Instruction tuning vs domain adaptation: two different fine-tuning goals

Presenting KrishGyan, farming advice in your voice and language
Presenting KrishGyan, farming advice in your voice and language

Originally published at https://pranjulrathour.scult.in/blog/instruction-tuning-vs-domain-adaptation. That copy is the canonical version and gets updates first.

"Fine-tune it on our data" hides two different requests. One is: make the model behave the way we want when asked. The other is: make the model fluent in our field. They need different datasets, different training setups and different tests. Confusing them is the most common design error I see in student projects.

Instruction tuning

The data is pairs — or conversations — of request and ideal response. The loss is computed on the response tokens. The goal is behaviour: follow the format, respect the constraints, respond in the house tone. A few thousand carefully written examples go a long way. This is what most QLoRA projects in FineTune Studio are.

At an Integral Startup Foundation hackathon
At an Integral Startup Foundation hackathon

Domain adaptation

The data is raw text from the field — manuals, papers, transcripts — and the loss is computed on all of it, the way the base model was pre-trained. The goal is vocabulary and fluency: the model stops stumbling over your acronyms. It needs far more text, and it does not by itself teach the model to answer questions politely. Usually you follow it with a short instruction-tuning stage.

How they differ in practice

  • Data volume — thousands of examples versus millions of tokens.
  • Loss masking — response-only versus full sequence.
  • Evaluation — task rubric on held-out instructions versus perplexity on held-out domain text, then task rubric after the instruction stage.
  • Risk — instruction tuning can overfit to phrasing; domain adaptation can erode general instruction-following if pushed too far.

What you probably want

If your problem is facts, neither — use retrieval, as argued in RAG vs fine-tuning. If your problem is behaviour, instruction tuning. If the model genuinely cannot read your domain's text, a modest domain-adaptation pass followed by instruction tuning. Decide which before you collect data, because the data you need is different.

Pranjul Rathour
Pranjul Rathour

Name the goal, and the recipe follows. Skip that step and you will train the wrong thing perfectly.

From my carousels
1,211 Prompts, Every One Stamped
1,211 Prompts, Every One Stamped, slide 11,211 Prompts, Every One Stamped, slide 2
1,211 Prompts, Every One Stamped, slide 31,211 Prompts, Every One Stamped, slide 4
Full carousel on Instagram and LinkedIn.
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 to a room
Presenting to a room
Pranjul Rathour
Pranjul Rathour
Pranjul Rathour at a campus event
Pranjul Rathour at a campus event

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

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

I built 15 free tools, 1,211 prompts and a 50,000-skill library — here's what's inside tools.scult.in

Image
Carousel · 12 slides I built every part of tools.scult.in myself — 15 browser tools, 1,211 version-stamped prompts, a 50,000+ agent-skills library, a public MCP server, and 40 retro photo prompts where I'm the face in every example. All of it free. No signup, no ads, no upload step. Swipe through for what's inside and why I built it this way. Link in the last slide. Slide 1 of 12 Slide 2 of 12 Slide 3 of 12 Slide 4 of 12 Slide 5 of 12 Slide 6 of 12 Slide 7 of 12 Slide 8 of 12 Slide 9 of 12 Slide 10 of 12 Slide 11 of 12 Slide 12 of 12 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. Email: pranjulrathour41@gmail.com Portfolio & ...
Read more