Bedovyy/Anima-INT8

Para usarlo en ComfyUI hay que clonar ComfyUI-Flux2-INT8 dentro del directorio custom_nodes. Después, cargar el modelo con el nodo Load Diffusion Model INT8 (W8A8) y dejar on_the_fly_qunatization en False.

No hay snippets oficiales disponibles para la librería Diffusion Single File en Hugging Face.

Configuración INT8 Tensorwise:

{
  "format": "comfy_quant",
  "block_names": ["net.blocks."],
  "rules": [
    {
      "policy": "keep",
      "match": ["blocks.0", "adaln_modulation", ".mlp.layer2"]
    },
    {
      "policy": "int8_tensorwise",
      "match": ["q_proj", "k_proj", "v_proj", "output_proj", ".mlp"]
    }
  ]
}

Configuración INT8 Rowwise:

{
  "format": "comfy_quant",
  "block_names": ["net.blocks."],
  "rules": [
    {
      "policy": "keep",
      "match": ["blocks.0.", "adaln_modulation", ".0.mlp", ".1.mlp", ".2.mlp", ".3.mlp"]
    },
    {
      "policy": "int8_rowwise",
      "match": ["q_proj", "k_proj", "v_proj", "output_proj", ".mlp"]
    }
  ]
}