Redirecting to original paper in 30 seconds...

Click below to go immediately or wait for automatic redirect

arxiv_cl 80% Match Research Paper / Competition Submission Participants in NLP Competitions,LLM Researchers,Machine Learning Engineers 2 weeks ago

Finetuning LLMs for EvaCun 2025 token prediction shared task

large-language-models β€Ί evaluation
πŸ“„ Abstract

Abstract: In this paper, we present our submission for the token prediction task of EvaCun 2025. Our sys-tems are based on LLMs (Command-R, Mistral, and Aya Expanse) fine-tuned on the task data provided by the organizers. As we only pos-sess a very superficial knowledge of the subject field and the languages of the task, we simply used the training data without any task-specific adjustments, preprocessing, or filtering. We compare 3 different approaches (based on 3 different prompts) of obtaining the predictions, and we evaluate them on a held-out part of the data.
Authors (2)
Josef Jon
OndΕ™ej Bojar
Submitted
October 17, 2025
arXiv Category
cs.CL
arXiv PDF

Key Contributions

This paper presents a submission for the EvaCun 2025 token prediction task, detailing the fine-tuning of three LLMs (Command-R, Mistral, Aya Expanse) on the provided data. It compares three different prompting approaches for prediction and evaluates their performance.

Business Value

Demonstrates practical application of LLMs in competitive settings and provides insights into effective fine-tuning and prompting strategies for specific NLP tasks.

Paper Metadata

Innovation Type

Application/Experimentation

Deployment Feasibility

High, as it focuses on applying existing LLMs and fine-tuning techniques.

Limitations Addressed

Addresses the challenge of adapting general-purpose LLMs to a specific, potentially unfamiliar task (token prediction) with limited domain knowledge and without extensive task-specific preprocessing.

Technical Tags

token predictionLLM fine-tuningEvaCun 2025Command-RMistralAya Expanseprompting strategiestask-specific adjustments

Research Topics

Natural Language ProcessingLarge Language ModelsMachine Learning CompetitionsModel Fine-tuning

Methods & Architectures

Fine-tuning LLMsPrompting strategies (3 variations)Held-out data evaluation Command-RMistralAya Expanse

Applications & Tasks

Token Prediction Task NLP Competitions Adapting LLMs to Specific TasksEvaluating Prompting Strategies Token PredictionEvaluating LLM Performance on a Shared Task

Datasets & Benchmarks

Datasets
EvaCun 2025 task data

Related Fields

Natural Language ProcessingMachine LearningCompetitions

Keywords

Large Language ModelsToken PredictionFine-tuningPromptingEvaCun 2025Command-RMistralAya ExpanseNLPMachine Learning Competition

Academic Context

#Natural Language Processing#Large Language Models#Machine Learning Competitions#Model Fine-tuning

Commercial Potential

Potential Products
Specialized models for token prediction tasksTools for LLM fine-tuning and evaluation
Target Industries
TechnologyResearch
Use Case Examples
Participating in and winning NLP challengesDeveloping models for specific language prediction tasks
Competitive Edge
Represents a practical application and evaluation of leading LLMs on a specific NLP task within a competitive framework.
Market Opportunity
Active market for NLP competitions and benchmark datasets.
Revenue Models
Prize money from competitionspotential for licensing fine-tuned models.

Resource Requirements

Compute Needs
GPU resources for fine-tuning and inference of LLMs.
Data Requirements
Task-specific training data (EvaCun 2025 dataset).
Deployment Constraints
Computational resources for fine-tuning large models.
Scalability
Scales with the underlying LLM architectures used.

Production Readiness

Maturity Level

Research/Application

Time to Market

Short (for applying existing models)

Patent Potential

Low

View Full Paper Back to Papers