DeepSeek V4.1 Flash: 552B parameters, 8B awake

Carousel · 9 slides

DeepSeek V4.1 Flash: 552B parameters, but only 8B active per token while reading (16B while writing).

→ 384 experts per layer, 6 fire per token
→ KV cache: 890 bytes per token, 437× smaller than DeepSeek V1
→ 1M-token context, trained on 45T tokens
→ Self-reported 74.2% on DeepSWE v1.1, level with Opus 5.0 and GPT-5.6 Sol
→ MIT license, 748K downloads in 30 days

And the 2-bit build runs fully in memory on two 128 GB Macs.

I'm Pranjul Rathour, GenAI engineer from Kanpur. Want me to judge your hackathon or speak at your college? pranjulrathour.scult.in/invite

DeepSeek V4.1 Flash: 552B parameters, 8B awake, slide 1 of 9
Slide 1 of 9
DeepSeek V4.1 Flash: 552B parameters, 8B awake, slide 2 of 9
Slide 2 of 9
DeepSeek V4.1 Flash: 552B parameters, 8B awake, slide 3 of 9
Slide 3 of 9
DeepSeek V4.1 Flash: 552B parameters, 8B awake, slide 4 of 9
Slide 4 of 9
DeepSeek V4.1 Flash: 552B parameters, 8B awake, slide 5 of 9
Slide 5 of 9
DeepSeek V4.1 Flash: 552B parameters, 8B awake, slide 6 of 9
Slide 6 of 9
DeepSeek V4.1 Flash: 552B parameters, 8B awake, slide 7 of 9
Slide 7 of 9
DeepSeek V4.1 Flash: 552B parameters, 8B awake, slide 8 of 9
Slide 8 of 9
DeepSeek V4.1 Flash: 552B parameters, 8B awake, slide 9 of 9
Slide 9 of 9
Pranjul Rathour
Pranjul Rathour
GenAI engineer, Kanpur · 3x first-prize hackathon winner · campus mentor
I ship production RAG pipelines, fine-tune LLMs and build agentic AI products end to end. I lead engineering at SCULT INDIA for a 14-member team and have mentored 200+ students through TechVerse Enclave.
Open to: GenAI roles, hackathon judging, mentorship sessions and guest talks at colleges.
On stage, at hackathons and on campus
Presenting BrandHive
Presenting BrandHive
A career session for students
A career session for students
Presenting to a room
Presenting to a room

Comments

Post a Comment

Popular posts from this blog

Forming a hackathon team: roles, skills and the mistake most teams make

Image
At an Integral Startup Foundation hackathon Originally published at https://pranjulrathour.scult.in/blog/hackathon-team-roles-and-formation . That copy is the canonical version and gets updates first. Team TechVerse — Raman Shukla, Saksham Gupta and me — won three first prizes in 2025 not because we were the strongest coders in the room, but because we had stopped overlapping. When I mentor teams before a hackathon, formation is the first thing I fix. In a packed college auditorium The four roles Builder — owns the core feature and the demo path. Usually backend or the AI piece. Interface — owns what the judge sees: the front end, the flow, the seeded data. Storyteller — owns the problem statement, the pitch, the slides and the Q&A plan; also does user research in hour one. Integrator — keeps the pieces working together, manages the repo, runs the demo rehearsals. In a team of three, the builder or interface person takes this. The mistake Five friends who all write R...
Read more

Reward modeling, explained simply: what it is and why most student projects don't need it

Image
At an Integral Startup Foundation hackathon Originally published at https://pranjulrathour.scult.in/blog/reward-modeling-basics-for-student-projects . That copy is the canonical version and gets updates first. A reward model is trained to score how good a response is, so it can guide further training of another model — the mechanism behind RLHF-style alignment. It's genuinely complex to build well, and most student projects don't need to build one. What it actually requires A dataset of paired responses with human or automated preference judgements — which output is better, and why. A separate training run to fit a model that predicts that preference, before it's ever used to guide anything else. Careful evaluation of the reward model itself, since a flawed reward model teaches the wrong thing to whatever it later guides. When a student project actually needs this Almost never, directly. Supervised fine-tuning or DPO on direct preference pairs gets most projects m...
Read more

Hello from Kanpur: what I build, and what I'll write about here

Image
On the mic Originally published at https://dev.to/pranjulrathour/hello-from-kanpur-what-i-build-and-what-ill-write-about-here-f2j . That copy is the canonical version and gets updates first. I'm Pranjul Rathour, a GenAI engineer from Kanpur, India. This is the first post on this blog, so here is who is writing, what I have actually shipped, and what you can expect if you follow along. What I build Between February and July 2026, at Next Upgrad Web Solutions, I designed, built and shipped five production AI applications end to end, alone, and deployed all of them with Docker, CI and observability: RAG.NextUpgrad : a retrieval-augmented generation service with hybrid dense plus BM25 retrieval, cross-encoder reranking, streaming cited answers, and a confidence gate that refuses to call the model when the evidence is weak. It runs in about 220 MB of memory, inside a free 512 MB instance. FineTune Studio : fine-tune open models with QLoRA in a browser. Upload a dataset, validate ...
Read more