Great question — answering as the UphillTech team:
the NPU of the AI Max 395+ has 50 TOPS (overall 126 TOPS) the 3070 8GB has around 163 TOPS (326 TOPS). So you should be OK with the speed if you only need to translate 20 Pages once (and you can even run it overnight) (Page with rough speed for a most likely bigger model https://github.com/XiongjieDai/GPU-Benchmarks-on-LLM-Inference ) however mind the memory you need especially if you want to increase the context window. The OOM meany out of memory the biggest model in the table only starts working with 96 GB. If you don't have enough memory you can sap memory in and out but remember that is very slow. So maybe a online Model is a better solution. You should be able to automate all the LLM...
Full answer by joerg-de: https://ai.stackexchange.com/a/49081 (CC BY-SA 4.0).
Need it done for your business? UphillTech.
