<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: uwimana Xaverine</title>
    <description>The latest articles on DEV Community by uwimana Xaverine (@uwimana_xaverine_44456038).</description>
    <link>https://dev.to/uwimana_xaverine_44456038</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4151926%2Faf093b26-bf7c-47ad-9241-b45deb825290.png</url>
      <title>DEV Community: uwimana Xaverine</title>
      <link>https://dev.to/uwimana_xaverine_44456038</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/uwimana_xaverine_44456038"/>
    <language>en</language>
    <item>
      <title>ASSGNIMENT</title>
      <dc:creator>uwimana Xaverine</dc:creator>
      <pubDate>Mon, 26 Oct 2026 14:20:02 +0000</pubDate>
      <link>https://dev.to/uwimana_xaverine_44456038/assgniment-376i</link>
      <guid>https://dev.to/uwimana_xaverine_44456038/assgniment-376i</guid>
      <description>&lt;p&gt;&lt;strong&gt;Greetings!&lt;/strong&gt;&lt;br&gt;
I have learned about &lt;strong&gt;Large language models&lt;/strong&gt; are AI systems used to model and process human language. They are called “large” because these types of models are normally made of hundreds of millions or even billions of parameters that define the model's behavior, which are pre-trained using a massive corpus of text data.&lt;/p&gt;

&lt;p&gt;In Three Models, the Project: Which Is Best for Kinyarwanda?&lt;br&gt;
Gemma is the model I would choose to improve our project, because it is the one most likely to understand Kinyarwanda. This blog explains the three models I compared, in simple words.&lt;br&gt;
Our project in one minute&lt;br&gt;
In our NLP class, we looked inside GPT-2, a small language model. We also built our own tiny GPT and trained it on Kinyarwanda news.&lt;br&gt;
We found three problems:&lt;/p&gt;

&lt;p&gt;*&lt;em&gt;•GPT-2 *&lt;/em&gt; splits Kinyarwanda words into many small pieces. It needs about 1.7 times more pieces than English for the same sentence.&lt;br&gt;
•GPT-2 learned almost only from English, so it cannot write good Kinyarwanda.&lt;br&gt;
•Our tiny GPT is very small, so it writes text that looks like Kinyarwanda but makes no sense.&lt;br&gt;
So we need a model that is better with Kinyarwanda. We also need a model we can open and look inside, because that is what our project is about.&lt;/p&gt;

