Gemma 4 12B IT Abliterated Uncensored - GemmAnima Argon Bridge
Korean model card | GitHub app source
Prototype Notice
This is the minimal v0.2.0 Argon bridge release for GemmAnima.
It contains only the evaluated Argon 100k direct Gemma 4 12B hidden-state bridge checkpoint plus metadata and license notices. It does not contain Gemma base weights, Anima base weights, GemmAnima chat LoRAs, vision/tagger LoRAs, planner adapters, old bridge experiments, or rollback checkpoints.
This is a public prototype artifact. It is not a production image model, not a safety-rated assistant, and not a commercial-ready release.
Uploaded Payload
| File | Role |
|---|---|
hiddenstage_bridge/argon_12b_100k_bridge.pt |
Argon 100k direct Gemma 4 12B hidden-state to Anima-conditioning bridge |
adapter_manifest_v0.2.0_argon.json |
SHA256 and byte-size manifest |
asset_manifest.json |
Minimal local placement manifest for this bridge |
model_sources.json |
Minimal source metadata for this bridge |
LICENSE_NOTICES.md |
Composite prototype license notice |
README.md / HF_MODEL_CARD.ko.md |
Model cards |
Base Models Are Not Included
Use the original pages for upstream base weights:
| Component | Source |
|---|---|
| Gemma 4 12B GGUF | ggml-org/gemma-4-12B-it-GGUF |
| Anima diffusion/VAE | circlestone-labs/Anima |
The GitHub app source owns full runtime setup and model download planning. This HF repository is only the Argon bridge artifact.
Evaluation Evidence
Local checks before this release:
| Check | Result |
|---|---|
| Argon 100k bridge final validation MSE | 0.0019557795094442554 |
| Cached smoke render | Pass |
OOD g/e pose render, 3 prompts, Qwen baseline comparison |
Pass |
The OOD test used fresh Gemma 4 12B hidden states for new prompts and rendered Qwen baseline vs Argon bridge outputs side by side. Results were judged usable for an Argon prototype: pose and composition were preserved well enough for this release line, while exact text rendering and strict semantic compliance still need more evaluation.
Intended Use
This bridge is intended for the GemmAnima standalone app and related local experiments that connect Gemma 4 12B hidden states to Anima-compatible conditioning.
Do not use this file as a standalone model. It requires the GemmAnima runtime code, upstream Gemma hidden-state generation, and Anima image-generation assets.
License And Restrictions
This bridge does not relicense upstream models. Users must comply with:
- Gemma/GGUF source license terms
- Anima source license terms
- NVIDIA Open Model License terms where applicable through Anima's upstream lineage
- GemmAnima adapter/checkpoint notices
- Dataset and tag-source restrictions relevant to their use
Because the current image path depends on Anima, treat this Argon prototype as non-commercial, non-production, limited-test material unless you have separately verified all upstream permissions.