System Report for CCL25-Eval Task 5: New Dataset and LoRA-Fine-Tuned Qwen2.5

2026-06-10 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Natural Language Processing · Depth: Advanced, quick

Summary

A new system report details the development of PoetryQwen, a domain-specialized large language model designed for classical Chinese poetry translation and emotional understanding. Addressing limitations in existing research that often neglects poetic appreciation's distinctive features and lacks high-quality domain-specific datasets, the task was decomposed into term interpretation, semantic interpretation, and emotional inference. Researchers constructed the Classical Chinese Poetry Instruction Pair Dataset (CCPoetry-49K), comprising 49,404 high-quality instruction-response pairs. PoetryQwen was created by applying Low-Rank Adaptation (LoRA) to fine-tune the Qwen2.5-14B model. Experimental results on the CCL25-Eval Task 5 benchmark show PoetryQwen achieved a score of 0.757, marking a 9.7% improvement over the Qwen2.5-14B-Instruct baseline's 0.690. This demonstrates a significant enhancement in precise translation and emotional understanding of classical poetry.

Key takeaway

For NLP Engineers developing specialized LLMs for culturally rich or niche domains, this work demonstrates a clear path to significant performance gains. You should consider decomposing complex tasks into subtasks and investing in constructing high-quality, domain-specific instruction datasets like CCPoetry-49K. Applying Low-Rank Adaptation (LoRA) to fine-tune a robust base model such as Qwen2.5-14B can yield substantial improvements, as shown by PoetryQwen's 9.7% benchmark increase.

Key insights

Domain-specific LoRA fine-tuning on Qwen2.5-14B with CCPoetry-49K significantly enhances classical Chinese poetry understanding and translation.

Principles

• Task decomposition aids domain-specific LLM optimization.
• High-quality, domain-specific datasets are critical.
• LoRA fine-tuning effectively adapts LLMs.

Method

The task was decomposed into term, semantic, and emotional interpretation. CCPoetry-49K, with 49,404 instruction-response pairs, was constructed. LoRA fine-tuning was then applied to the Qwen2.5-14B model to create PoetryQwen.

In practice

• Utilize CCPoetry-49K for classical poetry tasks.
• Apply LoRA to Qwen2.5-14B for domain adaptation.
• Decompose complex tasks for LLM specialization.

Topics

• Large Language Models
• Classical Chinese Poetry
• LoRA Fine-tuning
• Qwen2.5
• Domain Adaptation
• Instruction Datasets

Best for: AI Engineer, Research Scientist, AI Scientist, Machine Learning Engineer, NLP Engineer

Related on AIssential

• System Report for CCL25-Eval Task 5: New Dataset and LoRA-Fine-Tuned Qwen2.5 — Domain-specific datasets and LoRA fine-tuning significantly enhance LLM performance in specialized tasks like classical poetry analysis.
• The Evolution of Adaptation: From Full Fine-Tuning to LoRA Mastery — LoRA and QLoRA democratize LLM fine-tuning by drastically reducing computational and memory requirements.
• How Small Can You Go? LoRA Fine-Tuning 270M-8B Models for Merchant Information Extraction in Financial Transactions — Smaller LoRA-fine-tuned models can achieve near 8B performance for structured extraction, significantly reducing deployment costs.
• I Fine-Tuned an LLM to Reply Like My Wife — 10 Years of Chats, One Adapter, and Results That Amazed… — Fine-tuning LLMs with personal, messy data on consumer hardware can yield surprising persona learning.
• Modular Monolingual Adaptation using Pretrained Language Models — Modular PLM adaptation, freezing embeddings while tuning the rest, improves low-resource language performance.
• Fine-Tuning Qwen3.5 — Fine-tuning small vision-language models like Qwen3.5-0.8B on domain-specific datasets significantly improves specialized task performance.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.