Downloads · 30 days
7
8% of all-time downloads
yusraghazi/leesplank-noot-llama-3.2-3b-dpo-finetuned
leesplank-noot-llama-3.2-3b-dpo-finetuned is a text generation model from yusraghazi. Use it when you need the model to write or continue text. It is set up for transformers. The card lists the license as cc-by-nc-4.0.
What is this? A Dutch text-simplification model tuned towards clear, accessible B1-level language in the Leesplank Noot style (clear and simple, but not childish).
Downloads · 30 days
7
8% of all-time downloads
All-time downloads
85
Public
Parameters
3.2B
6.4 GB on disk
Likes
1
Public
Click a slice to open those files.
.safetensors6.4 GB · 100%
From the Hugging Face model README
What is this? A Dutch text-simplification model tuned towards clear, accessible B1-level language in the Leesplank Noot style (clear and simple, but not childish).
This repo contains a merged checkpoint (base model + trained adapter merged into one set of weights), so you can load it like a regular transformers model.
from transformers import AutoTokenizer, AutoModelForCausalLM
model_id = "yusraghazi/leesplank-noot-llama-3.2-3b-dpo-finetuned"
tok = AutoTokenizer.from_pretrained(model_id, use_fast=True)
mdl = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", torch_dtype="auto")
prompt = "Vereenvoudig: Indien u zich toch in dit gebied begeeft, bent u verplicht dit te melden aan de lokale autoriteiten."
inputs = tok(prompt, return_tensors="pt").to(mdl.device)
out = mdl.generate(**inputs, max_new_tokens=120, do_sample=False)
print(tok.decode(out[0], skip_special_tokens=True))
System:
Je bent een AI-assistent die Nederlandse teksten vereenvoudigt naar een helder, toegankelijk niveau voor iedereen, vergelijkbaar met de heldere taal die het Jeugdjournaal gebruikt. Behoud de betekenis en belangrijke informatie, maar gebruik eenvoudigere woorden en kortere zinnen. Schrijf niet kinderlijk, maar wel toegankelijk.
User:
Vereenvoudig:
If this repo includes a chat_template.jinja, you can use that as well.
Base model: UWV/leesplank-noot-llama-3.2-3b
On top of the base model, preference alignment with DPO was applied on prompt/chosen/rejected triplets. This teaches the model to prefer:
…over common failure modes such as information loss, meaning drift, hallucinations, and overgeneralization.
Trained on a Dutch DPO preference dataset with columns:
prompt: original sentence/textchosen: preferred simplificationrejected: less preferred (errorful/weaker) simplificationerror_type (optional): label describing the type of error in rejectedDataset: yusraghazi/INT-Extended-DPO-v1
trl.DPOTrainerbeta = 0.1Training used 4-bit quantization (QLoRA) + a LoRA adapter:
nf4, double quantization enabledr=16, lora_alpha=32, lora_dropout=0.05, bias="none", task_type="CAUSAL_LM"bf16 if supported, otherwise fp1625e-52 (per device)82565120.03paged_adamw_32bit42This release was primarily qualitatively evaluated on held-out examples (base vs tuned), compared against chosen.
Recommended next step: add reproducible metrics (e.g., SARI + meaning-preservation checks) on a fixed benchmark set.
Good for:
Not recommended for:
Even after DPO, the model can sometimes:
Keep a human in the loop when accuracy matters.
INT-Extended-DPO-v1)If you use this model, please cite/mention:
UWV/leesplank-noot-llama-3.2-3byusraghazi/INT-Extended-DPO-v1Maintainer: Yusraghazi
Feedback: open an issue/discussion on this Hugging Face repo.