MiniMax-H3-Pruned

About

What MiniMax-H3-Pruned is

MiniMax-H3, AdaLN-pruned MiniMaxAI/MiniMax-H3 in diffusers format with the AdaLN input projections reduced to their reachable rank. 52 GB smaller across the two DiT partitions, no visible change in output, and every published H3 LoRA loads - pruned-trained ones natively, released-trained ones projected. The conditioner is hosted here too, truncated: MiniMax-H3 conditions on the unnormalized hidden state after the Qwen3-VL conditioner's 50th decoder layer (hiddenstates[50]), so decoder layers 51-63 and the language-model head can never reach the conditioning and are dropped. textencoder/ ships decoder layers 0-50 - the 51st is kept so hiddenstates[50] stays a pre-norm intermediate rather than the final-norm output - and its embeddings are bitwise identical to the released conditioner's, text-only and vision-bearing presentations alike. It stays bfloat16; a quantized conditioner is planned as a separate variant, pending its own quality measurement. Both VAEs, both schedulers, the tokenizer and the processor are not duplicated here. modularmodelindex.json points those at MiniMaxAI/MiniMax-H3, so loading this repo pulls only the pruned partitions, the truncated conditioner, and the released components you already have. What was removed, and why it is safe Each of the 50 blocks holds adalnproj.linear, a Linear(2688 - 96768), and normout.linear adds one more. Together that is 13.03 B parameters, 39.3% of the checkpoint. All 51 of them read the same vector, silu(timeembedder(t)), which is a function of the scalar timestep alone - so however wide it is, the set of values it ever takes is a one-dimensional curve in R^2688. An 8-dimensional affine subspace covers that curve to a relative RMS error of 1.45e-5. One bfloat16 rounding step of the weights those projections are stored in is about 3.9e-3 - roughly 250x larger. So the projections are folded onto that subspace: W @ x + b with x = mean + c @ basis becomes (W @ basis.T) @ c + (b + W @ mean), the timestep MLP is replaced by a 1025-entry table of the coordinates c(t) (linearly interpolated), and every AdaLN projection takes an 8-wide input instead of a 2688-wide one. Measured fidelity Every folded projection, evaluate

Summarised from the published model card. Read the full card on the HuggingFace links below.

Specifications

What it is

How it works

How video models work

Variants

Sizes & precisions

Open weights ship in several sizes and precisions. One page, all the variants — pick the one that fits your GPU. VRAM figures are estimates from model size.

Using it via the API

Call it like any OpenAI endpoint

Once AxForge deploys multimodalart-minimax-h3-pruned for you, it answers on the OpenAI-compatible API — the same base URL and keys as every other model. (multimodalart-minimax-h3-pruned below is illustrative; you get the exact model name on deployment.)

$ curl -sS https://api.axforge.ai/v1/videos/generations \
  -H "Authorization: Bearer $AXFORGE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"multimodalart-minimax-h3-pruned","prompt":"a drone shot over a forest"}'

Details

Languages, data & research

Licence

Commercial licence needed

Sources

Weights & code

Explore

More video models