#llm

Articles tagged with #llm

One Answer, Four Different Scores: Learning LLM Evaluation the Hard Way
This week's material was about evaluating LLMs: offline metrics (tested against a fixed dataset) versus online metrics (measured from real user interaction). It sounds straightforward on the slides. O
Sep 25, 20268 min read6
Satu Jawaban, Empat Angka Berbeda: Belajar Evaluasi LLM dari Nol
Minggu ini materinya soal evaluasi LLM: offline metrics (diuji pakai dataset tetap) vs online metrics (diukur dari interaksi user sungguhan). Konsepnya kedengaran sederhana di slide. Begitu dicoba sen
Sep 25, 20267 min read1
Serving Your Own Model with llama.cpp: GPU Offload, Caching, and the Trap of a Stuck Wrong Answer
The model is fine-tuned. The next question: how do you actually use it — not inside a training notebook, but served like a real API, with a cache so the same question doesn't get recomputed over and o
Sep 25, 20267 min read1
Nge-serve Model Sendiri dengan llama.cpp: GPU Offload, Cache, dan Jebakan Jawaban yang Nyangkut
Model sudah selesai di-fine-tune. Langkah berikutnya: bagaimana cara benar-benar memakainya — bukan di notebook training, tapi disajikan seperti API sungguhan, dengan cache biar tidak menghitung ulang
Sep 25, 20266 min read2
What Does Fine-Tuning Actually Learn? Testing QLoRA with Deliberately Controlled Data
Fine-tuning is often treated as a way to "embed knowledge" into a model — but what actually happens under the hood can be far more layered than that. This post builds a small, controlled experiment —
Sep 25, 20267 min read4
Apa yang Sebenarnya Dipelajari Fine-Tuning? Menguji QLoRA dengan Data yang Sengaja Dikontrol
Fine-tuning sering dianggap sebagai cara "menanamkan pengetahuan" ke model — padahal yang sebenarnya terjadi bisa jauh lebih berlapis dari itu. Artikel ini membangun eksperimen kecil dan terkontrol —
Sep 25, 20267 min read1