我想要一天分享一點「LLM從底層堆疊的技術」,並且每篇文章長度控制在三分鐘以內,讓大家不會壓力太大,但是又能夠每天成長一點。
知識庫(KB)的目標是讓組織控制發送給模型的信息流,以引導其朝正確方向發展,你可以使用任何格式來構建 KB,例如 SQL Server、其他數據庫、JSON 文件、CSV 等,我將以 Snap-ML 諮詢公司作為範例來展示一個樣本 KB。
assert1 = {'role': 'assistant', 'content': 'Opening hours of Snap-LM Consulting :Monday through Friday 9am to 5pm. Services :expert systems, rule-based systems, machine learning, deep learning, transformer models.'}
assert2 = {'role': 'assistant', 'content': 'Services :expert systems, rule-based systems, machine learning, deep learning, transformer models.'}
assert3 = {'role': 'assistant', 'content': 'Services :Fine-tuning OpenAI GPT-3 models, designing datasets, designing knowledge bases.'}
assertn = {'role': 'assistant', 'content': 'Services:advanced prompt engineering using a knowledge base and SEO keyword methods.'}
kbt = []
kbt.append(assert1)
kbt.append(assert2)
kbt.append(assert3)
kbt.append(assertn)



















