English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Tools annual

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Finetuned large language models (such as ChatGPT and Qwen-chat) can generateChinese classical poetry following human's instructions. LLMs perform well incontent, but are usually lacking in format, with occasionally excess orinsufficient number of characters in each line. Since most SOTA LLMs aretoken-based, we assume that the format inaccuracy is due to the difficulty ofthe "token planning" task, which means that the LLM need to know exactly howmuch characters are contained in each token and do length-control planningbased on that knowledge. In this paper, we first confirm our assumption byshowing that existing token-based large language models has limited knowledgeon token-character relationship. We use a spelling bee probing procedure, andfind that Qwen-chat failed in nearly 15% Chinese spelling test. We then showthat a token-based model can be easily tailored into a token-free model (interms of Chinese), which can largely solve the format accuracy problem. Ourtailoring procedure removes long-tokens from the vocabulary and the languagemodel head, and keeps only character-level or byte-level tokens. As part of ourcontribution, we release the finetuned token-free model (which is based onQwen-chat-7B), which can generate chinese classical poetry following complexinstructions like LLMs (such as story paraphrasing), and also perform well informat. On the test set, our token-free model achives an format accuracy of0.96, compared to 0.84 for token-based equivalents and 0.38 for GPT-4.

Token-free LLMs Can Generate Chinese Classical Poetry with More Accurate Format