<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[MLWorks - Medium]]></title>
        <description><![CDATA[Learn Python, ML, and Software Engineering - Medium]]></description>
        <link>https://medium.com/mlworks?source=rss----197d34a8de74---4</link>
        <image>
            <url>https://cdn-images-1.medium.com/proxy/1*TGH72Nnw24QL3iV9IOm4VA.png</url>
            <title>MLWorks - Medium</title>
            <link>https://medium.com/mlworks?source=rss----197d34a8de74---4</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 17:35:05 GMT</lastBuildDate>
        <atom:link href="https://medium.com/feed/mlworks" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Deletion Tuning - The Surprising Effectiveness of Deleting Weights in LLM Post-Training]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/mlworks/deletion-tuning-the-surprising-effectiveness-of-deleting-weights-in-llm-post-training-ece41b1cd699?source=rss----197d34a8de74---4"><img src="https://cdn-images-1.medium.com/max/2600/0*Q8VktYocMoig5nBE" width="5472"></a></p><p class="medium-feed-snippet">During fine-tuning, we teach a pretrained language model a new capability, and we typically modify its weights to achieve this. There are&#x2026;</p><p class="medium-feed-link"><a href="https://medium.com/mlworks/deletion-tuning-the-surprising-effectiveness-of-deleting-weights-in-llm-post-training-ece41b1cd699?source=rss----197d34a8de74---4">Continue reading on MLWorks »</a></p></div>]]></description>
            <link>https://medium.com/mlworks/deletion-tuning-the-surprising-effectiveness-of-deleting-weights-in-llm-post-training-ece41b1cd699?source=rss----197d34a8de74---4</link>
            <guid isPermaLink="false">https://medium.com/p/ece41b1cd699</guid>
            <category><![CDATA[neurons]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[neural-networks]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Mayur Jain]]></dc:creator>
            <pubDate>Sat, 14 Nov 2026 09:20:49 GMT</pubDate>
            <atom:updated>2026-11-14T09:21:19.744Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[ Meta’s E-Means beats K-Means Clustering]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/mlworks/metas-e-means-beats-k-means-clustering-93c7658184b6?source=rss----197d34a8de74---4"><img src="https://cdn-images-1.medium.com/max/2600/0*DQgvj1AUDVgfLZSJ" width="7680"></a></p><p class="medium-feed-snippet">Introduction</p><p class="medium-feed-link"><a href="https://medium.com/mlworks/metas-e-means-beats-k-means-clustering-93c7658184b6?source=rss----197d34a8de74---4">Continue reading on MLWorks »</a></p></div>]]></description>
            <link>https://medium.com/mlworks/metas-e-means-beats-k-means-clustering-93c7658184b6?source=rss----197d34a8de74---4</link>
            <guid isPermaLink="false">https://medium.com/p/93c7658184b6</guid>
            <category><![CDATA[machine-learning-ai]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Mayur Jain]]></dc:creator>
            <pubDate>Wed, 11 Nov 2026 13:54:13 GMT</pubDate>
            <atom:updated>2026-11-11T13:55:17.068Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[What is A2A?]]></title>
            <link>https://medium.com/mlworks/what-is-a2a-protocol-2314e464aea9?source=rss----197d34a8de74---4</link>
            <guid isPermaLink="false">https://medium.com/p/2314e464aea9</guid>
            <category><![CDATA[a2a]]></category>
            <category><![CDATA[a2a-protocol]]></category>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[agents]]></category>
            <category><![CDATA[genai]]></category>
            <dc:creator><![CDATA[Kavya]]></dc:creator>
            <pubDate>Wed, 11 Nov 2026 07:06:17 GMT</pubDate>
            <atom:updated>2026-11-11T07:06:15.862Z</atom:updated>
            <content:encoded><![CDATA[<h4>Beginner’s story in the A2A protocol</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*EMGDWhHgCAzO2uBJ.png" /></figure><p>You know when people created an agent which can do one certain task successfully, there is always a need to have multiple agents who can do their specific tasks successfully and pass them to the output, either as a shared one or in a sequential way to provide the final answer.</p><p>Imagine a person named Maya, who is asking a chatbot a query: “Am I eligible to avail the discount on this product?”</p><p>The query looks simple, but in the background we have to check which product she is describing, evaluate her profile for that product (which requires fetching her stored information), and then compare it with the discount for the product. An agent that has to do all these might struggle, as it can do only one task. So then, how do we do it?</p><h4>A2A protocol</h4><p>That’s when they introduced the A2A protocol (aka Agent-to-Agent protocol), which simply means one agent interacts with another agent to share the output from one agent to the other.