NInferEZby 2beng2
THE NINFEREZ COLLECTION

Meet your next
local model.

Native NInfer conversions, ready to explore in Manager.
Choose by encoding, components, and the hardware you have.

Download Manager →Visit Hugging Face

Showing 4 artifacts

SNVFP4

Swift 1.5

Qwen3.8 · 27B

A complete text and vision artifact, with native NVFP4 weights and a bundled speculative draft for compatible Blackwell GPUs.

TextVisionMTPDFlash2
21.22 GiBArtifact size
Compatibility & license +

Native NVFP4 requires a compatible Blackwell target. Verified on RTX 5090; choose context to fit available VRAM.

License: Swift Open License v1.0 + component licenses

Download model ↓Model card & credits on Hugging Face ↗

Original weights: UkisAI · Swift 1.5 / ajgazin · NVFP4 source.
NInfer conversion by 2beng2. Full lineage and component credits in the model card.

OIQ3_XXS

OrcaRouter

Qwen3.8 · 27B · MTP

The compact OrcaRouter variant. Preserves source IQ3_XXS weights and includes MTP for speculative decoding.

TextMTP
10.06 GiBArtifact size
Compatibility & license +

Text only; the source vision projector is not included. Needs a compatible groupwise-int engine.

License: Apache 2.0

Download model ↓Model card & credits on Hugging Face ↗

Original weights: RentedNoodle · OrcaRouter source.
NInfer conversion by 2beng2. Full lineage and component credits in the model card.

OIQ3_XXS

OrcaRouter

Qwen3.8 · 27B · DFlash2

Adds a DFlash2 draft to OrcaRouter, with MTP available as an alternative. Explore the published long-context measurements.

TextMTPDFlash2
12.13 GiBArtifact size
Compatibility & license +

Text only. Long-context measurements used a 32 GB RTX 5090; configured capacity is not actual prompt length.

License: Apache 2.0

Download model ↓Model card & credits on Hugging Face ↗

Original weights: RentedNoodle · OrcaRouter source.
NInfer conversion by 2beng2. Full lineage and component credits in the model card.

HIQ2_XS

Heretic

Qwen3.8 · 27B

Our smallest featured 27B artifact. A text-only IQ2_XS conversion for a more compact model library.

Text
7.84 GiBArtifact size
Compatibility & license +

Artifact size is not a VRAM requirement. Check the model card and Manager’s memory estimate before loading.

License: See model card and original weight licenses

Download model ↓Model card & credits on Hugging Face ↗

Original weights: 0bserverx · Heretic source.
NInfer conversion by 2beng2. Full lineage and component credits in the model card.

File size is one part of the picture.

KV cache, draft components, runtime state, and workspace all need memory. Use Manager’s VRAM estimate and each model card to choose your context. Native NVFP4 requires a compatible Blackwell GPU.

KNOW WHAT YOU ARE DOWNLOADING

Clear origins.
Useful details.

These are conversions and packaged artifacts, built from existing upstream weights. They retain the original models’ authorship and applicable licenses.

Each model card explains components, provenance, license, and the available measurements. Benchmark results describe the tested artifact and configuration.

New to the format?

NInfer artifacts are different from GGUF or Transformers checkpoints. The Converter can prepare supported sources; the Engine runs compatible artifacts.

Prepare a model Check engine compatibility Start with Manager
YOUR NEXT STEP

Put your GPU to work.

Your models, your workspace, your next idea.
Start with NInferEZ Manager.

Download for WindowsAll downloads →
Windows x64 · Installer or Portable