MiniMax H3

Not merged yet. H3 support is only in miles-diffusion#154 and sglang-diffusion#34365. Neither repo has it on main. After a normal install, follow Setup to move both checkouts onto those PRs.

1. Model introduction

MiniMax-H3 is MiniMax’s omni-modal generative system: one transformer denoises a packed text + video + audio sequence and emits video with native stereo audio. MiniMax distilled CFG into the checkpoint, so the official setting has no classifier-free guidance — the recipe is unguided (--diffusion-guidance-scale 1.0). The unmerged PR trains the t2va (text-to-audio-video) workflow, video branch only — the audio stream is rolled out but not trained.

2. Supported variants

Model	HF ID	Notes
MiniMax-H3	`MiniMaxAI/MiniMax-H3`	t2va video-only Flow-GRPO on PR #154. `fl2va` / `ref2va` are not trained yet.

3. Setup

H3 is not on main of either repo. Finish Installation first (Docker image or from source), then move both halves onto the open PRs — a main trainer against a main engine will not run this recipe.

Repo	PR	What it adds
miles-diffusion	radixark/miles_diffusion#154	H3 family config, LoRA IPC map, 2-GPU recipe
sglang-diffusion	sgl-project/sglang#34365	MiniMax H3 `t2va` RL rollout

Paths below match the Docker image (/root/miles_diffusion, /sgl-workspace/sglang).

# miles-diffusion → PR 154
cd /root/miles_diffusion
git fetch https://github.com/radixark/miles_diffusion.git pull/154/head:pr-154
git checkout pr-154
pip install --force-reinstall -r requirements.txt   # H3 needs the diffusers git pin
pip install -e . --no-deps

# sglang-diffusion → PR 34365
cd /sgl-workspace/sglang
git fetch https://github.com/sgl-project/sglang.git pull/34365/head:pr-34365
git checkout pr-34365
pip install -e "python[all]"

Drop these checkouts once both PRs merge.

4. Launch

The 2-GPU recipe lives on PR #154, not main: scripts/run_diffusion_grpo_h3_t2va_2gpu.py — 2 colocate GPUs, reward colocated, 16:9 / 4 s (sglang pins short_edge=768, so the canvas is 1344×768 / 107 frames). Requires --use-lora --lora-ipc-weight-sync (the recipe already sets both). Status: 📈 V — Verified, based on the complete run reported on PR #154. The command does not exist on main.

# alignment check (optimizer frozen)
python3 scripts/run_diffusion_grpo_h3_t2va_2gpu.py \
  --cuda-visible-devices 0,1 \
  --num-rollout 2 --eval-interval 0 --debug-alignment

# short train
python3 scripts/run_diffusion_grpo_h3_t2va_2gpu.py \
  --cuda-visible-devices 0,1 --num-rollout 5 --eval-interval 0

# 2-GPU long run (resource-limited scale)
python3 scripts/run_diffusion_grpo_h3_t2va_2gpu.py \
  --rollout-batch-size 4 --n-samples-per-prompt 16 \
  --cuda-visible-devices 0,1

The commands above are the 2-GPU verification path on PR #154. No larger canonical H3 topology is committed to main.

5. Reference results

2-GPU long run from PR #154 (rollout_batch_size=4, n_samples_per_prompt=16):

train/log_prob_mean_abs_diff stays in 1.4e-5 .. 7.1e-5 (under 1e-4).
Eval PickScore (eval/pickscore_test, every 10 rollouts): 0.806 → 0.814, peak 0.816 at step 59.
Train rollout/reward/raw_mean sits around ~0.79 (64 samples / step; noisier than eval).

This is the batch-4 curve reported on PR #154; it is not evidence for a larger topology.

Getting Started

User Guide

Models

Advanced

Developer

1. Model introduction

2. Supported variants

3. Setup

4. Launch

5. Reference results

​1. Model introduction

​2. Supported variants

​3. Setup

​4. Launch

​5. Reference results

1. Model introduction

2. Supported variants

3. Setup

4. Launch

5. Reference results