antirez/h3.c

View on GitHub

MiniMax H3 inference engine for Mac computers

Stars: 2.7k
Forks: 211
Open beginner issues: 0
Indexed issues: 28

Dominant language: C
License: MIT
Last GitHub push: Aug 11, 2026
Latest indexed: Sep 13, 2026
Contributing guide: No contributing guide
Code of conduct: No code of conduct
Beginner labels: No beginner labels indexed
PR merge metrics: No merged PRs in 30d

Open All

28 issues indexed so far

[Community] h3.c-studio — a bilingual web GUI for h3-metal Open

antirez/h3.c#65 · 2 comments · 0 reactions · 0 assignees · Indexed Sep 14, 2026
Build fails on macOS 15 SDK: MTLGPUFamilyMetal4 undeclared Open

antirez/h3.c#66 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 14, 2026
Qwen causal GQA rounds scaled queries to BF16 before the QK contraction Open

antirez/h3.c#3 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
Questions about M4 Max support, 720p 15‑second generation time, and GUI Open

antirez/h3.c#5 · 7 comments · 1 reaction · 0 assignees · Indexed Sep 13, 2026
RES audio sampling accumulates low-step error on the video sigma grid Open

antirez/h3.c#6 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
Performance: 1344x768 / 5s renders wait-dominated (GPU idle per profile) on M5 Max — measured dataset Open

antirez/h3.c#13 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 13, 2026
Temporary Array is too large for kernel. Encode failed. Open

antirez/h3.c#15 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
Corrupted video (and audio) outputs Open

antirez/h3.c#16 · 7 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
M1 Ultra profile + two negative results (TensorOps unavailable, M5 gates are no-ops) Open

antirez/h3.c#17 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 13, 2026
M4 Pro 24GB verification: official sharded weights render cleanly; misaligned repacked weights produce garbage (confirms PR #14) Open

antirez/h3.c#18 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 13, 2026
Bug: Interactive session doesn't emit 'Done ->' line when stdout is a pipe (non-TTY) Open

antirez/h3.c#25 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
Is it possible to support the Windows OS? Open

antirez/h3.c#26 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
Sparse Attention roadmap / MPS SolAttn implementation Open

antirez/h3.c#27 · 0 comments · 2 reactions · 0 assignees · Indexed Sep 13, 2026
256×256 halved spatial RoPE swaps prompt subjects on the optimized INT8 package (fixed by --use-reference-rope) Open

antirez/h3.c#28 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
M1 Max 64gb Open

antirez/h3.c#29 · 8 comments · 1 reaction · 0 assignees · Indexed Sep 13, 2026
Mac air M5 32GB My first AI video with h3.c — an astronaut on Mars! Open

antirez/h3.c#30 · 2 comments · 1 reaction · 0 assignees · Indexed Sep 13, 2026
no work for ref-image Open

antirez/h3.c#36 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 13, 2026
Feature request: LoRA support Open

antirez/h3.c#37 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 13, 2026
Minimax Music 3 Open

antirez/h3.c#38 · 2 comments · 1 reaction · 0 assignees · Indexed Sep 13, 2026
Feature request: 4bit and 8bit support Open

antirez/h3.c#41 · 1 comment · 1 reaction · 0 assignees · Indexed Sep 13, 2026
CUDA backend for Linux/NVIDIA — interest in upstreaming? Open

antirez/h3.c#42 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
how to download the models in a batch file? where are the instructions? Open

antirez/h3.c#46 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 13, 2026
Qwen text encoder: sequence exceeds Metal threadgroup memory with reference video + ref images (≥150 frames) Open

antirez/h3.c#47 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
what is the practical use for this? I don't see it... Open

antirez/h3.c#49 · 3 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
PLEASE, do *NOT* add cuda support! ask author to fork the source, and make a new distribution, DON'T pollute this one, thanks! Open

antirez/h3.c#50 · 2 comments · 3 reactions · 0 assignees · Indexed Sep 13, 2026
M4 Pro 24GB: fold Turbo LoRA (PR #14) benchmark — 124f is the sweet spot on both 480P and 768P Open

antirez/h3.c#51 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
A potential nondeterministic issue Open

antirez/h3.c#52 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
how to generate 2K videoes Open

antirez/h3.c#54 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026

antirez/h3.c

Get new issues in your inbox