codemichaeld/minimax_vae

FP8 Model Conversion

• —Source: https://huggingface.co/Mamad8/MiniMax-H3-Image-VAE
• —Original File(s): minimax_h3_t1_image_vae_step1597.safetensors
• —Original Format: safetensors
• —FP8 Format: E5M2
• —FP8 File: minimax_h3_t1_image_vae_step1597-fp8-e5m2.safetensors

Usage

from safetensors.torch import load_file
import torch

# Load FP8 model
fp8_state = load_file("minimax_h3_t1_image_vae_step1597-fp8-e5m2.safetensors")

# Convert tensors back to float32 for computation (auto-converted by PyTorch)
model.load_state_dict(fp8_state)

Note: FP8 tensors are automatically converted to float32 when loaded in PyTorch. Requires PyTorch ≥ 2.1 for FP8 support.

Statistics

• —Total tensors: 562
• —Converted to FP8: 562
• —Skipped (non-float): 0