I am currently using the calendly app to allow employees to sign up for group meetings with various managers(staff in booking).  I would like to stop using this paid app and use something that is already integrated in my outlook calendar.

I need the bookings app to have the functionality for up to 6 employees (customers) to self-sign up for pre-set event times, given only a common link – not an individual invite.

Is this in the works?

With this in place, I believe the bookings app can then send reminders, allow customers to cancel or reschedule, allow managers/staff to cancel or reschedule.

​I am currently using the calendly app to allow employees to sign up for group meetings with various managers(staff in booking).  I would like to stop using this paid app and use something that is already integrated in my outlook calendar.I need the bookings app to have the functionality for up to 6 employees (customers) to self-sign up for pre-set event times, given only a common link – not an individual invite. Is this in the works? With this in place, I believe the bookings app can then send reminders, allow customers to cancel or reschedule, allow managers/staff to cancel or reschedule.   Read More

​

Hi all,

I have an external hard drive formatted by NTFS. I bought a new Mac recently but writing to NTFS is not supported by Mac. FAT32 is supported by both Windows 11 PC and Mac. Is it possible to convert NTFS to FAT32 without losing data on Windows 11/10 PC ?

It is a 4T drive and formatting is not the choice. I can’t afford to lose so much data. Please kindly make a suggestion if you know how to do this.

Thanks

​Hi all, I have an external hard drive formatted by NTFS. I bought a new Mac recently but writing to NTFS is not supported by Mac. FAT32 is supported by both Windows 11 PC and Mac. Is it possible to convert NTFS to FAT32 without losing data on Windows 11/10 PC ? It is a 4T drive and formatting is not the choice. I can’t afford to lose so much data. Please kindly make a suggestion if you know how to do this. Thanks  Read More

​

Is it possible  for all the custom fields from the booking to be included in the ‘New Booking’ confirmation email that is sent?

All this info is included on the email that the customer gets, so unsure why it’s not included in the confirmation email too.

​Is it possible  for all the custom fields from the booking to be included in the ‘New Booking’ confirmation email that is sent? All this info is included on the email that the customer gets, so unsure why it’s not included in the confirmation email too.   Read More

​

Introduction

Natural Language Processing (NLP)  has revolutionized how we interact with technology, enabling us to communicate with machines more human-likely. Language models, a critical component of NLP, have become increasingly sophisticated, allowing us to build applications that can understand, generate, and process human language. However, as language models become more complex, it’s essential to ensure that they are performing optimally and reliably.

Evaluating language models is a crucial step in achieving this goal. By assessing the performance of language models, we can identify areas of improvement, optimize their performance, and ensure that they are reliable and accurate. However, evaluating language models can be a challenging task, requiring significant expertise and resources.

That’s where Azure AI Studio comes in. Azure AI Studio is a comprehensive platform that provides a range of tools and features for building, deploying, and managing machine learning models, including language models. With Azure AI Studio, you can evaluate language models in an efficient and scalable way, using a range of evaluation metrics and datasets.

In this blog post, we’ll take a step-by-step look at how to evaluate language models using Azure AI Studio. We’ll cover the importance of evaluating language models, the benefits of using Azure AI Studio, and provide a comprehensive guide on how to evaluate language models using the platform. Whether you’re a data scientist, machine learning engineer, or NLP practitioner, this blog post will provide you with the knowledge and skills you need to evaluate language models using Azure AI Studio.

Evaluating Language Models with Azure AI Studio: A Step-by-Step Guide

As the demand for natural language processing (NLP) and language models continues to grow, it’s essential to ensure that these models are performing optimally and reliably. Evaluating language models is a crucial step in achieving this goal, and Azure AI Studio provides a comprehensive platform for doing so. In this blog post, we’ll take a step-by-step look at how to evaluate language models using Azure AI Studio, complete with screenshots and a conclusion.

Why Evaluate Language Models?

Language models are a critical component of many NLP applications, including chatbots, sentiment analysis, and language translation. However, these models can be complex and prone to errors, which can have significant consequences. Evaluating language models helps to identify areas of improvement, optimize performance, and ensure that the models are reliable and accurate.

