I will fine tune and train llms with pytorch and huggingface

I fine-tune open models on your data so they actually understand your terminology, your documents, and your use case.

What I do

• Fine-tune open-weights LLMs with LoRA or full fine-tuning, on PyTorch and HuggingFace
• Clean and structure your raw data into a usable training set usually the hardest part
• Build an evaluation harness so quality is measured, not guessed
• Export for production: ONNX, TorchScript, or a deployed inference API
• Run it on your infrastructure, so there are no per-token API costs

Recent work: fine-tuned a model on a client's internal documentation. Accuracy on their benchmark went from roughly 60% to over 90%, running on their own hardware.

How I work

I build the eval set before training anything. Without it there's no way to tell whether a run improved things or made them worse, and most fine-tuning projects skip this step.

I'll also tell you when fine-tuning isn't the answer sometimes RAG or better prompting gets you there for far less.

Message me with your use case and dataset size, and I'll tell you what's realistic.