The Neural Deep Dive
The memory wall has finally met its match. This week, we tear down TurboQuant, the new two-stage compression breakthrough from Google and Tether that achieves a massive 5x reduction in KV cache footprints to squeeze 70B models onto consumer-grade GPUs. From the technical wizardry of random orthogonal rotations to the market-disrupting implications for NVIDIA and local inference, we’re exploring whether this is a true "democratization" of AI or just a strategic power play by Tether.
181 episoder
Kommentarer
0Vær den første til at kommentere
Tilmeld dig nu og bliv en del af The Neural Deep Dive-fællesskabet!