neko-legends/Ornith-1.0-35B-AEON-Ultimate-Uncensored-NVFP4-GGUF-MTP

<div style="font-family: Inter, ui-sans-serif, system-ui, -apple-system, BlinkMacSystemFont, 'Segoe UI', sans-serif; border: 1px solid #2f2118; border-radius: 18px; overflow: hidden; background: #0b0b0f; box-shadow: 0 20px 48px rgba(0,0,0,0.26); margin: 0 0 28px 0;"> <div style="padding: 30px 28px 24px 28px; background: radial-gradient(circle at 8% 0%, rgba(255,122,26,0.34), transparent 34%), radial-gradient(circle at 92% 10%, rgba(255,184,107,0.18), transparent 26%), linear-gradient(135deg, #050507 0%, #111116 54%, #1f1209 100%); border-bottom: 1px solid rgba(255,122,26,0.35);"> <div style="display: flex; flex-wrap: wrap; gap: 14px; align-items: center; justify-content: space-between;"> <div> <div style="font-size: 11px; font-weight: 900; color: #ffb86b; letter-spacing: 1.8px; text-transform: uppercase;">Neko Legends local inference release</div> <h1 style="margin: 8px 0 0 0; color: #fff7ed; font-size: 30px; line-height: 1.12; font-weight: 950; border: 0;">Ornith-1.0-35B-AEON-Ultimate-Uncensored-NVFP4-GGUF-MTP</h1> </div> <div style="background: rgba(255,122,26,0.14); border: 1px solid rgba(255,122,26,0.72); color: #ffd7ad; font-size: 12px; font-weight: 900; padding: 8px 12px; border-radius: 999px;">RTX 5090 validated</div> </div> <p style="margin: 14px 0 0 0; max-width: 900px; color: #d6d3d1; font-size: 14px; line-height: 1.7;"> A text-generation GGUF package for recent <code style="color:#ffb86b;">llama.cpp</code> builds: AEON Ultimate Uncensored NVFP4 trunk/body weights with a compatible MTP block, ready for Blackwell native FP4 local serving. </p> </div>

<div style="display: grid; grid-template-columns: repeat(auto-fit, minmax(155px, 1fr)); gap: 1px; background: #2f2118;"> <div style="background:#111116; padding: 15px 16px;"><span style="display:block; color:#a8a29e; font-size:11px; font-weight:900; text-transform:uppercase;">Format</span><b style="display:block; margin-top:5px; color:#fff7ed; font-size:18px;">GGUF</b></div> <div style="background:#111116; padding: 15px 16px;"><span style="display:block; color:#a8a29e; font-size:11px; font-weight:900; text-transform:uppercase;">Quant</span><b style="display:block; margin-top:5px; color:#ffb86b; font-size:18px;">NVFP4</b></div> <div style="background:#111116; padding: 15px 16px;"><span style="display:block; color:#a8a29e; font-size:11px; font-weight:900; text-transform:uppercase;">Spec decode</span><b style="display:block; margin-top:5px; color:#fff7ed; font-size:18px;">draft-mtp</b></div> <div style="background:#111116; padding: 15px 16px;"><span style="display:block; color:#a8a29e; font-size:11px; font-weight:900; text-transform:uppercase;">Validated ctx</span><b style="display:block; margin-top:5px; color:#ffb86b; font-size:18px;">262k</b></div> <div style="background:#111116; padding: 15px 16px;"><span style="display:block; color:#a8a29e; font-size:11px; font-weight:900; text-transform:uppercase;">Target stack</span><b style="display:block; margin-top:5px; color:#fff7ed; font-size:18px;">llama.cpp</b></div> <div style="background:#111116; padding: 15px 16px;"><span style="display:block; color:#a8a29e; font-size:11px; font-weight:900; text-transform:uppercase;">Artifact</span><b style="display:block; margin-top:5px; color:#ffb86b; font-size:18px;">23.4 GB</b></div> </div> </div>

[!IMPORTANT] This repo publishes one recommended AEON-trunk MTP artifact. It was validated for text serving only; the original safetensors family is multimodal, but this GGUF card does not claim vision or multimodal serving support.