Step 1: Create a New Project

To get started with evaluating a language model in Azure AI Studio, you’ll need to create a new project. To do this, follow these steps:

Open Azure AI Studio and click on “New Project”
Choose “Language” as the project type
Select the language you want to evaluate

Step 2: Upload Your Model

Once you’ve created a new project, you’ll need to upload your language model to Azure AI Studio. To do this, follow these steps:

Click on “Upload” and select the model file
Azure AI Studio supports a range of model formats, including TensorFlow, PyTorch, and ONNX

Step 3: Configure Evaluation Settings

After uploading your model, you’ll need to configure the evaluation settings. This includes specifying the evaluation metric, dataset, and other parameters. To do this, follow these steps:

Click on “Configure” and select the evaluation metric
Choose the dataset you want to use for evaluation
Specify any additional parameters, such as the batch size and number of epochs

Step 4: Run Evaluation

Once you’ve configured the evaluation settings, you can run the evaluation by clicking on “Run”. Azure AI Studio will then execute the evaluation and provide the results.

Step 5: Analyze Results

After completing the evaluation, you can analyze the results to identify areas of improvement and optimize your language model. Azure AI Studio provides a range of visualization tools and metrics to help you understand your model’s performance.

The screenshot below shows an example of the evaluation results for a language model. In this example, the model is being evaluated on a sentiment analysis task, where the goal is to predict a given text’s sentiment (positive or negative).

In this image, we can see several key metrics and visualizations that help us understand the performance of the language model:

Accuracy: This metric shows the overall accuracy of the model in predicting the sentiment of the text. In this example, the accuracy is 85%, which indicates that the model is performing well.
Confusion Matrix: This visualization shows the number of true positives, false positives, true negatives, and false negatives. In this example, the confusion matrix shows that the model is correctly predicting positive sentiment in 80% of cases and negative sentiment in 90% of cases.
ROC Curve: This visualization shows the performance of the model at different thresholds. In this example, the ROC curve shows that the model is performing well, with an area under the curve (AUC) of 0.92.
Loss Curve: This visualization shows the loss function of the model over time. In this example, the loss curve shows that the model is converging, and the loss is decreasing over time.

By analyzing these metrics and visualizations, we can identify areas of improvement for the language model. For example, we may want to improve the accuracy of the model by fine-tuning the hyperparameters or adding more training data. We may also want to investigate why the model is performing poorly on certain types of text or sentiment.

Overall, the analysis of the evaluation results provides valuable insights into the performance of the language model and helps us to optimize and improve its performance.

Let’s Get hands-on with an example

GPT-4-0613 model from the Azure AI Studio model gallery.

 Let’s dive deeper into this model and evaluate its performance.

Model Details: GPT-4-0613 is a large language model developed by Meta AI, with 1.3 billion parameters. It’s a variant of the GPT-4 model, which is a family of transformer-based language models that have achieved state-of-the-art results on a range of NLP tasks.

Model Architecture: GPT-4-0613 uses a transformer-based architecture, with 24 layers and 16 attention heads. It’s trained on a massive dataset of text from the internet, with a focus on generating coherent and natural-sounding language.

Model Performance: GPT-4-0613 has achieved impressive results on a range of NLP tasks, including language translation, text summarization, and conversational dialogue. It’s particularly well-suited for tasks that require generating long-form text, such as writing articles or creating chatbot responses.

Step 2: Evaluate the Model To evaluate the performance of GPT-4-0613, we can use various metrics such as:

Metrics

Perplexity: measures how well the model predicts a test dataset
Accuracy: measures the proportion of correctly classified instances
F1-score measures the balance between precision and recall
ROUGE score: measures the quality of generated text

Let’s dive into a hands-on comparison of small language models, specifically GPT-4 and Phi-3.5.

We’ll explore how to evaluate and compare these models, highlighting their strengths and weaknesses.

To begin, let’s import the necessary libraries and load the models:

Next, let’s prepare a sample input text to evaluate the models:

Now, let’s evaluate the models using the input text:

We can now compare the results from both models:

This comparison will give us an idea of how the models perform on a specific task. We can further evaluate the models using various metrics, such as accuracy, F1-score, and perplexity.

Evaluation Metrics

To comprehensively evaluate the models, we can use various metrics, including:

Accuracy: Measures the proportion of correctly classified instances.
F1-score: Calculates the harmonic mean of precision and recall.
Perplexity: Evaluates the model’s ability to predict a sample of text.

Here’s an example of how to calculate these metrics:

By evaluating and comparing these metrics, we can gain a deeper understanding of the strengths and weaknesses of each model.

Comparison of GPT-4 and Phi-3.5

Now that we’ve evaluated the models, let’s compare their performance:

GPT-4: GPT-4 is a larger, more powerful model with a greater number of parameters (1.5 trillion) compared to Phi-3.5 (175 billion) [2]. This increased capacity enables GPT-4 to understand and generate more complex language patterns.
Phi-3.5: Phi-3.5, on the other hand, is a smaller, more efficient model that is better suited for deployment in resource-constrained environments. Despite its smaller size, Phi-3.5 still demonstrates impressive language understanding and generation capabilities.

