NUC8i7HVK with Radeon RX Vega M GH: eGPU recommended for RAG?

Beginners
1

Hi,

I have a PC with a NUC8i7HVK (i7-8809G) and integrated Radeon RX Vega M GH. This GPU has 1536 shading units, 96 texture mapping units and 64 ROPs. AMD has provided the Radeon RX Vega M GH with 4 GB of HBM2 memory, which is connected via a 1024-bit memory interface. The GPU operates at a frequency of 1063 MHz, which can be ramped up to 1190 MHz, while the memory runs at 800 MHz.

I would like to read in various documents (RAG → PDF, TXT, DOCX) on a specific topic and then summarize them in a single text without redundancy, i.e. aggregate the information from several documents. Perhaps I would also like to do some ‘fine tuning’.

Which solution would you recommend?
Which language model is recommended for my requirements when starting out?

Does it make sense to switch to a Thunderbolt3 eGPU? I am also flirting with the ASUS ROG NUC 970, would it be sufficient to fulfil the above requirements?

Thanks and kind regards
huggingtestit

2

Apparently my question is too difficult. Can someone perhaps recommend another community where such a question might be answered?

Thanks in advance and kind regards
huggingtestit