Using Local LLMs

The use of Lar­ge Lan­guage Models (LLMs) can be extre­me­ly hel­pful for many repe­ti­ti­ve busi­ness pro­ces­ses such as data inter­pre­ta­ti­on, wri­ting busi­ness texts, pro­gramming, or con­fi­gu­ring regu­lar expres­si­ons.

Pro­vi­ders on the inter­net, like Ope­nAI with ChatGPT and others, offer high­ly useful ser­vices for the­se pur­po­ses, but they are asso­cia­ted with poten­ti­al risks:

  • If you phra­se your requests poor­ly, it is pos­si­ble to inad­ver­t­ent­ly dis­c­lo­se cus­to­mer data.
  • In some cases, it may be unavo­ida­ble to include busi­ness infor­ma­ti­on in your queries.
  • The­se ser­vices are only available online.

I asked ChatGPT how to avo­id lea­king out sen­si­ti­ve data while still using LLMs. In respon­se, I recei­ved many useful tips and was poin­ted to the avai­la­bi­li­ty of local LLMs, among other things.

One recom­men­ded tool with easy usa­bi­li­ty is GPT4All, which can be instal­led on Win­dows, Mac, and Linux. (Ano­ther opti­on would be Oll­ama, which is more tar­ge­ted toward pro­gramm­ers.) GPT4All (and Oll­ama) run LLMs local­ly on your own com­pu­ter. This eli­mi­na­tes the risks to your own data.

Of cour­se, local models also come with dis­ad­van­ta­ges:

  • The best models are not available as they are retai­ned by the pro­vi­ders for their own ser­vices.
  • Run­ning the models is extre­me­ly memo­ry- and pro­ces­sor-inten­si­ve. Older com­pu­ters quick­ly reach their per­for­mance limits, and some­ti­mes you need to be very pati­ent befo­re recei­ving a respon­se.

After instal­ling GPT4All, you need to install a model (or optio­nal­ly seve­ral). GPT4All pro­vi­des a simp­le inter­face with a long list of available models.

Choo­sing the right one is not easy. But ChatGPT can help here again. You can ask which model is best sui­ted for your spe­ci­fic query, and you will recei­ve appro­pria­te recom­men­da­ti­ons. The models down­load and install quick­ly and are then available on your own com­pu­ter. After that, there’s not­hing stan­ding in the way of safe expe­ri­men­ta­ti­on.

Quick Tip: If you need help ope­ra­ting the tool, sim­ply ask ChatGPT or the local models them­sel­ves.

Categories
Latest News
Your contact

If you have any ques­ti­ons about our pro­ducts or need advice, plea­se do not hesi­ta­te to cont­act us direct­ly.

Tel.: +49 40 22 85 900-0
E-mail: info@statsoft.de

Gui­do Band­holz (Head of Sales)