&lt;p&gt;What is a model, and what is a token?&lt;br&gt;
A language model is a program that reads text and guesses the next word. A token is a small piece of text that the model reads. English words are often one token. Words from languages the model did not study are cut into many tokens, and that makes reading harder.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;1.Phi-3.5-mini&lt;/strong&gt;&lt;br&gt;
&lt;strong&gt;FeatureS&lt;/strong&gt;c&lt;br&gt;
&lt;strong&gt;•Creator:&lt;/strong&gt; Microsoft&lt;br&gt;
&lt;strong&gt;•Size:&lt;/strong&gt; 3.8 billion parameters&lt;br&gt;
&lt;strong&gt;•Context Window:&lt;/strong&gt; Up to 128,000 tokens&lt;br&gt;
&lt;strong&gt;•License:&lt;/strong&gt; MIT License (Highly permissive)&lt;br&gt;
•What it is good at: Despite a compact footprint, Phi-3.5 was trained on a heavily curated, high-quality dataset containing vast amounts of multilingual text and code data. It punches well above its weight class for reasoning, instruction following, and structured formatting.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Model 2: Qwen2.5-1.5B&lt;/strong&gt;&lt;br&gt;
Qwen is a newer model with 1.5 billion parameters. It learned from text in many languages, not only English.&lt;br&gt;
&lt;strong&gt;•Good&lt;/strong&gt;: it is much bigger than GPT-2, it knows more languages, and it needs no login to download.&lt;br&gt;
&lt;strong&gt;•Bad:&lt;/strong&gt; it is smaller than Gemma, and I have not tested how well it handles Kinyarwanda.&lt;br&gt;
&lt;strong&gt;•Role in our project:&lt;/strong&gt; it is our backup choice.&lt;br&gt;
&lt;strong&gt;Feature&lt;/strong&gt;&lt;br&gt;
&lt;strong&gt;•Creator:&lt;/strong&gt; Alibaba Group&lt;br&gt;
&lt;strong&gt;•Size:&lt;/strong&gt; 1.5 billion parameters&lt;br&gt;
&lt;strong&gt;•Context Window:&lt;/strong&gt; Up to 32,768 tokens&lt;br&gt;
&lt;strong&gt;•License:&lt;/strong&gt; Apache 2.0 / Qwen License (Permissive commercial and research use)&lt;br&gt;
• What it is good at: Trained on a massive multilingual dataset covering over 29 languages. It features strong instruction-following capabilities, structured data generation, and an exceptionally generous context window for its size class.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Model 3: Gemma-2-2b&lt;/strong&gt;&lt;br&gt;
Gemma is a model from Google with 2 billion parameters. It also learned from many languages. From what I know of the Gemma family, its list of tokens is very large, about 256,000 pieces, and it was made to work across many languages.&lt;br&gt;
&lt;strong&gt;•Good: *&lt;em&gt;it is the biggest of the three, and its large token list should cut Kinyarwanda words into fewer pieces.&lt;br&gt;
*&lt;/em&gt;•Bad:&lt;/strong&gt; you must accept a license and log in to Hugging Face to download it, and I have not tested it on Kinyarwanda yet.&lt;br&gt;
&lt;strong&gt;•Role in our project:&lt;/strong&gt; it is our main choice.&lt;br&gt;
&lt;strong&gt;Feature&lt;/strong&gt;&lt;br&gt;
&lt;strong&gt;•Creator:&lt;/strong&gt; Google&lt;br&gt;
&lt;strong&gt;•Size:&lt;/strong&gt; 2 billion parameters&lt;br&gt;
&lt;strong&gt;•Context Window: *&lt;em&gt;8,192 tokens&lt;br&gt;
*&lt;/em&gt;•License&lt;/strong&gt;: Gemma Terms of Use (Requires a free Hugging Face account and license acceptance)&lt;br&gt;
&lt;strong&gt;•What it is good at:&lt;/strong&gt; Built from research and technology similar to Google's larger production models, Gemma features a massive vocabulary size (roughly 256,000 tokens) designed explicitly to handle multilingual corpora and tokenization efficiency across diverse languages.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;The best one for our project:&lt;/strong&gt; Gemma&lt;br&gt;
I choose Gemma for four reasons.&lt;br&gt;
1.It should read Kinyarwanda better. Our biggest finding was that GPT-2 cuts Kinyarwanda into too many pieces. Gemma has a bigger token list made for many languages, so it should need fewer pieces.&lt;br&gt;
2.It already knows how language works. Our tiny GPT had to learn everything from a small amount of news text. Gemma already knows a lot, so we only need to teach it Kinyarwanda. We can do this with a small training step called LoRA, which changes only a few of the model's numbers.&lt;br&gt;
3.It is bigger. More parameters means more room to learn a new language.&lt;br&gt;
4.We can still look inside it. Gemma is an open model, so our attention plots and layer experiments still work. Closed models like ChatGPT or Gemini only give answers, so we could not study them the same way.&lt;br&gt;
Qwen is a close second. If Gemma's login is a problem, &lt;br&gt;
Qwen is a good choice.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;References&lt;/strong&gt;&lt;br&gt;
•Google DeepMind. (2024). Gemma 2: Improving Open Language Models at Size. Google AI Technical Report.&lt;br&gt;
•Alibaba Group. (2024). Qwen2.5 Technical Report. Qwen Documentation and GitHub Repository.&lt;br&gt;
•Microsoft. (2024). Phi-3.5 Technical Report: Highly Capable Open Models on Your Device. Microsoft Research.&lt;br&gt;
•Hugging Face Model Hub: Gemma-2-2b Model Card, Qwen2.5-1.5B Model Card, and Phi-3.5-mini Model Card.&lt;/p&gt;

