🖥️ Creating your own HomeLAB for trading

mazibee wrote: Thu Apr 16, 2026 5:53 am Thanks to @Xard for sharing the latest setup i.e HomeLAB_v4.0

After two days of trial and error, I finally managed to get it working.
HomeLab10.jpg
HomeLab11.jpg


I ran into several issues along the way, probably because I’m running this on my Dell All-in-One with the following specs:

Processor: Intel® Core™ 7 150U (1.80 GHz)
RAM: 32 GB
Graphics:
NVIDIA GeForce MX570 A (2 GB)

Intel Integrated Graphics (128 MB)
Windows 11 Pro

I installed gemma4:e2b (about 7.2 GB)

Initially, I ran out of free usage on Antigravity, so I upgraded my plan, after that, the setup process became much smoother.

That said, I’m still not fully confident everything is working correctly and might still have many errors.

I suspect there are still some issues with the Genome/ Hebbian values, as they consistently stay close to 0.50000. I haven’t seen them go above 0.60000 or below 0.40000, which makes me think something might not be behaving as expected.

I should mention that I’m completely non technical, and I managed to set all of this up with the help of AI tools.

Zipped file shared by @Pelle also helped me a lot.

I’m sharing my current machine setup (zipped file) in case it helps others who want to build or improve their own HomeLab environment.

Hopefully @Xard and rest of the community can refine and improve it further by adding more features.

Best
Mazi

xard777 wrote: Thu Apr 16, 2026 10:52 am HomeLAB v4 Update: Hardware Limits & The Road to v5

My apologies to anyone having a tough time running HomeLAB v4 lately.

As this is an experimental project, I’ve found that running local models—especially alongside MT5—is incredibly resource-intensive. This performance bottleneck is the main reason development on v4 slowed down.
The Resource Challenge

While the Gemma4:26B MoE model is powerful, it carries a heavy 17GB payload. On a standard setup, this leaves only about 7GB of VRAM for the workflow, MT5, and the Linux OS, which simply isn't enough for stable performance.
The Hardware Solution: Dual-GPU Setup

To solve this, I’m adding a second GPU to my rig. I've opted for the Intel B70 32GB ECC VRAM, which is ideal for high-capacity tasks:

Intel B70 (32GB): Will handle the 17GB Gemma4 payload + 15GB workflow buffer.

AMD 7900XTX (24GB): Will be dedicated to MT5 and the Linux OS.

Looking Ahead: HomeLAB v5
Once the new hardware is integrated, work will begin on HomeLAB v5. This version will utilize a Docker stack with 9 containers, including FluxDB memory. It’s a massive architectural leap over v4, achieved through a relatively small hardware upgrade.
IMPORTANT: Workaround for Smaller GPUs

If you are running a card with lower VRAM, you can still get great results!

Change your Gemma4B local analysis to DeepSeek-V3 Cloud analysis.
You will see a massive performance boost, and the API costs are negligible.

Best,
Xard777

xard777 wrote: Thu Apr 16, 2026 1:56 pm HomeLAB v4.0 – Cloud Edition Update
I’ve now fully moved HomeLAB v4 to 100% cloud-based using DeepSeek-V3.
Why the change?

Dropped the heavy local Gemma4:26B model (17GB VRAM).
Makes the system accessible to users with low-VRAM GPUs (2GB / 6GB cards).
My RX 7900 XTX is now running very cool at ~45°C.
DeepSeek-V3 gives clean, structured reasoning at very low cost.

What’s new / improved:
THE_VOID short status logic added: When the market stays in strong equilibrium (Hebbian ≈ 0.5000 and coil power very low) for 3 consecutive cycles, the system now shows a brief “Market in equilibrium – monitoring only” message instead of running a full analysis. This reduces screen clutter and saves API tokens during quiet periods.
Removed all News & Intelligence Tickers: I never really used them and they were causing minor lag. Removing them has noticeably cleaned up the interface and improved overall smoothness on the 65" screen.

The system feels lighter, quieter, and more focused.
Updated zipped package with all changes is attached.
Feedback is welcome — especially from anyone who had trouble running the old local 26B model.
Best
Xard777

xard777 wrote: Thu Apr 16, 2026 1:56 pm HomeLAB v4.0 – Cloud Edition Update
I’ve now fully moved HomeLAB v4 to 100% cloud-based using DeepSeek-V3.
Why the change?

Dropped the heavy local Gemma4:26B model (17GB VRAM).
Makes the system accessible to users with low-VRAM GPUs (2GB / 6GB cards).
My RX 7900 XTX is now running very cool at ~45°C.
DeepSeek-V3 gives clean, structured reasoning at very low cost.

What’s new / improved:
THE_VOID short status logic added: When the market stays in strong equilibrium (Hebbian ≈ 0.5000 and coil power very low) for 3 consecutive cycles, the system now shows a brief “Market in equilibrium – monitoring only” message instead of running a full analysis. This reduces screen clutter and saves API tokens during quiet periods.
Removed all News & Intelligence Tickers: I never really used them and they were causing minor lag. Removing them has noticeably cleaned up the interface and improved overall smoothness on the 65" screen.

The system feels lighter, quieter, and more focused.
Updated zipped package with all changes is attached.
Feedback is welcome — especially from anyone who had trouble running the old local 26B model.
Best
Xard777

XXXX wrote: Thu Apr 16, 2026 10:36 am Is gemma4:e2b working locally ok as it requires a lot more ram than 2GB.