CubicLabs/Testing-Models-Datasets-Etc
0 Bytes
Email: cubiclaboratories@gmail.com
CPU-first training, open-source research, and practical deployment — no data center required.
> click for the full AXL detail page
Extrapolates from AXL's measured $0.004 per-model cost on consumer CPU.
Estimated CPU tokens/sec by parameter count (byte-level transformer, single-thread baseline).
File size from parameter count and quantization level.
KV-cache memory required for a given context window.
Total estimated RAM: weights at your chosen quantization plus KV-cache overhead.
Backs out training time from the $0.004/model anchor at a given effective hourly CPU rate.
Side-by-side comparison of two AXL model sizes.