- This hands-on notebook is designed to rapidly validate the fundamental capabilities of GPT-5 in Azure OpenAI while providing direct comparisons with GPT-4x models. It supports streamlined testing of the Chat Completions API, structured output, function calling, and the Responses API (covering Function Calling, Code, Vision, and Background Tasks) for both GPT-5 series and GPT-4x models.
s. Go to notebook
์ด ํธ์ฆ์จ ๋ ธํธ๋ถ์ Azure OpenAI์์ GPT-5์ ๊ธฐ๋ณธ ๊ธฐ๋ฅ์ ๋น ๋ฅด๊ฒ ๊ฒ์ฆํ๊ณ ๊ธฐ์กด gpt-4x๋ชจ๋ธ๊ณผ ๋น๊ตํ๋ ๋ ธํธ๋ถ์ ์ ๊ณตํฉ๋๋ค. Chat Completions API, structured output, function calling, Responses API (Function Calling, Code, Vision, Background Task ๋ฑ) ์ฌ์ฉ ์ gpt-5 series ๋ชจ๋ธ๊ณผ gpt-4x๋ชจ๋ธ์ ์ง์ ๋น๊ตํด์ผํ๋ ๊ฒฝ์ฐ ์์ฝ๊ฒ ํ ์คํธ ๊ฐ๋ฅํฉ๋๋ค. Go to notebook
- This hands-on demonstrates how to configure trace in AI foundry. It make you understand how to set up the right instruments for Azure Open AI, Azure Inference SDK to view the performance and optimize the LLM application.
s. Go to notebook
์ด ํธ์ฆ์จ์ AI Foundry์ trace ๊ธฐ๋ฅ์ ์ค์ ํ๋ ๋ฐฉ๋ฒ์ ๋ด๊ณ ์์ต๋๋ค. ์ด ํธ์ฆ์จ์ ํตํด ์ฌ๋ฌ๋ถ์ tracing ๊ธฐ๋ฅ์ ํ์ฑํํ๋ ๋ฐฉ๋ฒ๊ณผ ์ค์ ํ๋ ค๋ ์ฝ๋์ ๋ฐ๋ผ ์๋ง์ instrument๋ฅผ ์ค์ ํ๋ ๋ฐฉ๋ฒ๊ณผ ํ ์คํธ๋ฅผ ์งํํ ์ ์์ต๋๋ค. Go to notebook
- This hands-on demonstrates how to optimize prompts using PromptWizard. PromptWizard, released as open source and paper by Microsoft, is a prompt optimization tool for maximizing the performance of LLM. It is a prompt optimization framework that employs a self-evolving mechanism in which LLM generates, critiques, refines, and continuously improves prompts and examples through feedback and synthesis.
s. Go to notebook
์ด ํธ์ฆ์จ์ PromptWizard๋ฅผ ์ฌ์ฉํ์ฌ ํ๋กฌํํธ๋ฅผ ์ต์ ํํ๋ ๋ฐฉ๋ฒ์ ๋ณด์ฌ์ค๋๋ค. ๋ง์ดํฌ๋ก์ํํธ์ ์คํ ์์ค์ ๋ ผ๋ฌธ์ผ๋ก ๊ณต๊ฐํ PromptWizard๋ LLM์ ์ฑ๋ฅ์ ๊ทน๋ํํ๊ธฐ ์ํ ํ๋กฌํํธ ์ต์ ํ ๋๊ตฌ์ ๋๋ค. LLM์ด ์ค์ค๋ก ํ๋กฌํํธ์ ์์ ๋ฅผ ์์ฑ, ๋นํ, ์ ๊ตํ๊ณ ํผ๋๋ฐฑ๊ณผ ํฉ์ฑ์ ํตํด ์ง์์ ์ผ๋ก ๊ฐ์ ํ๋ ์๊ธฐ ์งํ ๋ฉ์ปค๋์ฆ์ ์ฑํํ ํ๋กฌํํธ ์ต์ ํ ํ๋ ์์ํฌ์ ๋๋ค. Go to notebook
- This hands-on workshop is tailored for engineers seeking to deepen their understanding of the Azure AI Evaluation SDK. Participants will explore the distinctions between evaluators and simulators through practical code examples. The workshop will guide you in assessing the quality and safety of your generative AI applications using industry-standard metrics. Leveraging Azure AI Evaluation SDKโs built-in evaluators, you will learn how to compare different versions of your applications and select the optimal solution to meet your specific requirements. Go to notebook
