Models

dppan runs GGUF models. The table below is the curated catalog — models validated end-to-end on the network. It is not a registry mirror: any GGUF from Ollama or Hugging Face can still be tried (dppan pull hf:<owner>/<repo>), the catalog just marks what we vouch for. Check it from the CLI anytime:

dppan models          # table with local availability
dppan models --json   # for scripting

Supported models

all archsall registriesollamahugging faceinclavate0 / 0

Model	Arch	Layers	Size	Registries	Manifest

The table is fetched live from dl.inclavate.io/models/catalog.json — the same catalog the CLI uses (dppan models caches it for 24 hours and works offline from a built-in copy). Models larger than one machine use sharded download.

Coverage & roadmap

The engine supports 43 architecture families — enough to run 157 of the top-200 trending text-generation models on Hugging Face. Where the rest stands:

Tier	Covers	Status
Dense transformers	Llama 3.x · Qwen 2.5/3 · Gemma 2/3/4 · Phi-3/4 · GLM-4 · SmolLM — plus their distills and finetunes	✅ Supported
Dense classics	GPT-2 · Pythia · BLOOM · Falcon · Phi-2 · Nemotron · Command R7B	✅ Supported
Mixture-of-Experts	Qwen3-MoE (A3B/235B) · gpt-oss (attention sinks, MXFP4) · Gemma 4 A4B · OLMoE · Qwen1.5-MoE · GLM-4.5-Air and GLM-4.5 355B-A32B · Mixtral 8x7B/8x22B · Llama 4 Scout & Maverick	✅ Supported
MLA attention	DeepSeek-V2 · V2-Lite · Coder-V2-Lite · GigaChat3 (latent-compressed KV, absorbed & legacy)	✅ Supported
State-space & hybrid	Mamba (1 & 2) · FalconMamba · Codestral Mamba · Granite 4.0-H (Micro dense · Tiny/Small MoE) · Falcon-H1 (parallel attention + Mamba2, 0.5B–34B)	✅ Supported
Hybrid — single mixer per layer	Nemotron Nano 2 (Nemotron-H, mamba2 / attention / MLP interleaved) · Nemotron-H 47B Reasoning · Nemotron-3 MoE (30B-A3B · 120B-A12B · 550B-A55B) · LFM2 (350M–2.6B, gated short-conv) · LFM2-8B-A1B (MoE) · Jamba (Mamba-1 + attention + MoE; 900M · Mini 1.7 · Large 1.7)	✅ Supported
Granite dense & MoE	Granite 3.x/4.x dense (2B/8B) · Granite-MoE a400m/a800m (incl. PowerMoE-3b)	✅ Supported
Gated delta-net	Qwen3.5 dense (0.8B–27B) · Qwen3.6 · Qwen3.5-MoE (A3B/A10B/A17B) · Qwen3-Next-80B (Instruct/Thinking) · Qwen3-Coder-Next — interleaved linear-attention + gated attention hybrid	✅ Supported
MLA — frontier	DeepSeek-V3 · R1 · V3.2 · Kimi K2 · Mistral-Large-3 (675B) · GLM-5/5.2 (glm-dsa) — all absorbed MLA, 671B–1T. V3.2 and Mistral-Large-3 convert to the same deepseek2 arch; GLM-5 ships DSA indexer tensors that upstream loads and does not use at inference	📋 In Validation
Multimodal input	Image, audio, and ordered video frames on any model with a compatible projector (mmproj) — Gemma 4 · Gemma 3 · Qwen2-VL · Qwen2.5-VL (3B–72B) · Qwen3-VL (2B–32B, 30B-A3B, 235B-A22B; deepstack) · GLM-4.6V-Flash · DeepSeek-OCR · Ultravox. Input only; the model answers in text — including document OCR (DeepSeek-OCR reads a page image back as text/markdown)	✅ Supported
Document input	PDF and text/Markdown files, optional OCR, native server-side video decode	📋 Planned
Media generation	Generating images, speech, and video — plus the OpenAI-compatible Files, Images, and Audio endpoints and async job/artifact storage they need	📋 Planned
Gemma 4 E-series	Per-layer embeddings (needs a protocol extension)	📋 Planned

No dates are promised — items ship when they pass the same validation gate as everything in the catalog. Missing a model you need? Write to contact@inclavate.io — requests directly shape this list.

Model sources

dppan pull and dppan download resolve models from three places:

Source	Ref format	Auth
Ollama registry	llama3.3:70b	none
Hugging Face (GGUF repos)	hf:bartowski/Llama-3.3-70B-Instruct-GGUF:Q6_K	none for public repos; your own HF_TOKEN for gated ones
Inclavate-hosted	inclavate:<slug>[:<quant>]	none

• Hugging Face: only repos that ship GGUF files (bartowski, unsloth, lmstudio-community, …). The quant selector defaults to Q4_K_M, but any GGUF quantisation runs — pick another with hf:<owner>/<repo>:<QUANT> (e.g. :Q6_K, :Q8_0, :IQ4_XS); the engine keeps tensors in their native type and the right kernels are dispatched automatically. The quant shown in the catalog is simply the configuration we validated. Split quants (-00001-of-0000N.gguf) are merged automatically. For gated repos, accept the license on huggingface.co and set HF_TOKEN (or log in with huggingface-cli) — tokens are always your own.
• Inclavate-hosted is for models with no community GGUF: converted and quantized once, centrally, then served from dl.inclavate.io — so nodes never have to quantize anything themselves.

Integrity

A GGUF has no per-tensor checksums, so a partially-downloaded shard can't be checked against the source's whole-file hash. dppan's answer is a per-tensor manifest:

dppan manifest llama3.3:70b -o 70b.manifest.json   # once, on a machine with the full file
dppan pull llama3.3:70b --layers 27-53 --verify --manifest 70b.manifest.json

With --verify, every downloaded tensor is hashed and compared; any mismatch names the tensor and deletes the shard. Without a manifest, shards are still structurally validated and fetched over TLS — fine for machines you control. Manifests published at dl.inclavate.io/manifests/<sha256>.json are picked up automatically, no --manifest flag needed.