Quick Start

<div style="font-family: Inter, ui-sans-serif, system-ui, -apple-system, BlinkMacSystemFont, 'Segoe UI', sans-serif; display: grid; grid-template-columns: repeat(auto-fit, minmax(240px, 1fr)); gap: 14px; margin: 18px 0 26px 0;"> <div style="border:1px solid #3a2a1f; background:#111116; border-radius:14px; padding:16px;"> <div style="color:#ffb86b; font-size:12px; font-weight:950; letter-spacing:0.8px; text-transform:uppercase;">Download</div> <p style="margin:8px 0 0 0; color:#e7e5e4; font-size:13px; line-height:1.65;">Use <code>ornith-1.0-35b-aeon-ultimate-uncensored-nvfp4-gguf-mtp.gguf</code>. It is the AEON NVFP4 GGUF with grafted compatible MTP block.</p> </div> <div style="border:1px solid #3a2a1f; background:#111116; border-radius:14px; padding:16px;"> <div style="color:#ffb86b; font-size:12px; font-weight:950; letter-spacing:0.8px; text-transform:uppercase;">Serve</div> <p style="margin:8px 0 0 0; color:#e7e5e4; font-size:13px; line-height:1.65;">Run with a current CUDA 13.x <code>llama.cpp</code> build and enable <code>--spec-type draft-mtp</code> or the tuned <code>draft-mtp,ngram-mod</code> profile.</p> </div> <div style="border:1px solid #3a2a1f; background:#111116; border-radius:14px; padding:16px;"> <div style="color:#ffb86b; font-size:12px; font-weight:950; letter-spacing:0.8px; text-transform:uppercase;">Expect</div> <p style="margin:8px 0 0 0; color:#e7e5e4; font-size:13px; line-height:1.65;">On the tested RTX 5090 machine, <code>llama.cpp</code> initialized MTP at full 262k context and reported <code>BLACKWELLNATIVEFP4 = 1</code>.</p> </div> </div>

RTX 5090 Snapshot

RTX 5090, Windows, llama.cpp-b9267-cuda13.1, context 262144, generation 1024 tokens, temperature=0.6.

Tuning note: for the 10k prompt, draft-mtp with --spec-draft-n-max 2 reached 133.7 decode tok/s and 104.0 full-wall tok/s. The chart uses the single temp=0.6 draft-mtp,ngram-mod profile for both prompt sizes.

<div style="font-family: Inter, ui-sans-serif, system-ui, -apple-system, BlinkMacSystemFont, 'Segoe UI', sans-serif; border: 1px solid #3a2a1f; background: #0f0f13; border-radius: 16px; overflow: hidden; margin: 18px 0 28px 0;"> <div style="padding: 14px 18px; background: linear-gradient(90deg, #1a120d 0%, #2b1708 100%); border-bottom: 1px solid rgba(255,122,26,0.35); color: #ffd7ad; font-weight: 950;">Windows native GGUF and MTP benchmark chart</div> <a href="https://huggingface.co/neko-legends/Ornith-1.0-35B-AEON-Ultimate-Uncensored-NVFP4-GGUF-MTP/blob/main/images/aeon-ornith-windows-docker-vs-gguf.png" target="_blank" style="display:block; background:#050507;"> <img src="https://huggingface.co/neko-legends/Ornith-1.0-35B-AEON-Ultimate-Uncensored-NVFP4-GGUF-MTP/resolve/main/images/aeon-ornith-windows-docker-vs-gguf.png" alt="AEON Ornith Ultimate Uncensored NVFP4 Windows Docker vs native GGUF benchmark chart" style="display:block; width:100%; border:0;" /> </a> </div>

Files

Which File Should I Use?

