2023-09-16 08:00:00
We explain how to run your own ChatGPT on your own computer, from the most accessible method to the most advanced method.
❤️ Patreon
???? Discord
???? techcafe.fr
Electronic components and materials, semiconductors: discover all the offers from our partner Farnell France on fr.farnell.com
(Advertisement)
Baby level
A chatbot on your PC Why? Science ! Innovation! But above all, fun. LLM that: What is a language model? Section hands : tokens, context, vectors and samplers. Easy level: software that does everything (or almost) Faraday.dev et LM Studio. Equipment: do I have all the equipment to take care of my LLama?
Toddler level
GGML et GPTQ : JPEG for AI? It’s unclear… What to choose? A 34b in Q3_K_S or a 13b in Q4_K_M? Good question ! Average level : KoboldCpp et embarrassment of the choice of AI… The samplers in detail: top_ktop_p, tfs, mirostat and we’re moving on (don’t worry).
Bearded level
Difficult level: Oobabogastraps and plugins, callcpp and its grammars. Sampling speculative : the art of saving time by working for nothing.
PrivateGPT, and why shouldn’t I chat with my PDFs? Space bar odyssey: with Open-interpreter, put Hal 9000 in your PC. Llama, licences not very poetic, there are also the MPT, Openllama, Falcon, etc.
Bonus links
Some links to keep up to date: The trusted “chatGPT” models: llama2chat, Vicuna1.5, NousHermes-llama2. The “fine-tuned” model in French: Vigogne-2. 18+ and waifu friendly models: Mythomax-L2, Pygmalion2. Models that code: codellama-instruct, wizardlm-codellama. Available in the size and flavor of your choice thanks to TheBloke (Tim Jobbins). And to test the BIG Llama and the VERY BIG Falcon free…
Participants
A program prepared by Guillaume Poggiaspalla Presented by Guillaume Vendé
Become a VIP listener
Guarantee long-term distribution and independent content thanks to the virtuous model of crowdfunding with Patreon and enjoy a private stream without ads.
1694892130
#LLamasté #ChatGPT #Tech #Café