</p><p>In our case, it would be an agent who checks Maya’s eligibility for the product, which is shared with another agent comparing it with the discount, and another agent which is the orchestrator that shares the final response in the chat. So that’s how it is.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*QGOWVFyRvgO_VSmP0dmvhA.png" /></figure><h4>Client vs. Server Agents</h4><p>Let’s see how A2A works. The primary part of A2A is that it has one client agent (orchestrator) and one server agent (remote agent). The orchestrator agent is the client agent that assigns tasks based on the user query to the server agent to execute the task. It also has the instructions on how to respond back. An example would be: if the output needs to be a number, the client agent will have the prompt instructions set on responding back with only a number. There are multiple ways apart from this to do that as well.</p><p>Next, the server agent gets the input from the client agent, executes it in a task-based manner, and sends back the response. And this is the main concept behind it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*FEeodT-LehX8Eoh9.gif" /></figure><h4>How Agents Find Each Other (Agent Card)</h4><p>Next comes: how does the client agent identify which agent it needs to call based on the query? The answer is using an Agent Card. An Agent Card is kind of like a business card for each agent that tells what each agent does, its skills, input and output methods, authentication, and the URL (if any) to invoke the agent.</p><p>This Agent Card will be shared with the client agent, and using an LLM, it identifies which agent it has to call to get the correct response. Usually, it can be found in /.well-known/agent.json.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*j-e9DA4EXSTO83PB.png" /></figure><h4>How the Interaction Happens</h4><p>Now comes the part of how the server agent executes its work. The first step of identifying which agent to call is done; now comes the execution.</p><p>The server and client agents communicate with JSON-RPC over the network. The client agent puts the input query in a message and sends it to the server agent with a unique ID. The server agent receives it, and a task is created on the server agent to execute it. Until it shares the response, it sends messages in the form of SSE (Server-Sent Events) to share updates with the client. In case the server requires any extra information, it asks for it from the client. Once the server has generated the output, it sends the information back to the client agent in an artifact. So these are the communication parts that happen between the client and server agents:</p><p><strong>Task:</strong> It is a state that has been created to start working on the query to get the response from the server agent.</p><p><strong>Message:</strong> It is a conversational exchange between agents. It has parts and context.</p><p><strong>Part:</strong> It is an array of content chunks.</p><p><strong>Context:</strong> It holds the context that has been shared between agents; it can be session ID info (context ID) or a task ID reference.</p><p><strong>Artifact:</strong> It holds the output generated by the server agent.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*kp3RWcF-qqm7kaRh.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Fbgkf6iq8gq4C_Nb.png" /></figure><p>So how do we do this in our example above? In our example, Maya’s query is shared with the server agent who checks the eligibility criteria through the client agent after identifying its skills using the Agent Card. Then a task is created on the server agent’s side to start working to generate the output. Context is passed between them via SSE, and the final artifact that holds the response is passed back to the client agent. Based on its prompts, it shares the info with Maya in the proper format!</p><h4>Basic Python Project Structure</h4><p>How do we implement this in code? A basic Python flow uses three main files. It can differ based on project requirements; this is just a sample:</p><p><strong>main.py:</strong> This is where we define the Agent Card and the skills of the agent we use.</p><p><strong>agent_executor.py:</strong> This is where we execute the agent with A2A task execution.</p><p><strong>client.py:</strong> This resolves the remote Agent Card, builds the task message, and sends it over HTTP.</p><p>Detailed info on this is shared here:</p><ul><li><a href="https://github.com/a2aproject/a2a-samples">https://github.com/a2aproject/a2a-samples</a></li><li><a href="https://a2a-protocol.org/latest/tutorials/python/1-introduction/">https://a2a-protocol.org/latest/tutorials/python/1-introduction/</a></li></ul><h4>Useful Resources</h4><ul><li><a href="https://codelabs.developers.google.com/intro-a2a-purchasing-concierge">https://codelabs.developers.google.com/intro-a2a-purchasing-concierge</a></li><li><a href="https://medium.com/@genai_cybage_software/mastering-googles-a2a-protocol-the-complete-guide-to-agent-to-agent-communication-8d3ba985a10d">https://medium.com/@genai_cybage_software/mastering-googles-a2a-protocol-the-complete-guide-to-agent-to-agent-communication-8d3ba985a10d</a></li></ul><p>So this is the basic working of A2A, which uses a client agent to find the correct agent via Agent Cards (which hold skills), calls it by sharing messages, gets the response, and shares it with the user.