Tagged: knowledge-distillation

4 articles

Latest

Megatron-Turing Costs $100M to Replicate. Who Is It Even For?

Segmind's open-source SD-Small and SD-Tiny slash Stable Diffusion size by up to 55%

A KL loss tweak lets a single GPU distill a 120B model without melting

Google's Gemma 2 drops: A 27B open model that beats Llama 3 70B