LoRA Fine-Tuning Tiny LLMs as Expert Agents

Comments (3)

jamesbriggs · 1d ago

Sharing my walkthrough on fine-tuning LLMs with LoRA using NVIDIA's NeMo microservices. The result is a llama-3.2-1b-instruct model fine-tuned to be really good at function-calling, making it ideal for agent-use.

It was a ton of fun to figure it out and it brought back some nostalgia from the days of training ML models, tweaking learning rates, dropout, and watching loss charts in W&B.

Final performance was way better than any 1-3B parameter LLM I tried with agentic workflows in the past.

kordlessagain · 1d ago

Thank you for making this. I clicked through on the container page on the cookbook/gen-ai/training/lora/nvidia-nemo /nemo-lora-function-calling.ipynb and it was a 404. I did find this: https://catalog.ngc.nvidia.com/orgs/nim/teams/meta/container...

Can you point to a public version of this model you trained. I'd like to test with an agentic framework I'm working on.

jamesbriggs · 1d ago

My bad, the link was wrong - you found the right one. I've updated it in the repo too, thanks. Let me know how it goes!

Tell HN: Namecheap pre-purchasing searched domain names?

Ask HN: Arc is dead, where should we move now?

Tasks Per Day – A minimalist productivity app that works

Ask HN: What are you working on? (May 2025)

Ask HN: What projects do you donate to?

Ask HN: How are people using ChatGPT to increase productivity in personal life?

Ask HN: Management wants to talk to my Datalake. What's the best way to do this?

Can we take a moment to appreciate what kind of web experience we are building?

Ask HN: Career Plateau: Looking for Advice on How to Break Through

Ask HN: What's the best tool you've used for sprint planning in 2025?

Ask HN: Are AIs intentionally weak at debugging their code?

Ask HN: Ban streaming services from playing the same commercial constantly?

Ask HN: Anyone struggling to get value out of coding LLMs?

Google will delete OAuth clients falsely flagged as unused

Al-LLM powered eBPF based security platform

Ask HN: Why that many more US-based companies are hiring "US-only" remote?

Best Buy is selling a $400 "digital Ethernet" cable for "cleaner, clearer sound"

Ask HN: Is anyone using AI conversation partners?

Ask HN: Am I old?

Reverse engineering Twitter's new WASM-based "X-XP-Forwarded-For" antibot header

Ask HN: Go deep into AI/LLMs or just use them as tools?

Ask HN: Building LLM apps? How are you handling user context?

Ask HN: How do I start my own cybersecurity related company?

Lost in AI-translation

Ask HN: Any modern mobile friendly family tree / genealogy builders?

SQLite is 25 years old today

What's your "coworker didn't lock screen" procedure?

Ask HN: Is this Tetris variant novel? Would it be fun?

Ask HN: What HN posts inspired or changed your perspective the most?

Ask HN: Would you be interested in OpenRouter for MCPs?

What's been your biggest technical bottleneck as a small startup lately?

Ask HN: Merge Request Review Agent?

It Survives Because It Collapsed: Three AIs, One Residue

LoRA Fine-Tuning Tiny LLMs as Expert Agents

Comments (3)