modelscope / modelscope/DiffSynth-Studio

the training details of minimax h3 model

Open
#1,585 4 comments 1 reaction 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 am currently training a LoRA for the MiniMax-H3 FL2VA model. However, after training on 10k videos, both the video clarity and temporal consistency have noticeably degraded. I am unsure if this issue stems from the prompts, the dataset, or certain training parameters.
I noticed that DiffSynth open-sourced a LoRA checkpoint (https://modelscope.ai/models/DiffSynth-Studio/MiniMax-H3-LoRA-LineartAnime). Could you share the specific training details for this model? I am particularly interested in the prompt design (how to align with the original MiniMax prompt templates), training steps, batch size, timestep shift, learning rate, and LoRA rank. I would love to reference these details to improve my training quality and consistency. Thank you!

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

Start with the linked DiffSynth-Studio/MiniMax-H3-LoRA-LineartAnime checkpoint and the original MiniMax prompt templates referenced in the issue. Determine the prompt design, training steps, batch size, timestep shift, learning rate, and LoRA rank used, then document those details so the requester can reproduce or compare the training setup.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
machine-learning
Issue type
Documentation
Difficulty
3/5
Estimated time
1-2 days
Activity status
Active
Clarity
Mostly clear
Newbie friendliness
42/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.