跟讀練習: Accelerating generative AI training at scale with SageMaker HyperPod | Amazon Web Services - 通過影片學習英語口語

選擇練習模式

每種模式都會開啟全屏練習室。

15 句
如果句子過短或過長，請點選 Edit 進行調整。
1
Our approach to generative AI model training is really customer focused
我們訓練生成式AI模型的思路完全以客戶為中心。
2
and what you're really developing is the optimization that sort of almost grows the model.
而你真正在開發的是一種最佳化技術，這種技術幾乎可以讓模型實現自我成長。
3
It's very compute intensive and orchestrating and managing all that hardware is a very big task.
這需要極高的計算量，而協調和管理所有這些硬體是一項非常艱鉅的任務。
4
SageMaker HyperPod is like a supercomputer at your fingertips.
SageMaker HyperPod就像一臺觸手可及的超級計算機。
5
It essentially enables you to have very fast and very easy deployment
它本質上能讓你實現非常快速、便捷的部署。
6
of a training infrastructure that really lets you turn individual nodes into a fabric with really hundreds
這種訓練基礎設施真正能讓你將各個獨立節點整合成一個由數百個節點組成的網路架構
7
or even thousands of GPUs that really enable you to scale up your training to the largest workloads.
甚至數千個GPU，足以支援您將訓練擴充套件到最大規模的工作負載。
8
If you think about just the sheer magnitude of the infra
如果你只考慮基礎設施的龐大規模
9
and the DevOps networking expertise
以及DevOps網路領域的專業知識
10
that would be needed to otherwise kind of build this out based on just like a naked EC2 layer, that's enormous.
如果完全基於裸EC2層來搭建這套系統，所需的工作量會非常巨大。
11
So the fact that this now comes as like a managed service using Slurm, that's an AI researcher's dream.
如今它能以基於Slurm的託管服務形式提供，這簡直是AI研究人員的夢想。
12
HyperPod Recipes, it's a really awesome and right now unique training stack that we were really excited about at Salesforce.
HyperPod Recipes 是一套非常出色、目前獨一無二的訓練技術棧，我們在 Salesforce 時對它非常感興趣。
13
In particular, I'm really excited about the fact that you can get FSTP with LoRa,
尤其讓我感到興奮的是，你可以通過LoRa實現FSTP。
14
with Context Parallel, parallel all in one basically training run.
藉助上下文並行技術，基本上可以在一次訓練執行中完成所有並行操作。
15
Having that out of the box and ready to go really speeds up your iteration time.
開箱即可直接使用，確實能大幅加快迭代速度。

下載應用

流利口語所需的一切

AI 發音評分逐句打分
音標練習掌握每個音
詞彙學習建立你的詞庫
詞彙遊戲邊玩邊學
掃碼
掃碼
Download on theApp StoreGet it onGoogle Play

同頻道

SAP Concur migrates over 85,000 SQL Server databases to Amazon Web Services | Amazon Web ServicesC1
2

SAP Concur migrates over 85,000 SQL Server databases to Amazon Web Services | Amazon Web Services

Daily ConversationWorking Adults
Is prompt engineering dead? Advice on optimizing AI interactions | Amazon Web ServicesC1
3

Is prompt engineering dead? Advice on optimizing AI interactions | Amazon Web Services

Business EnglishUniversity StudentsWorking Adults
What's Next with AWS - Watch the highlights in 7 minutes | Amazon Web ServicesC1
2

What's Next with AWS - Watch the highlights in 7 minutes | Amazon Web Services

Business EnglishWorking Adults
Scaling Hiring at Amazon with Amazon Connect | Amazon Web ServicesB2

Scaling Hiring at Amazon with Amazon Connect | Amazon Web Services

Daily ConversationWorking Adults
FOR YOU

推薦影片

Expensive RTX 5090 for LLMs? NO. Use This Instead. (Tesla V100 SXM2 + Z8 G4, #RACERRRZ)C1
1
462 句

Expensive RTX 5090 for LLMs? NO. Use This Instead. (Tesla V100 SXM2 + Z8 G4, #RACERRRZ)

Daily ConversationUniversity StudentsWorking Adults
Amazon Athena explained in 120 seconds💡C1
15
29 句

