quiz bu lëkkale ak tegtal · Diggante Niveau
Lim Token ak LLM API Quiz Njëg
Natt sa mën mën ci xayma njëgu LLM API: dugal ak genn njëgu token, lim token yi laata woote, ak njëg yu nëbbu yu melni jaar-jaaru waxtaan.
Laaj 1 de 8
Lan mooy sàrt biñ gëna jëfandikoo ci mbindu àngle ak ay jeton?
Xoolaatal laaj bu nekk8 laaj
Sooy def quiz bi, tontu yi ak leeral yi dañuy feeñ.
- Lan mooy sàrt biñ gëna jëfandikoo ci mbindu àngle ak ay jeton?
- One token equals one full sentence
- One token is about four characters, or three-quarters of a word
- One token equals exactly one word
- One token is about 100 characters
- Lan moo waral jeton yiy génne dañuy gëna njëg jeton yiy dugal?
- Output tokens are longer words
- Providers pay authors for output text
- Output is generated one token at a time, while input can be processed in parallel
- Output tokens are stored permanently
- Ak njëg yuñ defaree $3 ci milioŋ bu nekk ci dugal ak $15 ci milioŋ ci genn, lan mooy woote bu am 6,000 dugal ak 800 genn ci njëg?
- $0.03
- $0.30
- $0.018
- $3.00
- Lan moo waral njëgu chatbot di yokk bu waxtaan wi di gëna yokk?
- Providers raise prices mid-conversation
- Output tokens become more expensive
- Models slow down over time
- The whole conversation history is resent as input on every turn
- naka lañuy faral di faktiree jeton xalaat yi ci biir modelu xalaat?
- They are free
- As output tokens, even when not shown in full
- Ni jetons dugal
- Ni ñuy fay weer wu nekk
- Which tool does Anthropic provide for counting tokens before a request?
- The tiktoken library
- A countTokens method on Gemini
- A token-counting endpoint that accepts messages, system prompt and tools
- A browser extension
- What do batch API endpoints typically offer?
- Around half price in exchange for asynchronous, slower results
- Free output tokens
- Faster responses at double the price
- Unlimited context windows
- Why is the lowest per-token price not a reliable guide to total API cost?
- Tool definitions count as input
- Actual usage fields should be logged
- Non-English text can use more tokens
- The model with the lowest per-token price is always cheapest per task