Browse categories
Explore
Fiverr Pro
English
$
USD
I will fine tune an llm on your data with lora and full evaluation
Adarsh SM
About this gig
Before you pay anyone to fine-tune a model: about half the projects that reach me don't need one. Better prompting or retrieval solves it for far less, and I'll tell you that for free rather than take the order.
When fine-tuning IS the answer - domain vocabulary, a specific output format, a tone the base model won't hold, or cutting inference cost - here's what I do.
I'm an ML and Generative AI engineer with 4+ years in production AI. I've fine-tuned GPT and BERT-family models to 30% higher contextual accuracy, and I currently work on a ~30B-parameter foundation model.
WHAT YOU GET
An honest read on whether fine-tuning beats prompting
Dataset design, cleaning and formatting, where most projects actually fail
LoRA / QLoRA or full fine-tuning on open models
An eval harness that measures what you care about, not loss curves
Before/after numbers on a held-out set
Weights, inference code and quantized artifacts you own
MODELS
Llama, Mistral, Qwen, Gemma, Phi, plus OpenAI hosted fine-tuning · PyTorch, Hugging Face, PEFT, Unsloth, Axolotl
Message me with what the model needs to do and what data you have, even if it's messy or you're not sure there's enough.
Get to know Adarsh SM
Adarsh SM
GenAI Engineer for RAG Chatbots, LLM Fine tuning and AI Agents
- FromIndia
- Member sinceMay 2025
- Avg. response time1 hour
Languages
English, Hindi, Kannada, Telugu
I build AI that works in production, not just in demos. I'm an ML & GenAI engineer with 5+ years of experience shipping RAG chatbots, fine-tuned LLMs, computer vision and AI agents on AWS, GCP and Azure. Past work: AI systems for DRDO, a healthcare imaging model (95% accuracy) and finance automation (40% less manual work). You get clean, documented code, deployment and a walkthrough call. Message me before ordering: I reply within an hour and will tell you honestly if AI is the right fit.
My Portfolio
FAQ
How much data do I need?
For style, format or tone, a few hundred good examples often works. For real domain knowledge, think thousands. If you're short, Premium includes synthetic data generation. Order Basic first and I'll tell you exactly where you stand.
Who pays for GPU compute?
Basic needs none. For Standard and Premium, either you provide cloud credits or I run it and bill compute at cost, typically $20-100 depending on model size. I'll quote it before starting, never after.
Do I own the resulting model?
Entirely. You get the weights, the training config and the code. Note that the base model's own licence still applies, and I'll flag any commercial-use restriction before we pick one.
