Downloads · 30 days
0
jekat/Ministral-3-3B-BF16-LT-Morphology-Accentuation-LoRA
Ministral-3-3B-BF16-LT-Morphology-Accentuation-LoRA is a machine learning model from jekat. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for transformers. The card lists the license as mit.
LoRA adapters for Lithuanian morphological analysis and accentuation tasks, trained on top of Ministral-3-3B-Instruct-2512 in BF16.
Downloads · 30 days
0
Access
Public
Updated May 19, 2026
Repo size
1.2 GB
Likes
0
Public
Click a slice to open those files.
.safetensors1.1 GB · 88%
From the Hugging Face model README
LoRA adapters for Lithuanian morphological analysis and accentuation tasks, trained on top of Ministral-3-3B-Instruct-2512 in BF16.
This work is part of a bachelor's thesis at Vilnius University, Faculty of Mathematics and Informatics.
| Adapter | Rank | Data Source | Training Examples | Description |
|---|---|---|---|---|
r16 | 16 | LRT, Wikisources | 7k | Small diverse dataset, standard LoRA |
r64 | 64 | BLKT | 142k | Overfitted — included for research purposes |
r32-blkt | 32 | BLKT | 142k | Single-source dataset, rsLoRA |
r32-mixed | 32 | LRT, Wikisources, BLKT | 149k | Best adapter — mixed dataset, rsLoRA |
| Adapter | Accuracy |
|---|---|
| Base Ministral 3B | 8.0% |
| r16 | 62.0% |
| r64 | 7.4% |
| r32-blkt | 60.3% |
| r32-mixed | 80.0% |
| GPT-5.5 | 13.9% |
| Gemini 3.1 Pro | 13.0% |
| Adapter | Lemma | POS | Features | Overall |
|---|---|---|---|---|
| Base Ministral 3B | 18.9% | 39.8% | 42.7% | 33.7% |
| r16 | 37.4% | 50.7% | 62.6% | 48.6% |
| r64 | 0.0% | 0.0% | 0.0% | 0.0% |
| r32-blkt | 34.7% | 42.7% | 64.8% | 42.0% |
| r32-mixed | 46.4% | 57.7% | 64.4% | 56.5% |
| GPT-5.5 | 36.1% | 49.1% | 59.8% | 48.4% |
| Gemini 3.1 Pro | 32.7% | 42.7% | 51.1% | 42.2% |
The adapters were trained on three tasks: accentuation, morphological analysis (CoNLL-U and Jablonskis formats), and lemmatization. Prompts were given in both Lithuanian and English.