I will fine tune your llm on custom data with lora and qlora

Fine-tuning is worth it when a model needs to speak in your domain, follow your format, or handle your terminology consistently. It is not worth it when you just need the model to look things up. I will tell you which one you actually need before you spend anything.

What I do:

LoRA and QLoRA fine-tuning of open models: Llama, Mistral, Qwen, Gemma, Phi

Full fine-tuning and instruction tuning when your data justifies it

OpenAI fine-tuning through their API

Domain adaptation for medical, legal, financial and technical text

RAG systems when retrieval is the better answer

Training data preparation, which is usually where projects succeed or fail

Stack: Hugging Face Transformers, PEFT, Unsloth, Axolotl, PyTorch, Weights and Biases for tracking.

You receive the trained weights hosted where you want them, the training script with full documentation, evaluation results comparing your tuned model against the base model so you can see what changed, an inference script, and a guide for retraining when your data grows.

You need roughly 50 to 100 examples for a light OpenAI fine-tune, or 500 or more for open models. Quality beats quantity, and I will help you build the set.