MikCil/PREMOVE_llama3.3-70b_float16

Hugging Face
TEXT GENERATIONPricing:Input $2.88 / Output $2.88Concurrent Unit Cost:4Model Size:70BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jan 11, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Warm

MikCil/PREMOVE_llama3.3-70b_float16 is a 70 billion parameter Llama 3.3 model, developed by Michele Ciletti, specifically fine-tuned for Word-Sense Disambiguation (WSD) of Latin and Ancient Greek preverbs. This model achieves a Micro F1 score of 0.8447 on this specialized task, demonstrating high precision and recall for disambiguating preverbed motion verbs in historical languages. It was trained using Unsloth and Huggingface's TRL library, optimizing for efficient fine-tuning.

Loading preview...

MikCil/PREMOVE_llama3.3-70b_float16 Overview

This model, developed by Michele Ciletti, is a 70 billion parameter Llama 3.3 variant specifically fine-tuned for Word-Sense Disambiguation (WSD) of Latin and Ancient Greek preverbs. It addresses the complex task of identifying the correct meaning of preverbed motion verbs in these historical languages.

Key Capabilities & Performance

  • Specialized WSD: Excels at disambiguating preverbs in Latin and Ancient Greek.
  • Performance Metrics: Achieves a Precision of 0.8538, Recall of 0.8358, and a Micro F1 score of 0.8447 on its target task.
  • Efficient Fine-tuning: The model was fine-tuned from unsloth/llama-3.3-70b-instruct-unsloth-bnb-4bit using Unsloth and Huggingface's TRL library, enabling faster training.

Research Context

The development and evaluation of this model are detailed in the paper "Probing Preverbs: Evaluating Large Language Models on Latin and Ancient Greek Preverbed Motion Verbs" by Farina, A., & Ciletti, M. (2025), presented at Historical Languages and AI. Further details and resources are available on the associated GitHub repository.

Ideal Use Cases

  • Classical Philology: Researchers and linguists working with Latin and Ancient Greek texts requiring precise semantic analysis of preverbs.
  • Digital Humanities: Applications involving automated analysis or annotation of historical language corpora.
  • Specialized NLP: Tasks that demand high accuracy in word-sense disambiguation for highly inflected and complex linguistic structures.

Finetunes
1 models