{"slug": "holo-fuse-class-conditional-diffusion-on-the-0-parameter-hsl-byte-substrate", "title": "HoLo-FuSe — class-conditional diffusion on the 0-parameter HSL byte substrate (minimal-scale baseline, honest results)", "summary": "HoLo-FuSe, a class-conditional diffusion model using a frozen 27-D HSL byte substrate with zero learned parameters, successfully generated cat and dog faces from AFHQ data at 128px resolution on a single Colab T4. The minimal-scale baseline run demonstrated proof of operation, matching the performance of a same-budget learned embedding control while using no learned conditioning parameters.", "body_md": "Following up on the HoLo line (byte-LM, speech): **HoLo-FuSe** tests whether the same frozen HSL\n\nsubstrate can serve as the **conditioning door of an image-generation carrier**. Same method as the\n\nother rooms: take a verified baseline, swap exactly one door for HSL, measure against controls.\n\n**Honest framing first.** This was trained on a single free Colab T4 (~35M U-Net, 16k steps/arm,\n\n128px). It is a **minimum-scale baseline run** — the point is *proof of operation*, not visual\n\nquality. Please read the samples with that in mind; compute, not the method, is the main\n\nquality ceiling here.\n\nHSL is a frozen, deterministic 27-D feature frame over bytes (value geometry + cross-byte flow +\n\nboundary + Fourier + phase; a 4.6 KB LUT, **0 learned parameters**). Per the family rule —\n\n*fixed substrate where possible, explicit lens where necessary* — the only lens this room needs is\n\na **label→condition readout**:\n\n`label bytes -> frozen 27-D HSL frame (0 params) -> small learned readout (2-layer MLP) -> condition embedding -> added to the DDPM timestep embedding`\n\nNo spatial lens: the conv U-Net carrier already owns spatial structure. The substrate stays frozen;\n\nthe readout is the only trained conditioning component, and it is budget-matched to the control.\n\nClass-conditional DDPM (cosine schedule T=250, multi-level U-Net + self-attention, EMA,\n\nclassifier-free guidance with cond-drop 0.15). Data: AFHQ animal faces at 128px, Cat 5153 / Dog\n\n4739 (CC BY-NC 4.0 — so weights and samples are **non-commercial**). Three arms, **same seed,\nsame data, same architecture surface, same budget**, compared at step 14000:\n\n| arm | conditioning | result (qualitative, seed-matched) |\n|---|---|---|\n`none` |\nunconditional baseline | readable cat+dog faces, classes mixed |\n`learned` |\nsame-budget `nn.Embedding` control |\n“Cat”→cats, “Dog”→dogs |\n`hsl` |\nfrozen HSL 27-D (0 learned params) + readout | “Cat”→cats, “Dog”→dogs |\n\nTwo observations from the seed-matched bench (image attached; positions share the same initial\n\nnoise):\n\nThe plan is to focus on **quality of this one model** (longer training, color balance, multi-seed\n\nevaluation) rather than adding more models. That said, the compute budget is what it is — a free\n\nT4 and a 4 GB laptop GPU — so please keep expectations modest. If the substrate claim survives\n\nbetter-trained carriers, that is the result we are after; prettier cats are a bonus.\n\nCode, training harness, bench scripts, and the POC record:\n\n[https://github.com/Woojiggun/HoLo-FuSe](https://github.com/Woojiggun/HoLo-FuSe)\n\nLive demo (ZeroGPU, generates in seconds; checkpoints are linked from the repo):", "url": "https://wpnews.pro/news/holo-fuse-class-conditional-diffusion-on-the-0-parameter-hsl-byte-substrate", "canonical_source": "https://discuss.huggingface.co/t/holo-fuse-class-conditional-diffusion-on-the-0-parameter-hsl-byte-substrate-minimal-scale-baseline-honest-results/177711#post_1", "published_at": "2026-07-12 14:51:29+00:00", "updated_at": "2026-07-12 15:16:57.297270+00:00", "lang": "en", "topics": ["artificial-intelligence", "machine-learning", "generative-ai", "computer-vision", "ai-research"], "entities": ["HoLo-FuSe", "AFHQ", "Colab T4", "DDPM", "U-Net"], "alternates": {"html": "https://wpnews.pro/news/holo-fuse-class-conditional-diffusion-on-the-0-parameter-hsl-byte-substrate", "markdown": "https://wpnews.pro/news/holo-fuse-class-conditional-diffusion-on-the-0-parameter-hsl-byte-substrate.md", "text": "https://wpnews.pro/news/holo-fuse-class-conditional-diffusion-on-the-0-parameter-hsl-byte-substrate.txt", "jsonld": "https://wpnews.pro/news/holo-fuse-class-conditional-diffusion-on-the-0-parameter-hsl-byte-substrate.jsonld"}}