</description>
      <category>ai</category>
    </item>
    <item>
      <title>Computer Vision</title>
      <dc:creator>uwimana Xaverine</dc:creator>
      <pubDate>Tue, 13 Oct 2026 13:38:27 +0000</pubDate>
      <link>https://dev.to/uwimana_xaverine_44456038/computer-vision-4kop</link>
      <guid>https://dev.to/uwimana_xaverine_44456038/computer-vision-4kop</guid>
      <description>&lt;p&gt;Computer vision is a part of artificial intelligence (AI). It helps computers see, process, and understand pictures and videos just like human eyes and brains do. &lt;/p&gt;




&lt;h2&gt;
  
  
  The Different Ways Computers Look at Things
&lt;/h2&gt;

&lt;p&gt;Depending on what we want the computer to do, computer vision jobs usually fall into a few clear categories:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Image Classification (What is this?):&lt;/strong&gt; The computer looks at a whole picture and gives it one simple label. For example: &lt;em&gt;"This photo is a cup of coffee."&lt;/em&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Object Detection (Where is it?):&lt;/strong&gt; The computer not only figures out what is in the picture, but draws a box around it to show where it is. For example: &lt;em&gt;"Here is a car, and here is a pedestrian."&lt;/em&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Image Segmentation (Tracing the exact lines):&lt;/strong&gt; This is like coloring inside the lines. The computer figures out the exact shape of an object down to the individual pixel.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;OCR / Reading Text:&lt;/strong&gt; The computer scans images or signs to read and pull out text, turning a picture of a street sign into actual typed words.&lt;/li&gt;
&lt;/ul&gt;




&lt;h2&gt;
  
  
  How It Works
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Taking in the Picture:&lt;/strong&gt; Cameras or sensors capture a picture as a grid of tiny dots called pixels, along with color and light details.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Cleaning It Up:&lt;/strong&gt; The system cleans the image, fixes the brightness, and removes any blurry spots.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Finding Patterns:&lt;/strong&gt; Computer programs scan the image to look for basic lines, shapes, and corners.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Understanding the Object:&lt;/strong&gt; Smart computer models match these patterns to figure out what is in the picture and give it a label.&lt;/li&gt;
&lt;/ul&gt;




&lt;h2&gt;
  
  
  How Do Computers "See"
&lt;/h2&gt;

&lt;p&gt;Unlike humans—who look at a friend's face or a stop sign and instantly know what it is—computers don't have eyes, brains, or feelings. Instead, they look at everything using numbers.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;The Grid of Pixels:&lt;/strong&gt; To a computer, a picture is just a giant checkerboard (grid) made of tiny dots called pixels.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Colors as Numbers:&lt;/strong&gt; Every dot has a secret number code that tells the computer how much Red, Green, and Blue (RGB) light to show, using numbers from 0 to 255.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Finding Patterns:&lt;/strong&gt; The computer runs simple math checks over the grid:

&lt;ul&gt;
&lt;li&gt;First, it finds basic things like edges, lines, and bright spots.&lt;/li&gt;
&lt;li&gt;Next, it connects those lines to see shapes and textures.&lt;/li&gt;
&lt;li&gt;Finally, it puts the shapes together to recognize whole objects.&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Computer Vision Models and Their Primary Functions
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Image Classification&lt;/strong&gt; (ResNet, EfficientNet, Vision Transformer): Assigns a single overarching label to an entire image by analyzing global patterns or pixel arrangements.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Object Detection&lt;/strong&gt; (YOLO, Faster R-CNN): Detects what objects are present and where they are located by drawing precise bounding boxes around them.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Image Segmentation&lt;/strong&gt; (U-Net, Mask R-CNN): Performs pixel-level analysis to isolate exact shapes, boundaries, and individual object instances.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Generative &amp;amp; Multimodal Models&lt;/strong&gt; (Stable Diffusion, CLIP): Generates brand new visual media from text prompts or bridges text and image understanding for tasks like zero-shot recognition.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;In simple terms, computer vision is how we teach computers to "see" and understand the world through pictures and videos. Instead of just looking at random dots, smart models help machines recognize what is in an image, find where things are located, trace exact shapes, or even create entirely new pictures. It turns basic numbers into useful intelligence, making technology much more helpful in everyday life&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Image classification and detection&lt;/strong&gt;
In practical work I use this image for showing Image classifier and detect
&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F4vxaljp9of273d2qm59e.png" alt=" " width="621" height="356"&gt;
&lt;/li&gt;
&lt;/ul&gt;

