释放你内心的程序员：人工智能辅助编程.pdf

1、https:/dell.to/3N6gO6nWhat are we running onSimple powershellhow much free spece I have on diskopen me a folder where drivers usually are on Windowschange path to the Windows System32 drivers directory with drivers and then list all of them starting on letter s”and having a number in the nameThats e

2、asy but not super usefull,lets download real programmer toolsvscode+python and continue extenstionsBasic web developmentindex.html based websiteflask as web servergenerate a website with 2 tables with reportsOk,now lets create some data for those reportsjupyterhttps:/ it works under the hood NVIDIA

3、SLIDESHEREKey take aways coding assistants can really work in multiple use-cases this is based on lot of input tokens and small amount of output tokens small models 7B-13B work best as latency is a key here proper use of caching is important(you opperate on the same code base all the time,no point i

4、n prompting with it all the time)proper prompt beats model and HW speed in terms of result use LLMs also for learning,especially if you use ObsidianIntent detection agentArchitecture details for RAG in generalUnified GenAI Platform:RAG ArchitectureChat componentsChat serverChat UIUser auth serviceSe

5、ssion cacheUser authSession managerConversation loopFeedback loggerMessage queueMessage queue(single queue-split for visualization)Agent orchestratorSelect conversation agentsQuery processorIntent detection agentAI agent response(fusion&eval)multiple response agentsAgent orchestratorResponse evaluat

6、orResponse fusionGenAI model cloud model servingAgent orchestratorMessage queue(single queue-split for visualization)Message queue(single queue-split for visualization)Response evaluatorAI DaaS data storesGuardrailsIntent classificationEmbeddingsText generationHallucination modelTranslationSemantic