Open Models Almanac
Methodspost-trainingsupervised fine-tuning

single mention · post-training

Joint multimodal supervised fine-tuning

The final stage performs supervised fine-tuning on multimodal instruction-following data, analogous to the language-only SFT pipeline.1 This stage aligns the model to downstream use cases such as visual question answering, document understanding, and multi-image reasoning, while simultaneously reinforcing the language-only instruction-following and conversational capabilities.2

How this definition was written

Summarised by glm-5.3-flash from 1 source; every sentence is checked against the quote it cites.

  1. The final stage performs supervised fine-tuning on multimodal instruction-following data, analogous to the language-only SFT pipeline (Section 4.2 — LFM2 Technical Report
  2. This stage aligns the model to downstream use cases such as visual question answering, document understanding, and multi-image reasoning, while simultaneously reinforcing the language-only instruction-following and conversational capabilities — LFM2 Technical Report

Also called Joint Multimodal SFT.

Filed under post-training › supervised fine-tuning