Explorer / Pools
Model pools
Groups of machines that serve one model together: who is online, how much GPU memory they share and how fast they measured the model.
Pools11 open to join · 0 serving
Members online01 contributing
Shared VRAM12.8 GBdeclared, GPU backends · 12.8 GB all backends
Average speed1.10 tok/sper measuring member (1) · self-measured
1 pool(s)
| Pool | Model version | Status | Members | Online | Shared VRAM | Avg tok/s | Total tok/s | Uptime 24 h | Open seats |
|---|---|---|---|---|---|---|---|---|---|
| bonsai-2-community | d49769de9a…9f6c7f on chain | OPEN | 1 / 16 | 0 | 12.8 GB | 1.10 tok/s | 1.10 tok/s | 1.6% | 15 |
How to read these numbers
Pools are published by the PQWERAI pool directories, not by the chain. Memory is what each member declared for its backend (GPU backends count as VRAM); tokens per second are measured by each member on its own machine for this exact model and signed by its node key. Neither is proven. A split pipeline runs no faster than its slowest stage, so a pool total is an upper bound; the average is per member that reported a measurement.
Source: https://pools-testnet.pqwerai.com · read 0 s ago · JSON