Lablup open-sources GLM-5.2-NVFP4-DFlash-ko, a draft model to speed up GLM-5.2 inference in Korean (report)

Verified (fact)

Lablup said on the 12th that it had open-sourced GLM-5.2-NVFP4-DFlash-ko, a draft model that speeds up LLM inference for Korean-language queries and Korean agent tasks, on the 9th, Hangul Day. The model is used alongside Z.ai's GLM-5.2 when it runs in NVIDIA's 4-bit floating-point format (NVFP4), and is built on the DFlash method released by z-lab. Draft models are used in speculative decoding, where a small model first proposes candidate tokens and a large model verifies them in a batch to speed up generation. Specific speedup figures are not given in the report summary.

Needs fact-check (claim)

Lablup claims: faster responses for Korean-language LLMs.

Why it matters (interpretation)

Korean-language processing speed and cost are obstacles when serving large open models directly in Korea. A publicly available draft model tuned for Korean lowers the burden of using open-weight models such as GLM-5.2 for Korean-language services.

Sources

• Press 래블업, '한국어 AI 추론 속도 향상' 드래프트 모델 공개 Electronic Times (Korea)