From b333b06772c89d96aacb5490d6a219fba7c09cc6 Mon Sep 17 00:00:00 2001
From: Mitja Felicijan <mitja.felicijan@gmail.com>
Date: Thu, 12 Feb 2026 20:57:17 +0100
Subject: Engage!

---
 llama.cpp/examples/training/README.md | 17 +++++++++++++++++
 1 file changed, 17 insertions(+)
 create mode 100644 llama.cpp/examples/training/README.md

(limited to 'llama.cpp/examples/training/README.md')

diff --git a/llama.cpp/examples/training/README.md b/llama.cpp/examples/training/README.md
new file mode 100644
index 0000000..df42527
--- /dev/null
+++ b/llama.cpp/examples/training/README.md
@@ -0,0 +1,17 @@
+# llama.cpp/examples/training
+
+This directory contains examples related to language model training using llama.cpp/GGML.
+So far finetuning is technically functional (for FP32 models and limited hardware setups) but the code is very much WIP.
+Finetuning of Stories 260K and LLaMA 3.2 1b seems to work with 24 GB of memory.
+**For CPU training, compile llama.cpp without any additional backends such as CUDA.**
+**For CUDA training, use the maximum number of GPU layers.**
+
+Proof of concept:
+
+``` sh
+export model_name=llama_3.2-1b && export quantization=f32
+./build/bin/llama-finetune --file wikitext-2-raw/wiki.test.raw -ngl 999 --model models/${model_name}-${quantization}.gguf -c 512 -b 512 -ub 512
+./build/bin/llama-perplexity --file wikitext-2-raw/wiki.test.raw -ngl 999 --model finetuned-model.gguf
+```
+
+The perplexity value of the finetuned model should be lower after training on the test set for 2 epochs.
-- 
cgit v1.2.3