</description>
      <category>ai</category>
      <category>python</category>
    </item>
    <item>
      <title>WORD EMBENDDING</title>
      <dc:creator>uwimana Xaverine</dc:creator>
      <pubDate>Wed, 30 Sep 2026 13:55:53 +0000</pubDate>
      <link>https://dev.to/uwimana_xaverine_44456038/word-embendding-1kid</link>
      <guid>https://dev.to/uwimana_xaverine_44456038/word-embendding-1kid</guid>
      <description>&lt;p&gt;I learned how the computer can understand words through "word embedding" &lt;br&gt;
Word embeddings are a technique used in Natural Language Processing (NLP) to help computers understand words and the relationships between them.&lt;/p&gt;

&lt;p&gt;Computers do not understand words in the same way humans do. When we see words such as mango and banana, we understand that they are both fruits. A computer needs a numerical representation to process and compare these words.&lt;/p&gt;

&lt;p&gt;Word embeddings solve this problem by converting words into vectors, which are groups of numbers. These numerical representations help computers identify similarities and relationships between words.&lt;/p&gt;

&lt;p&gt;Example: Mango and Banana&lt;/p&gt;

&lt;p&gt;Imagine that a word embedding model represents words using three numbers:&lt;/p&gt;

&lt;p&gt;Mango → [0.80, 0.75, 0.20]&lt;br&gt;
Banana → [0.78, 0.72, 0.25]&lt;br&gt;
Car → [0.10, 0.15, 0.90]&lt;/p&gt;

&lt;p&gt;These numbers do not directly mean "fruit" or "vehicle." Instead, they are learned from the way words are used in sentences.&lt;/p&gt;

&lt;p&gt;Because mango and banana are often used in similar contexts, their vector representations can be close to each other. Car is used in very different contexts, so its vector can be farther away.&lt;/p&gt;

&lt;p&gt;For example:&lt;/p&gt;

&lt;p&gt;"I ate a ripe mango."&lt;/p&gt;

&lt;p&gt;and&lt;/p&gt;

&lt;p&gt;"I ate a ripe banana."&lt;/p&gt;

&lt;p&gt;The words mango and banana appear in similar contexts. When an embedding model processes many sentences, it can learn that these words have a similar relationship.&lt;br&gt;
How Does Word Embedding Work?&lt;/p&gt;

&lt;p&gt;Word embeddings are created by training a model on a large collection of text. The model studies the context surrounding words and learns which words tend to appear in similar situations.&lt;/p&gt;

&lt;p&gt;The basic idea is:&lt;/p&gt;

&lt;p&gt;Words used in similar contexts → Similar vector representations&lt;/p&gt;

&lt;p&gt;For example:&lt;/p&gt;

&lt;p&gt;Mango → fruit, sweet, food, eat, ripe&lt;/p&gt;

&lt;p&gt;Banana → fruit, sweet, food, eat, ripe&lt;/p&gt;

&lt;p&gt;The model learns these relationships from the text. It does not need someone to manually tell it that mango and banana are fruits.&lt;/p&gt;

&lt;p&gt;Common Word Embedding Techniques&lt;/p&gt;

&lt;p&gt;Some popular word embedding techniques include:&lt;/p&gt;

