![\"\"](\"https://substackcdn.com/image/fetch/$s_!NNCh!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fdfd5ee24-27f0-4d06-a987-e1c7247057d3_4800x2700.png\")
Many people asked me over the past months to share my workflow for how I come up with the LLM architecture sketches and drawings in my articles, talks, and the LLM-Gallery. So I thought it would be useful to document the process I usually follow.
\nThe short version is that I usually start with the official technical reports, but these days, papers are often less detailed than they used to be, especially for most open-weight models from industry labs.
\nThe good part is that if the weights are shared on the Hugging Face Model Hub and the model is supported in the Python transformers library, we can usually inspect the config file and the reference implementation directly to get more information about the architecture details. And “working” code doesn’t lie.
\n
Figure 1: The basic motivation for this workflow is that papers are often less detailed these days, but a working reference implementation gives us something concrete to inspect.
\nI should also say that this is mainly a workflow for open-weight models. It doesn’t really apply to models like ChatGPT, Claude, or Gemini, where the weights and details are proprietary.
\nAlso, this is intentionally a fairly manual process. You could automate parts of it. But if the goal is to learn how these architectures work, then doing a few of these by hand is, in my opinion, still one of the best exercises.
\n![\"\"](\"https://substackcdn.com/image/fetch/$s_!evXx!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fc06a1e90-e1bf-411f-964b-ce89852491a3_4168x1699.png\")
Figure 2: At a high level, the workflow goes from config files and code to architecture insights.
\n", "content_markdown": "Many people asked me over the past months to share my workflow for how I come up with the LLM architecture sketches and drawings in my articles, talks, and the [LLM-Gallery](https://sebastianraschka.com/llm-architecture-gallery/). So I thought it would be useful to document the process I usually follow.\n\nThe short version is that I usually start with the official technical reports, but these days, papers are often less detailed than they used to be, especially for most open-weight models from industry labs.\n\nThe good part is that if the weights are shared on the Hugging Face Model Hub and the model is supported in the Python [transformers](https://github.com/huggingface/transformers) library, we can usually inspect the config file and the reference implementation directly to get more information about the architecture details. And “working” code doesn’t lie.\n\n[![](https://substackcdn.com/image/fetch/$s_!NNCh!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fdfd5ee24-27f0-4d06-a987-e1c7247057d3_4800x2700.png \"\")](https://substackcdn.com/image/fetch/$s_!NNCh!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fdfd5ee24-27f0-4d06-a987-e1c7247057d3_4800x2700.png)\n\nFigure 1: The basic motivation for this workflow is that papers are often less detailed these days, but a working reference implementation gives us something concrete to inspect.\n\nI should also say that this is mainly a workflow for open-weight models. It doesn’t really apply to models like ChatGPT, Claude, or Gemini, where the weights and details are proprietary.\n\nAlso, this is intentionally a fairly manual process. You could automate parts of it. But if the goal is to learn how these architectures work, then doing a few of these by hand is, in my opinion, still one of the best exercises.\n\n[](https://substackcdn.com/image/fetch/$s_!evXx!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fc06a1e90-e1bf-411f-964b-ce89852491a3_4168x1699.png)\n\nFigure 2: At a high level, the workflow goes from config files and code to architecture insights.\n", "dateCreated": "2026-09-27T19:29:15.293681927+00:00", "date_created": "2026-04-18T11:24:36+00:00", "date_modified": "2026-04-18T11:24:36+00:00", "date_published": "2026-04-18T11:24:36+00:00", "description": "A learning-oriented workflow for understanding new open-weight model releases", "external_url": "https://magazine.sebastianraschka.com/p/workflow-for-understanding-llms", "git": { "commit": "ebb0941c5c06a2ab2377342d607535262a3e969e", "path": "items/magazine-sebastianraschka-com/2026/04/2026-04-18-my-workflow-for-understanding-llm-architectures.md", "permalink": "https://github.com/aymericbeaumet/aggr-instance/blob/ebb0941c5c06a2ab2377342d607535262a3e969e/items/magazine-sebastianraschka-com/2026/04/2026-04-18-my-workflow-for-understanding-llm-architectures.md" }, "id": "urn:aggr:item:b12d52a86cdc648942c3fbfc3f979bb837213a9b", "isBasedOn": { "@id": "https://magazine.sebastianraschka.com/p/workflow-for-understanding-llms" }, "mainEntity": { "@id": "https://magazine.sebastianraschka.com/p/workflow-for-understanding-llms", "@type": "CreativeWork", "archivedAt": { "@id": "https://aggr.aymericbeaumet.com/items/magazine-sebastianraschka-com/2026-04-18-my-workflow-for-understanding-llm-architectures/#webpage" }, "author": [ { "@type": "Person", "name": "Sebastian Raschka, PhD" } ], "dateModified": "2026-04-18T11:24:36+00:00", "datePublished": "2026-04-18T11:24:36+00:00", "headline": "My Workflow for Understanding LLM Architectures", "timeRequired": "PT2M", "url": "https://magazine.sebastianraschka.com/p/workflow-for-understanding-llms", "wordCount": 252 }, "name": "Archived snapshot: My Workflow for Understanding LLM Architectures", "preview": null, "reading_minutes": 2, "replicated_at": null, "source": "magazine-sebastianraschka-com", "source_name": "Ahead of AI", "summary": "A learning-oriented workflow for understanding new open-weight model releases", "tags": [], "temporalCoverage": "2026-09-27T19:29:15.293681927+00:00", "title": "My Workflow for Understanding LLM Architectures", "url": "https://aggr.aymericbeaumet.com/items/magazine-sebastianraschka-com/2026-04-18-my-workflow-for-understanding-llm-architectures/", "word_count": 252 }