3 ms·
We build software acceleration for LLM, effectively running smaller llama2 models at the same performance on several L4's as on 1xA100.
by pavelstoev 3y ago
We build software acceleration for LLM, effectively running smaller llama2 models at the same performance on several L4's as on 1xA100.