New: PKC MARK Benchmark Tool (Beta)
Hey everyone,
I’m a Korean who’s not a developer,
but I somehow ended up building a local LLM benchmark tool.
The Situation
Started "coding" 3 months ago by telling AI what to do.
Built this benchmark tool because I needed it.
My personal version works perfectly. I use it daily.
But the public release needs:
- Final testing
- Documentation polish
- Some edge case fixes
- Actually clicking "publish"
And I just... can't. 🤷♂️
So I'm posting here hoping some validation will give me the energy to finish.
What It Does
Local LLM benchmark tool with all the things I wished existed:
- One-click setup (because I'm lazy)
- Real-time monitoring (VRAM, temp, power, TPS, TTFT)
- HuggingFace integration (search & download from UI)
- Model pipelines (serial/parallel/hybrid CPU+GPU)
- Supports both Transformers and GGUF
- GPU optimization wizard (for people like me who don't know CUDA)
Hardware:** RTX 2060 SUPER 8GB
How I built it:** Learned to command AI effectively
Time:"Estimated time remaining for the public release polish
(from personal version): about two more weeks, starting today."
Screenshots
The Irony
I built this tool to TEST language models for my main AI project.
Now the benchmark tool is more polished than the actual AI system. 💀
Why I'm Here
Things that are done:
- ✅ All features work (personal version in daily use)
- ✅ Cross-platform scripts (Win/Mac/Linux)
- ✅ Documentation (English + Korean)
- ✅ Single HTML file UI (no build process!)
- ✅ GPU auto-optimization
- ✅ It actually looks professional
- ✅ I'm still working on the cleanup/final polish. 😥
Things I need to do for public release:
- ⏳ Final testing pass
- ⏳ Clean up some code
- ⏳ Make GitHub repo
- ⏳ Write release notes
My motivation level: 📉📉📉
The Real Question
Is this even good?
I've been staring at it for so long I can't tell anymore.
Does anyone actually want this? Would it be useful?
Or should I just keep it private and use it myself?
**I genuinely need feedback to know if finishing this is worth the effort.**
Tech Details (for the curious)
- Backend: FastAPI + llama-cpp-python + transformers
- Frontend: Vanilla JS, single HTML file
- No build tools, no webpack, no npm hell
- Just download and run
- `OneClick_RUN.bat` is not a lie
The entire frontend is ONE HTML file. Is that crazy? Maybe. Does it work? Perfectly.
TL;DR:
non-developer built a benchmark tool using AI.
Personal version works great and I use it every day.
Now too burned out to polish it for public release.
**Posting here hoping someone says "omg publish this" so I get motivated.**
Is that pathetic? Yes. Will it work? Let's find out. 😅
P.S. If you're wondering "why doesn't he just finish it?" - Me too, buddy. Me too.
P.P.S. If this gets no response I'm blaming Reddit's algorithm and going back to my couch.
"I had an AI write this for me."