</p><p>Happy learning!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2314e464aea9" width="1" height="1" alt=""><hr><p><a href="https://medium.com/mlworks/what-is-a2a-protocol-2314e464aea9">What is A2A?</a> was originally published in <a href="https://medium.com/mlworks">MLWorks</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Arete-27B — Removing Chinese Propaganda from the Model]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/mlworks/arete-27b-removing-chinese-propaganda-from-the-model-dce2911c327d?source=rss----197d34a8de74---4"><img src="https://cdn-images-1.medium.com/max/1390/1*UYUL8g3de1dIjP7rWeNnDw.png" width="1390"></a></p><p class="medium-feed-snippet">NousResearch post-trained Alibaba&#x2019;s Qwen3.8&#x2013;27B to remove Chinese values and add Western values to create Arete-27B.</p><p class="medium-feed-link"><a href="https://medium.com/mlworks/arete-27b-removing-chinese-propaganda-from-the-model-dce2911c327d?source=rss----197d34a8de74---4">Continue reading on MLWorks »</a></p></div>]]></description>
            <link>https://medium.com/mlworks/arete-27b-removing-chinese-propaganda-from-the-model-dce2911c327d?source=rss----197d34a8de74---4</link>
            <guid isPermaLink="false">https://medium.com/p/dce2911c327d</guid>
            <category><![CDATA[china]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[usa]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Mayur Jain]]></dc:creator>
            <pubDate>Tue, 10 Nov 2026 11:26:41 GMT</pubDate>
            <atom:updated>2026-11-10T11:27:12.187Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Continual Learning: How Can AI Models Learn New Things Without Forgetting the Old?]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/mlworks/continual-learning-how-can-ai-models-learn-new-things-without-forgetting-the-old-9b5e9a834140?source=rss----197d34a8de74---4"><img src="https://cdn-images-1.medium.com/max/2600/7d804d2762ae51a5f70d2d92fa8b20777a9a9abd6cf87650adca689c1cd29387" width="2624"></a></p><p class="medium-feed-snippet">Imagine training an AI model to recognize cats and dogs. After it achieves good accuracy, you update it with a new dataset of cars, trucks&#x2026;</p><p class="medium-feed-link"><a href="https://medium.com/mlworks/continual-learning-how-can-ai-models-learn-new-things-without-forgetting-the-old-9b5e9a834140?source=rss----197d34a8de74---4">Continue reading on MLWorks »</a></p></div>]]></description>
            <link>https://medium.com/mlworks/continual-learning-how-can-ai-models-learn-new-things-without-forgetting-the-old-9b5e9a834140?source=rss----197d34a8de74---4</link>
            <guid isPermaLink="false">https://medium.com/p/9b5e9a834140</guid>
            <category><![CDATA[continual-learning]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Mayur Jain]]></dc:creator>
            <pubDate>Tue, 10 Nov 2026 08:01:03 GMT</pubDate>
            <atom:updated>2026-11-10T08:01:31.222Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[EmbeddingGemma 2: How Google Built a Small Multimodal Embedding Model That Understands Text…]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/mlworks/embeddinggemma-2-how-google-built-a-small-multimodal-embedding-model-that-understands-text-859d9e457aa8?source=rss----197d34a8de74---4"><img src="https://cdn-images-1.medium.com/max/2484/4eb1bfd5ecb8d5136034a6ee4ce45b0c7440a28d98b08d5fe0b371be8184dbc9" width="2484"></a></p><p class="medium-feed-snippet">Introduction: Why do we need multimodal embedding models?</p><p class="medium-feed-link"><a href="https://medium.com/mlworks/embeddinggemma-2-how-google-built-a-small-multimodal-embedding-model-that-understands-text-859d9e457aa8?source=rss----197d34a8de74---4">Continue reading on MLWorks »</a></p></div>]]></description>
            <link>https://medium.com/mlworks/embeddinggemma-2-how-google-built-a-small-multimodal-embedding-model-that-understands-text-859d9e457aa8?source=rss----197d34a8de74---4</link>
            <guid isPermaLink="false">https://medium.com/p/859d9e457aa8</guid>
            <category><![CDATA[embedding]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[llm]]></category>
            <dc:creator><![CDATA[Mayur Jain]]></dc:creator>
            <pubDate>Fri, 06 Nov 2026 14:04:07 GMT</pubDate>