Use ornith-1.0-35b-aeon-ultimate-uncensored-nvfp4-gguf-mtp.gguf for the AEON Ultimate Uncensored NVFP4 GGUF with MTP serving support in llama.cpp. This repository intentionally publishes only the AEON-trunk MTP artifact.

This is a text-generation GGUF. The original safetensors model family is multimodal, but this GGUF file was validated for text serving only.

MTP Provenance

AEON's compressed-tensors checkpoint advertises mtp_num_hidden_layers = 1 in config metadata, but the downloaded model.safetensors contained no mtp, nextn, or model.layers.40 tensor names. A direct conversion with MTP metadata failed in llama.cpp because blk.40.attn_norm.weight and the rest of the MTP block were absent.

The recommended MTP file in this repo was therefore built as a graft:

• —Base/trunk/body: local base-only GGUF intermediate converted from `AEON-7/Ornith-1.0-35B-AEON-Ultimate-Uncensored-NVFP4`, not published in this repo
• —MTP block donor: 20 blk.40.* tensors from `s-batman/Ornith-1.0-35B-NVFP4-MTP-GGUF`
• —Result metadata: qwen35moe.block_count = 41, qwen35moe.nextn_predict_layers = 1, GGUF.tensor_count = 993

Local validation confirmed llama.cpp initializes draft-mtp successfully at full 262k context and reports BLACKWELL_NATIVE_FP4 = 1 on RTX 5090.

SHA256 for ornith-1.0-35b-aeon-ultimate-uncensored-nvfp4-gguf-mtp.gguf:

3F0545EE14ED3B01A18E794945E33FFE6876F9A3C3787316A652C6CFDE4BDDE3

Example llama.cpp Command

$LlamaServer = Join-Path "<path-to-llama.cpp-build-folder>" "llama-server.exe"
$Model = Join-Path "<path-to-model-folder>" "ornith-1.0-35b-aeon-ultimate-uncensored-nvfp4-gguf-mtp.gguf"

& $LlamaServer `
  --model "$Model" `
  --alias aeon-ornith-1.0-35b-nvfp4-aeon-mtp `
  --host 127.0.0.1 `
  --port 39199 `
  --device CUDA0 `
  --gpu-layers all `
  --gpu-layers-draft all `
  --ctx-size 262144 `
  --cache-type-k q4_0 `
  --cache-type-v q4_0 `
  --cache-type-k-draft q4_0 `
  --cache-type-v-draft q4_0 `
  --flash-attn on `
  --parallel 1 `
  --cont-batching `
  --jinja `
  --metrics `
  --slots `
  --spec-type draft-mtp `
  --spec-draft-n-max 2 `
  --spec-draft-p-min 0.0

For very long prompts, draft-mtp,ngram-mod with --spec-draft-n-max 3 was the better measured high-context profile in this run.

RTX 5090 Windows Benchmark Details

Censorship Smoke Test

A short local smoke test against ornith-1.0-35b-aeon-ultimate-uncensored-nvfp4-gguf-mtp.gguf on 2026-06-28 asked for neutral factual summaries of politically sensitive history/current-affairs topics. The model returned direct factual answers with no refusal or evasion markers detected. This is a small smoke test, not a formal safety or truthfulness evaluation.

Source And Credits

• —AEON NVFP4 source: `AEON-7/Ornith-1.0-35B-AEON-Ultimate-Uncensored-NVFP4`
• —MTP block donor: `s-batman/Ornith-1.0-35B-NVFP4-MTP-GGUF`
• —Base lineage: `deepreinforce-ai/Ornith-1.0-35B`
• —Local benchmark/launcher work: `neko-legends/nvidia-local-llm-profiles`

Responsible Use

This is an uncensored/abliterated model family. You are responsible for downstream usage, deployment policy, and any application-level safeguards. Older llama.cpp builds may not load current GGUF/NVFP4 files correctly.