README.md

---
library_name: diffusers
tags:
- fp8
- safetensors
- precision-recovery
- diffusion
- converted-by-gradio
---
# FP8 Model with Precision Recovery
- **Source**: `https://huggingface.co/spacepxl/Wan2.1-VAE-upscale2x`
- **File**: `Wan2.1_VAE_upscale2x_imageonly_real_v1.safetensors`
- **FP8 Format**: `E5M2`
- **Architecture**: vae
- **Precision Recovery Type**: Correction Factors
- **Precision Recovery File**: `Wan2.1_VAE_upscale2x_imageonly_real_v1-correction-vae.safetensors`
- **FP8 File**: `Wan2.1_VAE_upscale2x_imageonly_real_v1-fp8-e5m2.safetensors`
## Usage (Inference)
```python
from safetensors.torch import load_file
import torch
# Load FP8 model
fp8_state = load_file("Wan2.1_VAE_upscale2x_imageonly_real_v1-fp8-e5m2.safetensors")
# Load precision recovery file
recovery_state = load_file("Wan2.1_VAE_upscale2x_imageonly_real_v1-correction-vae.safetensors") if "Wan2.1_VAE_upscale2x_imageonly_real_v1-correction-vae.safetensors" else {}
# Reconstruct high-precision weights
reconstructed = {}
for key in fp8_state:
    fp8_weight = fp8_state[key].to(torch.float32)
    if recovery_state:
        # For LoRA approach
        if "lora_A" in recovery_state:
            if f"lora_A.{key}" in recovery_state and f"lora_B.{key}" in recovery_state:
                A = recovery_state[f"lora_A.{key}"].to(torch.float32)
                B = recovery_state[f"lora_B.{key}"].to(torch.float32)
                lora_weight = B @ A
                reconstructed[key] = fp8_weight + lora_weight
            else:
                reconstructed[key] = fp8_weight
        # For correction factor approach
        elif f"correction.{key}" in recovery_state:
            correction = recovery_state[f"correction.{key}"].to(torch.float32)
            reconstructed[key] = fp8_weight + correction
        else:
            reconstructed[key] = fp8_weight
    else:
        reconstructed[key] = fp8_weight
```
> Requires PyTorch ≥ 2.1 for FP8 support.