&lt;p&gt;Word2Vec&lt;br&gt;
GloVe&lt;br&gt;
FastText&lt;/p&gt;

&lt;p&gt;These techniques use different methods to learn useful numerical representations of words.&lt;br&gt;
Word2Vec&lt;/p&gt;

&lt;p&gt;Word2Vec is one of the most popular traditional word embedding methods. It learns word representations by looking at the words that appear around a particular word.&lt;/p&gt;

&lt;p&gt;For example:&lt;/p&gt;

&lt;p&gt;"I like eating a ripe mango."&lt;/p&gt;

&lt;p&gt;If the model sees many sentences where mango appears near words such as ripe, eating, fruit, sweet, and food, it learns a vector representation for mango based on these relationships.&lt;/p&gt;

&lt;p&gt;Word2Vec has two main approaches:&lt;/p&gt;

&lt;p&gt;CBOW (Continuous Bag of Words): Predicts a missing word using the surrounding words.&lt;/p&gt;

&lt;p&gt;Skip-gram: Uses a word to predict the words that appear around it.&lt;/p&gt;

&lt;p&gt;Why Are Vectors Important?&lt;/p&gt;

&lt;p&gt;Once words are converted into vectors, computers can perform mathematical operations to compare them.&lt;/p&gt;

&lt;p&gt;For example:&lt;/p&gt;

&lt;p&gt;Mango ↔ Banana → High similarity&lt;/p&gt;

&lt;p&gt;Mango ↔ Car → Lower similarity&lt;/p&gt;

&lt;p&gt;This allows NLP systems to identify relationships between words.&lt;/p&gt;

&lt;p&gt;The important point is that the model learns from patterns and contexts, rather than simply memorizing the spelling of a word.&lt;br&gt;
Applications of Word Embeddings&lt;/p&gt;

&lt;p&gt;Word embeddings are useful in many NLP applications, including:&lt;/p&gt;

&lt;p&gt;*Sentiment analysis&lt;br&gt;
*Text classification&lt;br&gt;
*Search engines&lt;br&gt;
*Chatbots&lt;br&gt;
*Machine translation&lt;/p&gt;

&lt;p&gt;They help computers process language and identify relationships between words.&lt;br&gt;
 Word Embeddings and AI&lt;/p&gt;

&lt;p&gt;Word embeddings are an important development in Artificial Intelligence and Natural Language Processing because they allow computers to represent language mathematically.&lt;/p&gt;

&lt;p&gt;Before word embeddings, techniques such as one-hot encoding represented each word as a separate category. This made it difficult for computers to understand relationships between words.&lt;/p&gt;

&lt;p&gt;With word embeddings, related words can have similar representations.&lt;/p&gt;

&lt;p&gt;For example:&lt;/p&gt;

&lt;p&gt;Mango → Fruit&lt;/p&gt;

&lt;p&gt;Banana → Fruit&lt;/p&gt;

&lt;p&gt;Apple → Fruit&lt;/p&gt;

&lt;p&gt;These words can have similar representations because they occur in related contexts.&lt;/p&gt;

&lt;p&gt;Conclusion&lt;/p&gt;

&lt;p&gt;Word embeddings transform words into numerical vectors that computers can process and compare. They help NLP models learn relationships between words based on their context and usage.&lt;/p&gt;

&lt;p&gt;The example of mango and banana makes the idea simple. They are different words, but they are both fruits and can appear in similar sentences. A word embedding model can learn this relationship from large amounts of text.&lt;/p&gt;

&lt;p&gt;The main idea is:&lt;/p&gt;

&lt;p&gt;Words → Numerical Vectors → Similarities → Relationships → NLP Understanding&lt;/p&gt;

&lt;p&gt;Therefore, word embeddings are an important concept in Natural Language Processing and Artificial Intelligence, helping computers work with human language in a more meaningful way.&lt;/p&gt;

</description>
      <category>ai</category>
      <category>machinelearning</category>
    </item>
  </channel>
</rss>