์ด ์ค์ต ์ํฌ์ต์ Azure AI ํ๊ฐ SDK๋ฅผ ์ดํดํ๊ณ ์ ํ๋ ์์ง๋์ด๋ฅผ ์ํ ๋ง์ถคํ ์ํฌ์ต์ ๋๋ค. ์ฐธ๊ฐ์๋ ์ค์ ์ฝ๋ ์์ ๋ฅผ ํตํด Evaluator์ Simulator์ ์ฐจ์ด์ ์ ์ดํด๋ด ๋๋ค. ์ด ์ํฌ์ต์์๋ ์ ๊ณ ํ์ค ๋ฉํธ๋ฆญ์ ์ฌ์ฉํ์ฌ ์์ฑ AI ์ ํ๋ฆฌ์ผ์ด์ ์ ํ์ง๊ณผ ์์ ์ฑ์ ํ๊ฐํ๋ ๋ฐฉ๋ฒ์ ์๋ดํฉ๋๋ค. Azure AI ํ๊ฐ SDK์ ๊ธฐ๋ณธ ์ ๊ณต ํ๊ฐ๊ธฐ๋ฅผ ํ์ฉํ์ฌ ๋ค์ํ ๋ฒ์ ์ ์ ํ๋ฆฌ์ผ์ด์ ์ ๋น๊ตํ๊ณ ํน์ ์๊ตฌ ์ฌํญ์ ์ถฉ์กฑํ๋ ์ต์ ์ ์๋ฃจ์ ์ ์ ํํ๋ ๋ฐฉ๋ฒ์ ๋ฐฐ์๋๋ค. Go to notebook
- Test the most basic way to use the o1(GA) with Vision model, Structured Output and gradio sample application as your playgournd Go to notebook
๋น์ ๋ชจ๋ธ, ๊ตฌ์กฐํ๋ ์ถ๋ ฅ ๋ฐ ๊ทธ๋ผ๋์ค ์ํ ์ ํ๋ฆฌ์ผ์ด์ ์ ํ๋ ์ด๊ทธ๋ผ์ด๋๋ก ์ฌ์ฉํ์ฌ o1(GA)๋ฅผ ์ฌ์ฉํ๋ ๊ฐ์ฅ ๊ธฐ๋ณธ์ ์ธ ๋ฐฉ๋ฒ์ ํ ์คํธํด ๋ณด์ธ์. Go to notebook
- Added Audio Data Augmentation using Audiomentations. Audiomentations supports both mono and stereo audio and integrates seamlessly with common audio processing workflows. It's lightweight, efficient, and helps simulate real-world audio conditions for better generalization in models.
Please do not forget to install the audiomentations package. Install with pip install audiomentations or see requirements.txt.
- Refactored to make it easier to test custom models for a given language by adding language-specific settings. Added a function to the 3_evaluate_custom_model notebook to retrieve detailed WER information from the notebook based on whether there are insertions, substitutions, or deletions. Go to notebook
์ธ์ด๋ณ ์ค์ ์ ์ถ๊ฐํ๋ฉด ๊ฐ๋จํ ํด๋น ์ธ์ด์ ๋ง๋ ์ปค์คํ ๋ชจ๋ธ์ ํ ์คํธํด๋ณผ ์ ์๋๋ก ๋ฆฌํํ ๋งํ์ต๋๋ค. insertion, substitution, deletion ์ฌ๋ถ์ ๋ฐ๋ผ ์์ธํ WER์ ๋ณด๋ฅผ ๋ ธํธ๋ถ์์ ์กฐํํ๋ ํจ์๋ฅผ 3_evaluate_custom_model ๋ ธํธ๋ถ์ ์ถ๊ฐํ์ต๋๋ค. Go to notebook
- Azure AI Speech is a managed service that provides speech capabilities such as speech-to-text, text-to-speech, voice translation, and speaker recognition. In this lab, you will learn the entire end-to-end process of training a custom speech-to-text (STT) model optimised for a specific language and use case based on synthetic data. You can practice generating synthetic text data (phi3.5), converting generated text files to audio files (text-to-speech), training(speech-to-text), evaluating, and deploying custom AI speech models based on synthetic text/audio files. In addition to generating synthetic data, you can also upload the speech data you use in the field to a specific folder and upload it to the storage account with simple notebook code to proceed with dataset creation, training, and evaluation. If you're looking to train custom speech models with different types of datasets to improve your word error rate (WER), this Python SDK and REST API-based handson makes it easy to automate your end-to-end model training and evaluation pipeline and scale your transformations.
