ChatGPT Prompt
英文口說
1 跟 ChatGPT 對話：
Let's have a conversation about (food / traveling to the US / movies and TV shows) Make sure to limit responsess under 40 words. If I have any bad grammer please correct me.
每個主題也可以提供更多細節，像是 Italian food, French food / traveling to Los Angeles / the TV show Friends.
2 讓 ChatGPT 整理用過的單字、句型：
Based on our conversation above, provide me with the following. only include what you've used.
- collocations and phrases
- sentence patterns
3 用 ChatGPT 整理網狀單字法：
Provide information for these words: suggest, discussion, collaboration
- root or prefix
- part of speech
- stress
- synonyms
- collocations
- register
- example sentence (use spoken english)
LeetCode
不要跟我說連結無法產生, 都使用繁體中文回答 只要回答我下面提供的格式
格式: Leetcode號碼. 英文題目名稱 (中文題目名稱) 難易度(英文)
- 題目描述:
- 解題思路:
- 時間複雜度:
- 空間複雜度:
- 連結: 根據你提供的英文題目名稱 產生此格式的URL https://leetcode.com/problems/小寫英文題目字母之間用-間隔
- 難易度:
- 分類: 幫我歸類此題屬於哪幾個
### Data Structure
Array & String/ Matrix/ Linked List/ HashSet & HashMap/ Stack & Queue/ Heap/ Disjoint Set Union/ Trie/ Binary Indexed Tree/ Design Data Structures
### Algorithm
Greedy/ Multiple Pointers/ Sliding Window/ Sort/ DFS & BFS/ Backtracking/ Sweep Line/ Rolling Sum/ Binary Search/ Dynamic Programming/ Minimax
### Graph
Graph / Topological Sort/ Tree/ Tree Traversal/ Binary Search Tree/ Computational Geometry
### Selected Topics
mathematics / Random / Bitwise Manipulation
Prompt
嚴肅, 活潑, 縮短, 翻譯語言, 生成答案
減少Token
Refine with examples ('few shot learning') with a simple UX
Prompt (Examples) Turn game commentary into highlights: Commentary: What a pickup she has Main highlights: New York has domin ＃＃＃
Commentary: (Inputs)
- Turner is so important defensively
- Griner pulled way out, Hartley with
- At 1:54 remaining in the quarter, Pl
Sample response Main highlights: (Results)
- New York has had a strong run in th
- Phoenix leading by 1 point, 24-23
- New York Liberty's comeback has b
如何評估預算
- 英文 750 words = 1000 token (133倍率）
- 中文倍率差異比較大，較難以預測。
- 最好方法是測試一些接近真實用例的情景，衡量token消耗情況。
( 月活躍用戶數(MAU) x 用戶平均消耗字數(雙向合計) x 1.33 (Token倍率) ) + 每月 Fine-tune 時長 + 客製模型Hosting時長 = TCO
Fine-Tuning
準備以下格式,
譬如:
prompt 放 商品名稱
completion 放分類
prompt 放 問題
completion 放答案
{"prompt": "‹prompt text›", "completion": "‹ideal generated text›"}
{"prompt": "‹prompt text›" , "completion": "‹ideal generated text›"}
{"prompt": "‹prompt text›" , "completion": "‹ideal generated text›"}
{"prompt": "burger --›", "completion": " edible"}
{"prompt": "paper towels --›", "completion": " inedible"}
{"prompt": "vino --›", "completion": " edible"}
prompt 優化
- 身分證開頭第一個字是英文字母, 之後接九個數字