You need to agree to share your Hugging Face information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this model content.

Polyvalent Forensic Deepfake Specialist (Gated Model)

Gated Access & Permission Notice: Release of these weights is gated for authorized academic research only, pending formal upstream licensing clearance from Veritas and HydraFake dataset authors. Under no circumstances are HydraFake facial image splits or the sft_36k.json dataset redistributed in this repository.

Model Description

  • Specialist Type: Multimodal facial manipulation detection and forensic explanation.
  • Base Backbone: OpenGVLab/InternVL3-2B (pinned at revision 899155015275a9b7338c7f4677e19c784e0e5a21).
  • Components:
    • llm_adapter/: Language model LoRA ($r=64$, $\alpha=128$, 73.8M parameters).
    • vision_adapter/: Vision encoder LoRA ($r=16$, $\alpha=32$, 6.3M parameters).
    • custom_heads/: 5 bare classification heads (attention pooling, frequency analysis, custom projector, domain router, classification head; 13.3M parameters).

Loading and Usage

Requires loader.py and wrapper.py shipped with this repository:

import torch
from loader import load_trained_wrapper

wrapper, tokenizer = load_trained_wrapper(
    base_dir="OpenGVLab/InternVL3-2B",
    llm_adapter_path="./llm_adapter",
    vision_adapter_path="./vision_adapter",
    heads_path="./custom_heads/custom_heads.safetensors",
    device="cuda" if torch.cuda.is_available() else "cpu"
)

Citation

@article{polyvalent2026,
  title={Polyvalent: Single-Base Multi-Specialist Architecture with Dynamic Adapter Residency},
  author={Polyvalent Research Team},
  year={2026}
}
Downloads last month
-
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for Pushkarrrr15151/PolyVL-deepfake-lora