mazafard / esco-gemma4

esco-gemma4

This model is a fine-tuned version of gemma-4 designed to map job descriptions and user inputs to official ESCO (European Skills, Competences, Qualifications and Occupations) taxonomies.

The model was trained efficiently utilizing the unsloth library and merged into 16-bit format for optimal inference performance.

Usage (Inference)

You can load this model directly using unsloth or standard transformers for sequence generation. For best results, use the identical prompt formatting applied during fine-tuning (see the provided inference.py script).

Loading with Unsloth

from unsloth import FastLanguageModel

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name="mazafard/esco-gemma4",
    max_seq_length=2048,
    dtype=None,
    load_in_4bit=True,
)
FastLanguageModel.for_inference(model)

Training Framework

• Base Model : Google Gemma 4 (4-bit)
• Fine-Tuning : LoRA / PEFT
• Engine : Unsloth