modelscope / modelscope/DiffSynth-Studio

oom on minimax h3

Open
#1,682 1 comment 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
13.1k
Forks
1.3k
Avg merge
13h 12m
Merged PRs (30d)
45

Description

I train h3 on 768p 10s videos with 9 reference images using parameter here:

--height 1344
--width 768
--num_frames 243 \

using 4*8h100 (84G memory), I get oom error. So I am curious about that one sample is very large, and should be splitted into different gpus with sequence parallel? I am not sure whether it supports sequence or context paralle.

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

No source file, test, or entry point is named. Start by reproducing the Minimax H3 training command with 768×1344 resolution, 243 frames, nine reference images, and four 84G H100 GPUs; then trace the H3 training entry point and its parallelism configuration. Done means the OOM cause and whether sequence or context parallelism is supported are established, with a documented fix or limitation.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
machine-learning, performance
Issue type
Bug
Difficulty
5/5
Estimated time
Over a week
Activity status
Active
Clarity
Needs clarification
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.