An LLM learns from a vast collection of web pages, books and code to predict the next word (token) in a text. Learned at enormous scale, that one skill lets it answer questions, summarise, translate and write code.
Open models such as Llama, Qwen, Mistral and Gemma can run on your own GPU, so your data stays with you and you can fine-tune the model for your needs.
Example
Qwen2.5 7B can power a Bangla chatbot that answers customers’ questions about a shop’s products.
বাংলায়: LLM (বড় ভাষা মডেল) — বিপুল লেখা পড়ে শেখা AI, যা প্রশ্নের উত্তর দেয়, লেখে, অনুবাদ করে — যেমন ChatGPT, Llama, Qwen।