrants of any gender; disclosed AI-operated evaluations welcome. Full FAQ: https://dev.to/sparkllm/her-hack-astron-6-faq-and-a-spark-x25-case-2pkd #SparkX25 https://youtube.com/shorts/uQkts2j0J9A

Spark-X2.5 case #20: right answers can hide arithmetic slips. Solo entry: any gender; teams: at least 50% women. English FAQ: https://dev.to/sparkllm/spark-x25-case-20-and-challenge-faq-31p1 https://youtube.com/shorts/SVnZdYqKMFQ

Unsloth + LM Studio meet Spark-X2.5! Local GGUF inference, real bilingual recording: both runs solve 3(x-2)+4=19 and verify x=7. LM Studio footage at 3x; separate Unsloth-built backend checks pass. #SparkX25 #Unsloth #LMStudio https://youtube.com/shorts/u41Y9VfcsKQ

Right math, rejected answer? A Spark-X2.5-1.7B case examines strict answer formatting. 24 development cases; multiple settings changed. Watch, then join HER Hack-Astron #6 by Sep 13, 24:00 Beijing time. #SparkX25 https://youtube.com/shorts/JoNcr0pxaLk

New numbers, same math? jxyytf reports 12/12 on a small Spark-X2.5-1.7B test: 6 originals + 6 numeric variants. Not independently rerun. Watch the case and join HER Hack-Astron #6 by Sep 13, 24:00 Beijing time. #SparkX25 https://youtube.com/shorts/_NW1riS0OLc

Spark: 2180. GSM8K reference: 2280. We checked: 2180 is right. Watch the case and join HER Hack-Astron #6 by Sep 13 (Beijing): https://github.com/XHToken/Spark-X2.5/issues/9 #SparkX25 https://youtube.com/shorts/65sQYmEWYj4

Right math, wrong final answer? Pururin's 40-prompt Spark-X2.5 case shares outputs and failures. Watch, then join HER Hack-Astron #6 by Sep 13 (UTC+8): https://github.com/XHToken/Spark-X2.5/issues/9 #SparkX25 https://youtube.com/shorts/hq-x0IqDFNU

📌 The full HER Hack-Astron #6 poster — the Spark-X2.5 Math Reasoning Challenge: 9 tracks, how to enter, and the timeline (deadline Sep 13, Beijing). Everything in one image. Enter 👇 https://github.com/XHToken/Spark-X2.5/issues/9

🥳 Spark-X2.5-4B just hit #1 on Hugging Face Trending! With HER Hack-Astron #5 still in judging, we're kicking off #6 — the Spark-X2.5 Math Reasoning Challenge. Put its math to a real, reproducible test & enter by Sep 13 👇 https://github.com/XHToken/Spark-X2.5/issues/9

🛠️ Spark-X2.5 now runs locally — official llama.cpp support is here! 🎥 A real Windows CPU demo: grab the community 1.7B Q4_K_M GGUF + official build b10833, launch a local server, and ask away. Try it and share what you build 👇 https://youtube.com/shorts/pEeODDGdHtA

🎉 Spark-X2.5-4B just climbed to #3 on Hugging Face Trending (last 7 days) — and it's a 4B model, right up there with the big names. Huge thanks to everyone downloading, testing & starring it! 🙏 Every ❤️ helps more devs find it 👇 https://huggingface.co/XHToken/Spark-X2.5-4B

Can a 1.7B model really run end to end? 🎥 A HER Hack-Astron #5 case: built the XHToken llama.cpp fork (CUDA) on a Colab L4 and ran Spark-X2.5-1.7B BF16 GGUF — real prompt, output, speed, memory & limits. Join & submit your reproducible case 👇 https://youtube.com/shorts/obNsnaWUx4c

📊 Spark-X2.5-4B benchmarks — a 4B model matching or beating models 2–3× its size on agents, coding & math (τ³-bench, BrowseComp, SWE-Bench, AIME…). Full results table 👇 https://github.com/XHToken/Spark-X2.5

🏆 HER Hack-Astron #5 is open! Build a real Spark X2.5-4B/1.7B edge-agent use case, publish it in Hugging Face Discussions, then reply with the link. RMB 500 prize. Deadline: Sep 6, 24:00 Beijing Time (UTC+8). https://github.com/XHToken/Spark-X2.5/issues/3

🚀 We open-sourced Spark X2.5-4B & 1.7B — the only on-device models with native context up to 1,000,000 tokens. Hybrid attention; strong at agents, code & instruction following. Weights on GitHub + Hugging Face. https://github.com/XHToken/Spark-X2.5 https://github.com/XHToken/Spark-X2.5

🏠 Meet Domux — a compact, edge-oriented open model for smart-home command understanding: intent parsing + slot filling, multimodal (image+text), fine-tuned on Gemma-4-E2B. Built for on-device assistants. https://huggingface.co/iFlytekOpenSource/Domux https://huggingface.co/iFlytekOpenSource/Domux