FastVideo-FastH3-8-Step-V2-NVFP4
This UniServe deployment checkpoint applies calibrated NVIDIA Model Optimizer NVFP4 W4A4 to the fused gate/up and down projections in all 50 H3 denoiser gated MLPs and to the qkv, attention-output, fused gate/up, and down projections in all 36 Video VAE decoder Transformer blocks; denoiser attention, the BF16 text encoder, token refiner, modulation, normalization, latent and conditioning boundaries, convolutions, residual/affine paths, and audio decoder remain BF16/FP32. Calibration used 1,000 prompts with complete eight-forward denoising trajectories and deployment-distribution VAE decoding, ModelOpt commit
6a4b3f147e14a6fec690fedbced8df402344085d, E2M1 values, K16 FP8 E4M3 block scales, FP32 tensor scales, and calibration-derived activation tensor scales.
This repository is a UniServe deployment checkpoint in NVIDIA Model Optimizer's unified Hugging Face layout, derived from FastVideo/FastVideo-FastH3-8-Step-V2. It generates synchronized video and audio with eight transformer forwards and retains the source checkpoint's VSA-H3 execution contract, including video scheduler shift 10 and VSA sparsity 0.8.
Run with UniServe
The quantized transformer and vae components store packed E2M1 values in .weight, K16 FP8 E4M3 block scales in .weight_scale, the FP32 weight scale in .weight_scale_2 and the static activation scale in .input_scale, and each declares a quantization_config (quant_method: modelopt, quant_algo: NVFP4) in its config.json; layers listed under ignore stay dense. UniServe detects this configuration automatically. The checkpoint does not require NVIDIA Model Optimizer at deployment time and rejects runtime precision presets or component overrides that would conflict with its calibrated weights and scales. The tensors are the calibrated values and scales of this repository's earlier packed layout, renamed and described bit for bit, so the validation below applies unchanged.
uniserve serve skx618/FastVideo-FastH3-8-Step-V2-NVFP4 \
--served-model-name MiniMax-H3 \
--dtype bfloat16
The checkpoint targets the source model's four-B200 execution contract. The GPU count must divide H3's 56 attention heads.
Validation status
The denoiser and Video VAE components were exported after their complete 1,000-record Phase A and Phase B calibration jobs. An executable-checkpoint review generated five 5-second, 1344×768, 24-fps synchronized audio/video samples from this checkpoint and five same-prompt, same-seed quality-reference samples successfully while preserving the source model's VSA sparsity 0.8 contract. Across those five pairs, mean RGB PSNR was 11.92 dB, SSIM 0.3647, AlexNet LPIPS 0.5184, and audio log-mel cosine 0.9346. These prompts came from the calibration-source dataset, so the measurements describe trajectory drift and generation viability, not independent held-out quality acceptance; no perceptual-equivalence claim is made.
License
This derivative inherits the MiniMax H3 Community License and the source checkpoint's usage conditions.