Bedovyy/Anima-FP8

Para usar torch.compile, utiliza el nodo TorchCompileModelAdvanced de KJNodes, configura el modo en max-autotune-no-cudagraphs y asegúrate de que dynamic esté en false.

Capas cuantizadas FP8:

{
  "format": "comfy_quant",
  "block_names": ["net.blocks."],
  "rules": [
    {
      "policy": "keep",
      "match": ["blocks.0", "blocks.1."]
    },
    {
      "policy": "float8_e4m3fn",
      "match": ["q_proj", "k_proj", "v_proj", "o_proj", "output_proj", ".mlp"]
    },
    {
      "policy": "nvfp4",
      "match": []
    }
  ]
}

Capas cuantizadas NVFP4Mixed:

{
  "format": "comfy_quant",
  "block_names": ["net.blocks."],
  "rules": [
    {
      "policy": "keep",
      "match": ["blocks.0."]
    },
    {
      "policy": "float8_e4m3fn",
      "match": ["v_proj", "adaln_modulation", ".mlp"]
    },
    {
      "policy": "nvfp4",
      "match": ["k_proj", "q_proj", "output_proj"]
    }
  ]
}