Amazon Athena explained in 120 seconds💡

Daily ConversationUniversity StudentsWorking Adults
Why Coding Agents Keep Making Your Codebase WorseC1
1
853 句

Why Coding Agents Keep Making Your Codebase Worse

Daily ConversationWorking Adults
How AI is helping brands identify consumer trends fasterC1
2
74 句

How AI is helping brands identify consumer trends faster

Daily ConversationUniversity StudentsWorking Adults
Introducing GPT-Image-2.5 in the APIB2
4
12 句

Introducing GPT-Image-2.5 in the API

Daily ConversationWorking Adults
Learn 10X Faster In 10 MinutesC1
2
171 句

Learn 10X Faster In 10 Minutes

Daily ConversationUniversity StudentsWorking Adults
Data Ingestion Best Practices! How to Ingest Data Efficiently at Enterprise Scale!C1
1
133 句

Data Ingestion Best Practices! How to Ingest Data Efficiently at Enterprise Scale!

Daily ConversationWorking Adults
Agentic AI Frameworks Explained: Workflows, Multi-Agent, & ProductionC1
1
124 句

Agentic AI Frameworks Explained: Workflows, Multi-Agent, & Production

Daily ConversationUniversity StudentsWorking Adults
I Tried PewDiePie's Odysseus AI So You Don't Have ToC1
1
341 句

I Tried PewDiePie's Odysseus AI So You Don't Have To

Daily ConversationWorking Adults
Trust the Process: How to Design Teams That Actually WorkC1
1
346 句

Trust the Process: How to Design Teams That Actually Work

Daily ConversationUniversity StudentsWorking Adults
The Race To Power AI Is Reshaping America's Electric GridC1
1
137 句

The Race To Power AI Is Reshaping America's Electric Grid

Daily ConversationWorking Adults
Demystifying MCP | Architect InsightsC1
1
128 句

Demystifying MCP | Architect Insights

Daily ConversationWorking Adults

真實場景：生成式AI訓練的技術討論

這段影片對話圍繞亞馬遜雲服務的SageMaker HyperPod展開，探討如何高效擴充套件生成式AI模型訓練。對話中涉及技術術語、行業痛點及解決方案，是練習職場英語聽力與口語的絕佳素材。通過模仿此類專業對話，不僅能提升英語發音，還能掌握科技領域的表達邏輯。

實用短語與搭配

  • compute intensive：計算密集型（描述任務對計算資源的高需求）
  • orchestrating and managing：編排與管理（常用於技術場景中對資源的統籌）
  • scale up：擴充套件（指擴大業務或技術能力）
  • managed service：託管服務（雲端計算領域的常見術語）
  • iteration time：迭代時間（技術開發中最佳化效率的關鍵指標）

這些短語在科技類對話中高頻出現，熟練掌握能讓你的英語口語練習更貼近實際應用場景。

影子跟讀挑戰：現在就行動！

開啟影片，找到對話片段，跟著音訊進行英語影子跟讀。先聽一遍完整句子，注意語調與重音，再逐句模仿，力求發音與節奏一致。重複3-5次後，嘗試脫稿複述，檢查是否能準確使用上述短語。若想提升效果，可藉助shadowing site或shadowspeak工具，對比自己的發音與原音差異。只需10分鐘，你就能感受到口語流暢度的提升！

堅持影子跟讀，不僅能提高英語發音，還能培養語感，讓你在專業對話中更自信。現在就開始，享受快速進步的樂趣吧！

影片中的語法

說話人最常用的結構，並附上影片中的原話：

結構影片中的用法
被動語態 be + 過去分詞 — 強調發生了什麼，而不是誰做的would be needed

什麼是跟讀法？

跟讀法 (Shadowing) 是一種有科學依據的語言學習技巧，最初開發用於專業口譯員的培訓，並由多語言者Alexander Arguelles博士普及。這個方法簡單而強大：您在聽英語母語原聲的同時立即大聲重複——就像是一個延遲1-2秒緊跟說話者的影子。與被動聽力或語法練習不同，跟讀法強迫您的大腦和口腔肌肉同時處理並模模擬實的講話模式。研究表明它能顯著提高發音準確性，語調，節奏，連讀，聽力理解和口語流利度——使其成為雅思口語備考和真實英語交流最有效的方法之一。

影子跟讀法: 閱讀完整分步指南 →