Hacktoberfest 2026: the issues maintainers tagged for October, open and beginner-friendly. Browse Hacktoberfest issues

model: MiniMax-M3.1-Flash (minimax-m3 shapes, sparse attention on every layer, text + vision)

Open
  • Is something wrong?
#698 3 comments 0 reactions 0 assignees View on GitHub

Maintainers usually reply within 1 day

Nobody has claimed this yet.

Assessment

Difficulty
5/5
Estimated time
Over a week
Newbie friendliness
25/100
Issue type
Feature
Clarity
Mostly clear
Activity status
Active
Tech stack
rust
Domain
backend, machine-learning

Research direction

Start with the existing MiniMax-M3 model code in crates/model/src/arch/minimaxm3 and compare its attention and FFN assumptions with the four M3.1 Flash differences listed here. The issue calls for a design round covering attention and FFN kinds, GGUF and oracle routing, and placement across two cards plus the host tier; contributors should comment first to coordinate. Done means the maintainers’ design choices are agreed before the text path and then vision work begin.

Written by the indexing model from the issue text.

Description

model

MiniMax's M3.1 Flash: a multimodal MoE (text + vision), the same shapes as MiniMax-M3. Weights: MiniMaxAI/MiniMax-M3.1-Flash at 76faee1b.

  • Safetensors only, 250 GB: MXFP8 weights, NVFP4 routed experts. A 2.3 GB DSpark draft model (5 layers) ships beside it.
  • We already describe MiniMax-M3 on the model side (crates/model/src/arch/minimaxm3). M3.1 Flash differs in four facts: every layer runs MiniMax Sparse Attention (M3 ran it only on the MoE layers), layers 0-2 keep a dense FFN, there is no MTP, and rope theta is 1e7.
  • No GGUF and no llama.cpp support yet. llama.cpp's minimax-m3 graph ties sparse attention to the leading dense count, so the M3 path would build layers 0-2 wrong.
  • License: MiniMax's own (custom, not OSI).

The maintainers take this one. We start with a design round: the reader split of attention kind and FFN kind, the GGUF and oracle route, and the placement on two cards plus the host tier. Then the text path, then vision. Contributors are welcome on any stage. Comment first so we do not duplicate work.

(Written with AI help.)

Dominant language
Rust
Stars
21
Forks
4
Avg merge
2h 32m
Merged PRs (30d)
530

Getting set up

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.

More from midagedev/bloomery

All issues in midagedev/bloomery

Similar issues

More Rust issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.