Benchmark: MiniMind2-65M & Samsung A15 - "Infinity" Result.

[4-Vit]

Environment

• Device: Samsung Galaxy A15 (8GB RAM)
• Backend: MNN (CPU)
• State: in the process of optimizatiON.

Data: (MiniMind2-MNN)

• RAM: 356 MB
• Prefill: 228.4 t/s
• Decode: INFINITY t/s ± NaN
• Total Time: 1.686 s

Summary

NO_comment:🤷🏿‍♂️
4 Developers:
)Господа, "Queen_is_delighted":👍

[Juude]

what is the app's version

[4-Vit]

Sent from [Proton Mail](https://proton.me/mail/home) for Android.

…

-------- Original Message --------

On Friday, 04/03/26 at 07:44 Jinde.Song ***@***.***> wrote: Juude left a comment [(alibaba/MNN#4335)](#4335 (comment)) what is the app's version — Reply to this email directly, [view it on GitHub](#4335?email_source=notifications&email_token=CBAGCRYDRZ4QIODUJODSMJD4T46TXA5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTIMJYGE4DKNBQHA4KM4TFMFZW63VGMF2XI2DPOKSWK5TFNZ2KYZTPN52GK4S7MNWGSY3L#issuecomment-4181854088), or [unsubscribe](https://github.com/notifications/unsubscribe-auth/CBAGCRYRAFAGXUONUUEXODT4T46TXAVCNFSM6AAAAACXINOOW6VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHM2DCOBRHA2TIMBYHA). You are receiving this because you authored the thread.Message ID: ***@***.***>

[wangzhaode]

Thanks for the report. The "Infinity" decode speed likely means the generation completed in an extremely short time (possibly only 1 token), causing a divide-by-zero-like display.

Could you provide:

1. The full config.json content used for MiniMind2-MNN
2. The complete terminal output of llm_bench
3. Verify the model generated correct content (not empty/single token)

[wangzhaode]

感谢反馈。Infinity 结果通常是因为生成 token 极少导致的显示问题，非功能 bug。如有具体问题请提供完整日志后重新提交。

[4-Vit]

Я прекрасно понимаю -ЧТО- это обозначает | скорее всего - бюджетное "железо" не успевало считать | отчет "хотел" опубликовать "Gemini", а не я | я -НЕМНОГО- занят: воюю / это когда человечки, которых убивают другие человечки — особого желания заниматься всяким хламом у меня НЕТ. Sent from [Proton Mail](https://proton.me/mail/home) for Android.

…

-------- Original Message --------

On Wednesday, 04/15/26 at 09:41 王召德 ***@***.***> wrote: wangzhaode left a comment [(alibaba/MNN#4335)](#4335 (comment)) Thanks for the report. The "Infinity" decode speed likely means the generation completed in an extremely short time (possibly only 1 token), causing a divide-by-zero-like display. Could you provide: - The full config.json content used for MiniMind2-MNN - The complete terminal output of llm_bench - Verify the model generated correct content (not empty/single token) — Reply to this email directly, [view it on GitHub](#4335 (comment)), or [unsubscribe](https://github.com/notifications/unsubscribe-auth/CBAGCR4RNEXM6FWKTEPTRGT4V4VITAVCNFSM6AAAAACXINOOW6VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHM2DENBZHA2TGOBTHE). You are receiving this because you authored the thread.Message ID: ***@***.***>

[4-Vit]

* я -НЕ- писал, что считаю этот результат вашей ошибкой. Sent from [Proton Mail](https://proton.me/mail/home) for Android.

…

-------- Original Message --------

On Wednesday, 04/15/26 at 10:04 王召德 ***@***.***> wrote: wangzhaode left a comment [(alibaba/MNN#4335)](#4335 (comment)) 感谢反馈。Infinity 结果通常是因为生成 token 极少导致的显示问题，非功能 bug。如有具体问题请提供完整日志后重新提交。 — Reply to this email directly, [view it on GitHub](#4335 (comment)), or [unsubscribe](https://github.com/notifications/unsubscribe-auth/CBAGCRYLWHWUYSJKEZOCLZL4V4X57AVCNFSM6AAAAACXINOOW6VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHM2DENBZHE2TSOBZG4). You are receiving this because you authored the thread.Message ID: ***@***.***>