https://coobiw.github.io/tags/mllm-training-framework-supporting-multimodal-pipeline-parallel/ MLLM Training Framework (Supporting Multimodal Pipeline Parallel) Hugo Blox Builder (https://hugoblox.com)en-usMon, 25 Dec 2023 00:00:00 +0000 https://coobiw.github.io/media/icon_hu_557f799f97e4e28d.png https://coobiw.github.io/tags/mllm-training-framework-supporting-multimodal-pipeline-parallel/ https://coobiw.github.io/project/mpp-llava/ Mon, 25 Dec 2023 00:00:00 +0000 https://coobiw.github.io/project/mpp-llava/ <p>The Repo supports {video/image/multi-image} {single/multi-turn} conversations. All 7B/14B llava-like training is conducted on 3090/4090 GPUs. <em><strong>To prevent poverty (24GB of VRAM) from limiting imagination</strong></em>, I implemented an MLLM version based on deepspeed Pipeline Parallel.</p>