# Neroued/ninfer

High-performance single-GPU inference for selected model checkpoints and GPUs.

Repository: https://github.com/Neroued/ninfer
Canonical: https://ross.abutalabs.com/products/ninfer
Language: C++
License: Apache-2.0
License Family: permissive
Last push: 2026-09-02T17:14:09+00:00

## Health v2 (maintenance only)
Score: 58/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 100, release rhythm 35, longevity 4
- inputs: {"age_days": 68, "days_push": 0, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, young
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1252, forks 241 (observed 2026-09-03T02:15:06.881349+00:00)

## What it is
NInfer is a from-scratch C++/CUDA inference engine optimized for maximum single-GPU performance on a narrow, explicitly registered set of Qwen model checkpoints, targeting an NVIDIA RTX 5090. It supports text, image, and video prompts via a local CLI and OpenAI-/Anthropic-compatible HTTP APIs.

## Use cases
- run Qwen models at maximum speed on a single RTX 5090
- serve Qwen checkpoints through an OpenAI-compatible API
- run multimodal image and video prompts locally
- decode with high concurrency using MTP and CUDA graphs
- use NVFP4-quantized model weights for faster inference
- self-host a minimal LLM inference server

## When to choose
- you own an RTX 5090 and want peak single-GPU throughput for supported Qwen checkpoints
- you want a lightweight, purpose-built engine instead of a general model runtime
- you need OpenAI- or Anthropic-compatible local serving endpoints
- you want multimodal (text, image, video) prompting with quantized weights

## When to avoid
- you need to run arbitrary models or checkpoints outside the registered set
- you have a different GPU than the targeted NVIDIA hardware
- you need multi-GPU or distributed inference
- you want broad ecosystem compatibility like vLLM or llama.cpp

## Facets
- artifact type: library
- maturity: active
- function: llm-inference, http-server, cli, gpu-computing
- domain: large-language-models, machine-learning, developer-tools
- platform: cpp, cli
- tags: cuda, qwen, single-gpu, openai-compatible-api, anthropic-compatible-api, quantization, nvfp4, rtx-5090, multimodal, mtp, linux, gpu, web-server

## Member repositories
- Neroued/ninfer (main) score 58

## Provenance
- Observed fields: from GitHub, fetched 2026-09-03T02:15:06.881349+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T06:58:44.643368+00:00, confidence not recorded.
  - readme: https://github.com/Neroued/ninfer (fetched 2026-09-03T02:15:06.881349+00:00, sha 522c12f36bb7)
- Data as of 2026-08-30T08:39:29.467469+00:00.