julienp79/ occitan-gemma-4-e2b-it-lora-sfttrainer:Q8_0

4 yesterday

Gemma 4 2B fine-tuned on Occitan via standard LoRA (r=16) with SFTTrainer. Q2_K, Q4_K_M, Q5_K_M, Q8_0, f16 quants available. 

ollama run julienp79/occitan-gemma-4-e2b-it-lora-sfttrainer:Q8_0

Details

2 days ago

7c70e4b15b6a · 5.0GB ·

model
gemma4
·
4.65B
·
Q8_0
system
Ès un escritor e grammarista occitan lengadocian. Respon unicament en occitan lengadocian. Escriu d
params
{ "num_ctx": 768, "stop": [ "<bos>", "<|turn>", "<turn|>", "
template
{{ if .System }}<bos><|turn>system {{ .System }}<turn|> {{ end }}{{ if .Prompt }}<|turn>user {{ .Pro

Readme

Occitan Lengadocian — Gemma 4 E2B LoRA

Fine-tune of Gemma 4 E2B Instruct on Occitan in the Lengadocian dialect (IEO grafia classica norm) using standard LoRA.

Reliable entry point. Earlier model in the E2B series, superseded on most metrics by the RS-LoRA r=32 version but still a solid and fast model. Good literary vocabulary with consistent Lengadocian register.

Strengths

Produces clean literary prose with authentic vocabulary (roginassa, esparnhament de lum, plumegàs, nolor de frescum, refolimentit). Fast inference, smallest footprint in the collection.

When to prefer the RS-LoRA version

If you have the same hardware, the RS-LoRA r=32 model (occitan-gemma-4-e2b-it-rslora-sfttrainer) outperforms this one on all evaluated tasks and is the recommended E2B model.

System prompt

Ès un escritor e grammarista occitan lengadocian. Respon unicament en
occitan lengadocian. Escriu dirèctament lo tèxte demandat, sens cap
d'introduccion, de comentari ni d'explicacion sus ton trabalh. Pas de
preamble. Pas de version multiplas. Pas de traduccion.

Recommended quantisation

Quant Size Use case
Q4_K_M ~1.6 GB Recommended
Q5_K_M ~1.9 GB Slightly better quality
Q8_0 ~2.5 GB Near-lossless
Q2_K ~1.1 GB Minimal RAM

Training

LoRA · r=16 · α=32 · block_size=384 · 2535 steps · 5 epochs
RTX 3060 12GB · ~3h · FastVisionModel + SFTTrainer