Multilingual Text Generation With 128k Context

via “multilingual text generation with language-specific tokenization”

text-generation model by undefined. 1,00,53,835 downloads.

Unique: Uses a unified SentencePiece tokenizer trained on mixed-language corpus, enabling efficient multilingual generation without language-specific branches; Qwen3 specifically optimizes for Chinese-English code-switching through instruction-tuning on bilingual examples

vs others: Better Chinese support than Llama 3.2 or Mistral due to native training on Chinese data; more efficient than separate monolingual models due to shared parameters, though with slight quality tradeoff vs language-specific models