<![CDATA[Stories by Aditya Inamdar on Medium]]> https://medium.com/@inamdaraditya?source=rss-2bfbfa3bca8f------2 https://cdn-images-1.medium.com/fit/c/150/150/1*Is_87f5wg-TVQQech9xK3Q.png Stories by Aditya Inamdar on Medium https://medium.com/@inamdaraditya?source=rss-2bfbfa3bca8f------2 Medium Sat, 14 Nov 2026 16:13:38 GMT <![CDATA[Build an n8n Agent That Can Explain Why It Refused a Refund]]>

Wire a chat model to two tools, inspect a wrong decision, and fix it by changing what the model is allowed to pass between them.

Continue reading on Medium »

]]> https://inamdaraditya.medium.com/build-an-n8n-agent-that-can-explain-why-it-refused-a-refund-bb2f8327a053?source=rss-2bfbfa3bca8f------2 https://medium.com/p/bb2f8327a053 Sat, 14 Nov 2026 07:20:54 GMT 2026-11-14T07:21:20.842Z <![CDATA[How to Spend $4 of AMD’s $100 GPU Credit on a Benchmark Worth Keeping]]>

Run your first ROCm experiment on an Instinct MI300X and find out what batching does to vector-search throughput.

Continue reading on Medium »

]]> https://inamdaraditya.medium.com/how-to-spend-4-of-amds-100-gpu-credit-on-a-benchmark-worth-keeping-3b2e267c6d0a?source=rss-2bfbfa3bca8f------2 https://medium.com/p/3b2e267c6d0a Sat, 14 Nov 2026 06:51:45 GMT 2026-11-14T06:52:09.703Z <![CDATA[Stable Diffusion Starts With 32 KiB of Noise. Where Does the GPU Memory Go?]]>

Follow a single prompt through Stable Diffusion v1, from 77 text positions to 4,096 image positions. Along the way, a few lines of Python…

Continue reading on Medium »

]]> https://inamdaraditya.medium.com/stable-diffusion-starts-with-32-kib-of-noise-where-does-the-gpu-memory-go-40f49f6d67a6?source=rss-2bfbfa3bca8f------2 https://medium.com/p/40f49f6d67a6 Sat, 14 Nov 2026 05:25:27 GMT 2026-11-14T05:25:55.184Z <![CDATA[Your Stable Diffusion Autoencoder Can Pass Its Test With the Wrong Latents]]>

Follow one image, one prompt, and one noise prediction through Stable Diffusion v1. Then reproduce a scaling bug that an image round trip…

Continue reading on Medium »

]]> https://inamdaraditya.medium.com/your-stable-diffusion-autoencoder-can-pass-its-test-with-the-wrong-latents-a617bac682c9?source=rss-2bfbfa3bca8f------2 https://medium.com/p/a617bac682c9 Fri, 13 Nov 2026 14:25:33 GMT 2026-11-13T14:25:57.780Z <![CDATA[A Diffusion Model Sees the Number 1. Should It Push Left or Right?]]>

Build a denoising diffusion probabilistic model from its training target to its sampling loop, then debug it against a distribution whose…

Continue reading on GoPenAI »

]]> https://blog.gopenai.com/a-diffusion-model-sees-the-number-1-should-it-push-left-or-right-04873c04490f?source=rss-2bfbfa3bca8f------2 https://medium.com/p/04873c04490f Fri, 13 Nov 2026 09:45:01 GMT 2026-11-13T09:45:01.720Z <![CDATA[How a Feature With Zero Weight Can Break an 8-Bit Linear Layer]]>

Inside LabML’s GPT-NeoX implementation of LLM.int8(), with a CPU lab that lets you watch a projection lose its answer and recover it.

Continue reading on Medium »

]]> https://inamdaraditya.medium.com/how-a-feature-with-zero-weight-can-break-an-8-bit-linear-layer-59c6cb5b3ca5?source=rss-2bfbfa3bca8f------2 https://medium.com/p/59c6cb5b3ca5 Fri, 13 Nov 2026 07:21:35 GMT 2026-11-13T07:22:07.074Z <![CDATA[We Tracked the Shape of a Neural Network as It Learned. The Shape Didn’t Tell Us Enough.]]>

Persistent homology can reveal how neural-network weights reorganize during training. But can those geometric changes distinguish learning…

Continue reading on Medium »

]]> https://inamdaraditya.medium.com/we-tracked-the-shape-of-a-neural-network-as-it-learned-the-shape-didnt-tell-us-enough-0128dba52a7b?source=rss-2bfbfa3bca8f------2 https://medium.com/p/0128dba52a7b Thu, 12 Nov 2026 08:15:15 GMT 2026-11-12T08:15:15.252Z <![CDATA[Qwen Got 31 Fewer Math Problems Right When Attention Heads Dropped Out in Bursts]]>

Two runs can lose the same planned amount of attention and take different paths through a calculation. A paired experiment on 250 math…

Continue reading on Medium »

]]> https://inamdaraditya.medium.com/qwen-got-31-fewer-math-problems-right-when-attention-heads-dropped-out-in-bursts-0c8e1b817769?source=rss-2bfbfa3bca8f------2 https://medium.com/p/0c8e1b817769 Thu, 12 Nov 2026 06:36:39 GMT 2026-11-12T06:37:05.979Z <![CDATA[The User Is Testing Your Code. Should the LLM Keep Their Conversation in Memory?]]>

An LLM server has to decide which conversation it can afford to forget. Intent-aware caching turns the pause between messages into an…

Continue reading on Medium »

]]> https://inamdaraditya.medium.com/the-user-is-testing-your-code-should-the-llm-keep-their-conversation-in-memory-d6cdd3081483?source=rss-2bfbfa3bca8f------2 https://medium.com/p/d6cdd3081483 Wed, 11 Nov 2026 15:22:02 GMT 2026-11-11T15:22:32.400Z <![CDATA[The AI Changed Its Answer 25.8% of the Time. Nobody Changed the Question.]]>

Researchers tested a 421-million-parameter decision model by rearranging the same answer choices. The model often changed its mind…

Continue reading on GoPenAI »

]]> https://blog.gopenai.com/the-ai-changed-its-answer-25-8-of-the-time-nobody-changed-the-question-d547a3f70b1b?source=rss-2bfbfa3bca8f------2 https://medium.com/p/d547a3f70b1b Wed, 11 Nov 2026 13:04:44 GMT 2026-11-11T13:05:17.013Z