#qlora

Fine-tuning an open 7B LLM into a GST question-answering assistant, with a reproducible eval harness measuring the gain over the base model.

Production-grade LLM fine-tuning with LoRA/QLoRA — dataset synthesis, rigorous eval, cost analysis

Kendi fine-tune ettiğim Türkçe LLM — Qwen2.5-3B + QLoRA (Unsloth), Kaggle T4'te eğitildi, Ollama ile yerelde çalışıyor

Archive of small projects and exercises done while following the [AI Engineer Core Track: LLM Engineering, RAG, QLoRA, Agents] Udemy Course

Self-hosted local-AI stack over Ollama: private chat, a structured-output API, and a QLoRA fine-tune → serve → chat loop. No cloud, no secrets.

Does an LLM's willingness to blackmail depend on its persona — and on whether that persona is prompted or trained into the weights? Two-part study on Gemma-3-12B: 16 prompted personas (n=100) + 10 QLoRA persona adapters, in the Lynch et al. shutdown scenario.

Reproducible QLoRA SFT and DPO experiments for open-weight LLM reasoning, with locked selection, matched controls, and audit-grade provenance.

Serve DeepSeek and Qwen, run an on-demand model library, and fine-tune with Unsloth QLoRA on 2x NVIDIA DGX Spark. TP=2 vLLM lanes behind one OpenAI-compatible endpoint for OpenCode, Cursor, and Hermes. Honest benchmarks, committed artifacts.