Sebastian Raschka

Summary: Staff Research Engineer at Lightning AI. Author of "Python Machine Learning," "Machine Learning with PyTorch and Scikit-Learn," "Build a Large Language Model (from Scratch)." PhD from Michigan State University. Writes exceptional technical deep-dives (blog, Substack). Published detailed explainer on DeepSeek DSA. Strong educator + practitioner — potential PhD advisor or industry mentor.


Profile

Field Details
Current Role Staff Research Engineer, Lightning AI
PhD Michigan State University (Computer Science / ML)
Key Books "Python Machine Learning" (3rd ed), "Machine Learning with PyTorch and Scikit-Learn", "Build a Large Language Model (from Scratch)"
Blog/Substack sebastianraschka.com, magazine.sebastianraschka.com
Twitter/X @rasbt
GitHub rasbt
Research Interests LLM training, efficient attention, ML education, PyTorch, Lightning

Significance for You

Multiple connection points:

Connection Relevance
DSA explainer author Your DSA-ViT project directly builds on DSA; his blog is a reference
Lightning AI Industry lab with PhD-style research; potential internship/collab
Educator + practitioner Rare combo — great for learning how to communicate research
Book author (LLM from Scratch) Your BSAI background matches his teaching style
PhD from MSU Academic track → industry research; good perspective on both paths

Target tier: Tier 2-3 (industry research, not traditional PhD advisor, but excellent mentor/network node)


Key Technical Content (from DSA blog)

His DeepSeek DSA blog covers:

  1. Indexer — Lightning indexer: I_{t,s} = sum w^I_{t,j} · ReLU(q^I_{t,j} · k^I_s)
  2. Token Selector — Top-k on index scores: S_t = {c_s mid I_{t,s} ∈ Top-k}
  3. Core Attention — MQA on selected tokens
  4. MLA Integration — DSA instantiated under MLA MQA mode (shared latent KV)
  5. Training Stages — Dense warmup (KL on indexer) → Sparse training (KL on selected + LM loss)

Your DSA-ViT project can cite this as accessible reference.


Contact Strategy

When: Anytime (active on X, responds to thoughtful replies) How: X reply / DM / email (listed on website) Angle:

  • Reference his DSA blog in your project
  • Ask specific technical question about DSA→ViT adaptation
  • Share your DSA-ViT project page
  • Lightning AI internship inquiry (Summer 2025)

Not for: Direct PhD supervision (industry role), but excellent for:

  • Industry mentorship
  • Internship referral
  • Technical feedback
  • Networking to academic collaborators

Related Wiki Pages

  • dsa-attention — Concept page (his blog is a source)
  • paper-deepseek-v3-2 — Original DSA paper
  • vision-transformer-dsa-integration — Your project applying DSA to ViT
  • lightning-ai — His employer (create if needed)
  • target-labs — Your PhD target list

Sources

  • web-sebastianraschka-dsa: Blog post "DeepSeek Sparse Attention (DSA) Explained"