[{"data":1,"prerenderedAt":118},["ShallowReactive",2],{"glossary-en-llm-processing":3},{"id":4,"title":5,"body":6,"description":107,"extension":108,"meta":109,"navigation":113,"path":114,"seo":115,"stem":116,"__hash__":117},"en_glossary/en/glossary/llm-processing.md","LLM Data Processing: Tokens, Embeddings, Temperature, Hallucination",{"type":7,"value":8,"toc":96},"minimark",[9,14,18,23,26,30,33,48,52,55,69,73,76],[10,11,13],"h2",{"id":12},"data-processing-and-output-generation","Data Processing and Output Generation",[15,16,17],"p",{},"These terms are frequently encountered during the interaction between the user and the model.",[19,20,22],"h3",{"id":21},"tokens","Tokens",[15,24,25],{},"Models read text not word-by-word, but in small chunks called \"tokens\". Generally, 1000 tokens equal approximately 750 words. This is the unit of currency for LLM compute.",[19,27,29],{"id":28},"embeddings","Embeddings",[15,31,32],{},"The conversion of words or sentences into numerical vectors that a computer can understand. Semantically similar words are positioned close to each other in this vector space.",[34,35,36],"ul",{},[37,38,39,43,44,47],"li",{},[40,41,42],"strong",{},"Application",": Embeddings allow for \"Semantic Search\". Instead of keyword matching, you can find records based on meaning. This could be applied to search through historical alarm logs in a system like ",[40,45,46],{},"ZMA",".",[19,49,51],{"id":50},"temperature","Temperature",[15,53,54],{},"A setting that controls the creativity or randomness of the model's output.",[34,56,57,63],{},[37,58,59,62],{},[40,60,61],{},"Low (0.1)",": More consistent, logical, and deterministic. Better for code or technical data.",[37,64,65,68],{},[40,66,67],{},"High (0.8+)",": More creative and unexpected. Better for brainstorming.",[19,70,72],{"id":71},"hallucination","Hallucination",[15,74,75],{},"The situation where a model confidently fabricates information that is not true.",[34,77,78],{},[37,79,80,83,84,91,92,95],{},[40,81,82],{},"Warning",": In industrial settings, minimizing hallucination is critical. When interpreting sensor data from a ",[40,85,86],{},[87,88,90],"a",{"href":89},"/en/products/gdt-digital-transmitter","GDT Digital Transmitter",", an AI system must be strictly grounded (often using ",[40,93,94],{},"RAG",") to avoid reporting false faults.",{"title":97,"searchDepth":98,"depth":98,"links":99},"",2,[100],{"id":12,"depth":98,"text":13,"children":101},[102,104,105,106],{"id":21,"depth":103,"text":22},3,{"id":28,"depth":103,"text":29},{"id":50,"depth":103,"text":51},{"id":71,"depth":103,"text":72},"Key terms in how AI processes input and generates output: Tokens, Vector Embeddings, and managing Hallucinations.","md",{"tags":110},[111,22,29,51,72,112],"LLM","AI Processing",true,"/en/glossary/llm-processing",{"title":5,"description":107},"en/glossary/llm-processing","Q-XJh2xRkPWf-sPov571DDMXb7myd7nGVgJCVsL3FrI",1778229654816]