            <atom:updated>2026-11-06T14:04:38.717Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[An Easy Introduction To CUDA]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/mlworks/an-easy-introduction-to-cuda-0a7859470071?source=rss----197d34a8de74---4"><img src="https://cdn-images-1.medium.com/max/1003/dddab79c5a23134bca1bb7bace4ea10866bbb532de14dbe870e495cd7a4f26fa" width="1003"></a></p><p class="medium-feed-snippet">CUDA C++ is just one of the ways you can create massively parallel applications with CUDA. It lets you use the powerful C++ programming&#x2026;</p><p class="medium-feed-link"><a href="https://medium.com/mlworks/an-easy-introduction-to-cuda-0a7859470071?source=rss----197d34a8de74---4">Continue reading on MLWorks »</a></p></div>]]></description>
            <link>https://medium.com/mlworks/an-easy-introduction-to-cuda-0a7859470071?source=rss----197d34a8de74---4</link>
            <guid isPermaLink="false">https://medium.com/p/0a7859470071</guid>
            <category><![CDATA[cuda]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[nvidia]]></category>
            <dc:creator><![CDATA[Mayur Jain]]></dc:creator>
            <pubDate>Wed, 04 Nov 2026 15:09:55 GMT</pubDate>
            <atom:updated>2026-11-04T15:10:25.473Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[ LLMs can read and edit neural net weights for better performance!]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/mlworks/llms-can-read-and-edit-neural-net-weights-for-better-performance-5eed8055bd99?source=rss----197d34a8de74---4"><img src="https://cdn-images-1.medium.com/max/1344/ff43a0e53513a386d02c7bc2e72d1ed4cb2bc333c93fcc726718ddfe94d93056" width="1344"></a></p><p class="medium-feed-snippet">The field of AI is undergoing drastic changes and innovation, things that seemed not practical.</p><p class="medium-feed-link"><a href="https://medium.com/mlworks/llms-can-read-and-edit-neural-net-weights-for-better-performance-5eed8055bd99?source=rss----197d34a8de74---4">Continue reading on MLWorks »</a></p></div>]]></description>
            <link>https://medium.com/mlworks/llms-can-read-and-edit-neural-net-weights-for-better-performance-5eed8055bd99?source=rss----197d34a8de74---4</link>
            <guid isPermaLink="false">https://medium.com/p/5eed8055bd99</guid>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Mayur Jain]]></dc:creator>
            <pubDate>Tue, 03 Nov 2026 17:57:59 GMT</pubDate>
            <atom:updated>2026-11-03T17:58:31.670Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[DeepSeek-V4.1-Flash - Manage 1  Million Token Context in 1 GB RAM]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/mlworks/deepseek-v4-1-flash-manage-1-million-token-context-in-1-gb-ram-5257c03288bc?source=rss----197d34a8de74---4"><img src="https://cdn-images-1.medium.com/max/1672/44093caab6ae3745ca6306fb4d7b4732ad750bea5e53b790bda4da3483acefb4" width="1672"></a></p><p class="medium-feed-snippet">When an LLM generates text, it uses the attention mechanism to understand the relationship between tokens. For every token generated, the&#x2026;</p><p class="medium-feed-link"><a href="https://medium.com/mlworks/deepseek-v4-1-flash-manage-1-million-token-context-in-1-gb-ram-5257c03288bc?source=rss----197d34a8de74---4">Continue reading on MLWorks »</a></p></div>]]></description>
            <link>https://medium.com/mlworks/deepseek-v4-1-flash-manage-1-million-token-context-in-1-gb-ram-5257c03288bc?source=rss----197d34a8de74---4</link>
            <guid isPermaLink="false">https://medium.com/p/5257c03288bc</guid>
            <category><![CDATA[deepseek]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Mayur Jain]]></dc:creator>
            <pubDate>Mon, 02 Nov 2026 11:31:33 GMT</pubDate>
            <atom:updated>2026-11-02T14:11:10.722Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[ Beyond Flat Pixels: 4 Mind-Blowing ChatGPT Image Styles You Need to Try Today]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/mlworks/beyond-flat-pixels-4-mind-blowing-chatgpt-image-styles-you-need-to-try-today-ab6caf4a2b9f?source=rss----197d34a8de74---4"><img src="https://cdn-images-1.medium.com/max/2600/0*kvsqFpffXtyStPwo" width="4608"></a></p><p class="medium-feed-snippet">Imagine holding a cozy, needle-felted miniature of your pet in your hand, or viewing your profile picture rendered as a weathered marble&#x2026;</p><p class="medium-feed-link"><a href="https://medium.com/mlworks/beyond-flat-pixels-4-mind-blowing-chatgpt-image-styles-you-need-to-try-today-ab6caf4a2b9f?source=rss----197d34a8de74---4">Continue reading on MLWorks »</a></p></div>]]></description>
            <link>https://medium.com/mlworks/beyond-flat-pixels-4-mind-blowing-chatgpt-image-styles-you-need-to-try-today-ab6caf4a2b9f?source=rss----197d34a8de74---4</link>
            <guid isPermaLink="false">https://medium.com/p/ab6caf4a2b9f</guid>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[ai-image-generator]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[generative-ai-tools]]></category>
            <dc:creator><![CDATA[Mayur Jain]]></dc:creator>
            <pubDate>Tue, 27 Oct 2026 07:28:12 GMT</pubDate>
            <atom:updated>2026-10-27T07:28:48.014Z</atom:updated>
        </item>
    </channel>
</rss>