3173 - 选择合适的LLM模型（自带设备办公）.pdf

1、Orlando,FLOctober 69IBM TechXchange 2025Session 3173Michael Roudnitski,Wojciech(Victor)Fulmyk,Rav AhujaSoftware Engineer,Data Scientist,Global Program Director&CCOIBM Skills NetworkStay agile:Adapting to rapid advances in LLMsAgenda010203Key Concept OverviewHands-on Workshop Q&AIBM TechXchange|2025

2、IBM CorporationWhat you will learn in this session0102030405LLM DifferencesProblem RequirementsEver-Changing LandscapeChallenges in AdaptingTesting with PromptfooIBM TechXchange|2025 IBM CorporationGranite?GPT?Claude?Llama?Gemini?Mistral?IBM TechXchange|2025 IBM Corporation4Choosing the right LLM is

3、 HardIBM TechXchange|2025 IBM Corporation5Choosing the right LLM is HardIBM TechXchange|2025 IBM Corporation6https:/lmarena.ai/leaderboardOctober 3,2025Choosing the right Golf Club is EasyIBM TechXchange|2025 IBM Corporation7Choosing the right Golf Club is EasyIBM TechXchange|2025 IBM Corporation8Ch

4、oosing the right Golf Club is EasyIBM TechXchange|2025 IBM Corporation9Selecting the most suitable Large Language Model(LLM)is challenging.IBM TechXchange|2025 IBM Corporation10Each model is uniqueSelecting the most suitable Large Language Model(LLM)is challenging.IBM TechXchange|2025 IBM Corporatio

5、n11Each model is uniqueNew Models are always coming out Selecting the most suitable Large Language Model(LLM)is challenging.IBM TechXchange|2025 IBM Corporation12Every Model is Different.How?IBM TechXchange|2025 IBM Corporation13Context WindowsAbilities and SpecializationsQuality and ConsistencyVend

6、or and Licensing TermsCostSpeed and LatencyContext WindowsRead this 1 million LoC codebase and suggest improvementsSummarize this 5-minute news articleIBM TechXchange|2025 IBM Corporation14Cost per Token or RequestDifferent models may require more tokens to produce the same output.Not just price per