Create accountLogin

Recent searches

No recent searches

Search options

Only available when logged in.
techhub.social is one of the many independent Mastodon servers you can use to participate in the fediverse.
A hub primarily for passionate technologists, but everyone is welcome

Administered by:

Server stats:

5.3K
active users

techhub.social: AboutStatusProfiles directoryPrivacy policy

Mastodon: AboutGet the appKeyboard shortcutsView source codev4.3.7

#benchmark

7 posts5 participants0 posts today
Continued thread
Public
Osma Suominen

Indeed, the CPU-only performance is even worse. The LocalScore on the tiny 1B model is only 16, with a text generation speed of 7.7 tokens/second.

localscore.ai/result/235

Let's see if I can run this on a Raspberry Pi for comparison...

www.localscore.aiLocalScore - Test #235 ResultsLocalScore benchmark results for test #235. This is for the accelerator Intel Core i7-8550U CPU @ 1.80GHz (skylake)
#LocalScore#llm#benchmark
Public
Osma Suominen

My hobby: running LocalScore.ai to benchmark how fast (ehm) my 2018 laptop runs a tiny 1B LLM. The laptop has a NVIDIA MX150 mobile GPU, 2GB VRAM. I guess it was intended for Photoshop filters or CAD stuff.

I got a LocalScore of 101 on the tiny model using the GPU (13.5 tokens/second for generation). A value of around 1000 is considered passable.

localscore.ai/accelerator/234

Still, I think it's marginally better than CPU-only on the same laptop.

www.localscore.aiNVIDIA GeForce MX150 ResultsLocalScore benchmark results for NVIDIA GeForce MX150 with 2GB of memory.
#llm#benchmark#LocalLlama
Public
Quinn Comendant 🐧

Proof or Bluff? Evaluating LLMs on 2025 USA Math Olympiad.

“[T]he first comprehensive evaluation of full-solution reasoning for challenging mathematical problems. Using expert human annotators, we evaluated several state-of-the-art reasoning models on the six problems from the 2025 USAMO within hours of their release. Our results reveal that all tested models struggled significantly, achieving less than 5% on average.”

arxiv.org/abs/2503.21934v1

Table 1: Main results of our evaluation. Problems are scored out of 7 points, with the total score being at most 42. Listed scores are averaged over all four runs. We measure cost in USD, and report the average score across all generations and graders for each problem. Summary of the results by model: - R1: Scored 2.0 points. Cost per run: $2.03. - Flash-Thinking: Scored 1.8 points. Cost not available. - Claude 3.7: Scored 1.5 points. Highest cost among low performers at $9.03. - QWQ: Scored 1.2 points. Cheapest model at $0.42. - O1-Pro: Also scored 1.2 points, but at a very high cost of $203.44. - O3-Mini: Lowest total score at 0.9 points. Cost: $1.11. Takeaway: All models performed poorly overall, scoring between 0.9 and 2.0 out of 42. There’s no clear trade-off between cost and performance; the most expensive model (O1-Pro) performed no better than the cheapest (QWQ).
#ai#llm#benchmark
Public
MasterTrend Info

 - CPU-Z 2.09 nueva versión del benchmark
 - #CPUZ209 #BenchmarkingRevolution #TechPerformance - #Software - #BenchMark #EvergreenContent #PCPerformance
 - 🌟 ¡Descubre CPU-Z 2.09! La nueva versión del popular benchmark trae mejoras en la detección de hardware y correcciones de errores, haciendo tu experiencia aún más precisa. 🖥️ Con soporte para los últimos procesadores Intel y AMD, podrás medir el rendimiento de tu...
mastertrend.info/cpu-z-2-0-nue

Public
VictoriaMetrics

#VictoriaLogs usually needs less RAM and disk space than comparable solutions, and it performs typical queries over logs at faster speed. See these #benchmark results.

📊 Learn in this blog post, provided by #VictoriaMetrics Co-founder Aliaksandr Valialkin, why VictoriaLogs is a great solution for #LogManagement.

itnext.io/why-victorialogs-is-

ITNEXT · Why VictoriaLogs is a better alternative to Grafana Loki?By Aliaksandr Valialkin
ExploreLive feeds
About