In terms of performance, GPT-4 generally outperforms Phi-3.5 in tasks that require complex language understanding and generation. However, Phi-3.5’s smaller size and efficiency make it a more viable option for certain applications.

Evaluating Phi-3.5 with AI Studio

In this walkthrough, we’ll demonstrate how to evaluate the Phi-3.5 model using AI Studio. We’ll cover the necessary steps to prepare the input data, load the Phi-3.5 model, and calculate evaluation metrics.

Step 1: Prepare the Input Data

First, let’s prepare a sample input text to evaluate the Phi-3.5 model:

Step 2: Load the Phi-3.5 Model and Tokenizer

Next, we’ll load the Phi-3.5 model and tokenizer using the transformers library.

Step 3: Encode the Input Text

Now, let’s encode the input text using the Phi-3.5 tokenizer:

Step 4: Evaluate the Phi-3.5 Model

We’ll evaluate the Phi-3.5 model using the encoded input text:

Step 5: Calculate Evaluation Metrics

Finally, let’s calculate the perplexity of the Phi-3.5 model:

This walkthrough demonstrates how to evaluate the Phi-3.5 model using AI Studio. By following these steps, you can evaluate the performance of the Phi-3.5 model on your input data.

Comparing OpenAI GPT and Microsoft Phi-3 models in Azure AI Studio is a great way to evaluate their performance and understand their strengths and weaknesses.

Based on the search results, here’s a summary of the comparison:

OpenAI GPT Model

Advantages:

Higher accuracy on complex tasks
Better handling of long-range dependencies
More flexible and adaptable to different tasks

Disadvantages:

Requires more computational resources and memory
Slower inference times
More prone to overfitting

Microsoft Phi-3 Model

Advantages:

More efficient and lightweight, requiring fewer computational resources
Faster inference times
Less prone to overfitting

Disadvantages:

Lower accuracy on complex tasks
Struggles with long-range dependencies
Less flexible and adaptable to different tasks

Here’s a sample comparison table for the blog:

Model

Accuracy

Inference Time

Computational Resources

OpenAI GPT

92.5%

500ms

High

Microsoft Phi-3

88.2%

100ms

Low

Note that the actual numbers may vary depending on the specific task, dataset, and evaluation metrics used.

Here’s a sample Python code to compare the models in Azure AI Studio:

Conclusion

Evaluating language models using Azure AI Studio provides a comprehensive and efficient way to assess and improve your models. By following these steps, you can ensure that your language models are performing optimally and reliably. Whether you’re building a chatbot, or sentiment analysis.

Microsoft Resources

Evaluating the quality of AI document data extraction with small and large language models (microsoft.com)

Azure AI Studio documentation | Microsoft Learn

Evaluation of generative AI applications with Azure AI Studio – Azure AI Studio | Microsoft Learn

What is Azure AI Studio? – Azure AI Studio | Microsoft Learn

Azure AI Studio – Generative AI Development Hub | Microsoft Azure

Azure OpenAI Service models – Azure OpenAI | Microsoft Learn

​Microsoft Tech Community – Latest Blogs –Read More 

I would like the marked cell to come up empty when copying the formula to empty rows… what is wrong with the formula? 

All help is appreciated!

​I would like the marked cell to come up empty when copying the formula to empty rows… what is wrong with the formula?  All help is appreciated!  Read More

​

I am developing a type of import template. In which I would like a color coding to indicate which cells they need to fill in dependent on another cell and its drop-down list. 

Given the value in column L, I would like the columns O:BV to change colors.

I have made it to work for one cell with formula:
=count.if(Blad2!$B:$B;L7)>0

But can’t get the as in this instance L7 to change for L8 and so on. With this formula my thought was to make one for each column O:BV, but perhaps there is a better way?

Blad2 looks like this:

​I am developing a type of import template. In which I would like a color coding to indicate which cells they need to fill in dependent on another cell and its drop-down list.  Given the value in column L, I would like the columns O:BV to change colors.I have made it to work for one cell with formula:=count.if(Blad2!$B:$B;L7)>0But can’t get the as in this instance L7 to change for L8 and so on. With this formula my thought was to make one for each column O:BV, but perhaps there is a better way?Blad2 looks like this:  Read More

​