Go to notebookAzure AI Speech๋ ์์ฑ ํ ์คํธ ๋ณํ, ํ ์คํธ ์์ฑ ๋ณํ, ์์ฑ ๋ฒ์ญ, ํ์ ์ธ์๊ณผ ๊ฐ์ ์์ฑ ๊ธฐ๋ฅ์ ์ ๊ณตํ๋ ๊ด๋ฆฌํ ์๋น์ค์ ๋๋ค. ๋ณธ ํธ์ฆ์จ์์๋ ํน์ ์ธ์ด์ ์ ์ค์ผ์ด์ค์ ์ต์ ํ๋ Custom STT(Speech To Text)๋ชจ๋ธ ํ์ต์ End2End ์ ์ฒด๊ณผ์ ์ ํฉ์ฑ๋ฐ์ดํฐ(Syntethic data)๊ธฐ๋ฐ์ผ๋ก ์ค์ตํฉ๋๋ค. ํฉ์ฑ ํ ์คํธ ๋ฐ์ดํฐ ์์ฑ(phi3.5), ์์ฑ๋ ํ ์คํธํ์ผ์ ์ค๋์คํ์ผ๋ก ๋ณํ (Text to Speech), ํฉ์ฑ ํ ์คํธ/์ค๋์คํ์ผ ๊ธฐ๋ฐ์ Custom AI Speech ๋ชจ๋ธ ํ์ต(Speech to Text), ํ๊ฐ, ๋ฐฐํฌ๋ฅผ Python SDK์ REST API๊ธฐ๋ฐ์ผ๋ก ์ค์ตํด๋ณผ ์ ์์ต๋๋ค. ํฉ์ฑ๋ฐ์ดํฐ๋ฅผ ์์ฑํ๋ ๊ฒ ์ธ์๋ ํ์ฅ์์ ํ์ฉํ๊ณ ์๋ ์์ฑ๋ฐ์ดํฐ๋ฅผ ํน์ ํด๋์ ์ ๋ก๋ํ๋ฉด ๊ฐ๋จํ ๋ ธํธ๋ถ ์ฝ๋๋ก Storage Account์ ์ ๋ก๋ ๋ฐ ๋ฐ์ดํฐ์ ์์ฑ, ํ์ต ๋ฐ ํ๊ฐ ๊ณผ์ ์ ์งํํด๋ณผ ์๋ ์์ต๋๋ค. WER(๋จ์ด ์ค๋ฅ์จ)์ ๊ฐ์ ํ๊ธฐ ์ํด ๋ค์ํ ์ ํ์ ๋ฐ์ดํฐ์ ์ผ๋ก ๋ง์ถคํ ์์ฑ ๋ชจ๋ธ์ ํ์ต์ํค๋ ค๋ ๊ฒฝ์ฐ, Python SDK ๋ฐ REST API๊ธฐ๋ฐ์ ๋ณธ ํธ์ฆ์จ์ ํ์ฉํ์ฌ ์๋ํฌ์๋ ๋ชจ๋ธ ํ์ต ๋ฐ ํ๊ฐ ํ์ดํ๋ผ์ธ์ ์ฝ๊ฒ ์๋ํํ๊ณ ๋ณํ์ ํ์ฅํ ์ ์์ต๋๋ค. Go to notebook
- This hands-on workshop is designed to help engineers who have difficulty developing UI-based Promptflow in Azure ML Studio, AI Studio, and VS Code. Based on the Python Promptflow SDK, you will learn how to develop and run chat, flows with context, phi3 model integration deployed serverlessly, evaluation flows, and filter inappropriate prompts using content safety. Go to notebook
์ด ํธ์ฆ์จ ์ํฌ์ต์ Azure ML Studio, AI Studio, VS Code์์ Promptflow๋ฅผ ui๊ธฐ๋ฐ์ผ๋ก ๊ฐ๋ฐํ๋๋ฐ ์ด๋ ค์์ ๋๋ผ๋ ์์ง๋์ด๋ค์ ์ง์ํ๊ณ ์ ๊ฐ๋ฐ๋์์ต๋๋ค. Python Promptflow SDK๋ฅผ ๊ธฐ๋ฐ์ผ๋ก chat, context๊ฐ ํฌํจ๋ flow, serverless๋ก ๋ฐฐํฌํ phi3๋ชจ๋ธ ์ฐ๋, evaluation flow ๊ฐ๋ฐ ๋ฐ ์คํ ๋ฐฉ๋ฒ๊ณผ ๋ถ์ ์ ํ prompt๋ฅผ content safety์ ํ์ฉํด filteringํ๋ ์ค์ต์ ๋ด๊ณ ์์ต๋๋ค. Go to notebook
- Based on the current version of the Azure AI Search Python client library, azure-search-documents==11.6.0b4. This code sample demonstrates various code patterns to implement AI search using Azure OpenAI and Azure AI Search to create indexes, vector search, change search algorithms, cross-field search, multi-vector search, filtering, hybrid, and reranking. Go to notebook
ํ์ฌ Azure AI Search Python client library ์ต์ ๋ฒ์ ์ธ azure-search-documents==11.6.0b4๋ฅผ ๋ฐํ์ผ๋ก ์์ฑ๋์์ต๋๋ค. ์ด ์ฝ๋ ์ํ์ Azure OpenAI์ Azure AI Search๋ฅผ ์ฌ์ฉํ์ฌ ์ธ๋ฑ์ค ์์ฑ, vector search, ๊ฒ์ ์๊ณ ๋ฆฌ์ฆ ๋ณ๊ฒฝ, Cross-Field๊ฒ์, Multi-Vector๊ฒ์, filtering, Hybrid, Reranking์ ํ์ฉํ AI ๊ฒ์์ ๊ตฌํํ๋ ๋ค์ํ ์ฝ๋ ํจํด์ ๋ณด์ฌ์ค๋๋ค. Go to notebook
Date of creation: 15-Oct-2024
Updated: 06-Apr-2025
Hyo Choi | hyo.choi@microsoft.com | https://www.linkedin.com/in/hyogrin/
Daekeun Kim | daekeun.kim@microsoft.com | https://www.linkedin.com/in/daekeun-kim/