# alibaba/rtp-llm

RTP-LLM: Alibaba's high-performance LLM inference engine for diverse applications.

Repository: https://github.com/alibaba/rtp-llm
Canonical: https://ross.abutalabs.com/products/rtp-llm
Language: Cuda
License: Apache-2.0
License Family: permissive
Topics: gpt, inference, llama, llm, llm-serving, llmops, model-serving
Last push: 2026-08-26T18:06:34+00:00

## Health v2 (maintenance only)
Score: 66/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 99, release rhythm 22, longevity 70
- inputs: {"age_days": 980, "days_push": 7, "days_rel": 306, "gap_med": null, "n_releases_24m": 1}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1316, forks 266 (observed 2026-08-28T04:04:20.925423+00:00)

## What it is
RTP-LLM is Alibaba's high-performance LLM inference engine written in C++/CUDA, optimized with kernels like PagedAttention and FlashAttention and supporting INT8/INT4 quantization. It is production-proven across Alibaba business units and serves models via an OpenAI-compatible API.

## Use cases
- serve llama models with low latency on nvidia gpus
- deploy an openai-compatible llm api server
- run quantized int4/int8 llm inference
- batch serve llm requests in production
- self-host large language model inference on v100 gpus

## When to choose
- you need production-grade LLM serving on NVIDIA GPUs, especially V100
- you want built-in quantization and KV cache optimization
- you deploy within Alibaba ecosystem or need prefill/decode separation

## When to avoid
- you need broad multi-hardware support like AMD ROCm or Apple silicon today
- you prefer a Python-native stack like vLLM for extensibility
- you need Windows or macOS support

## Facets
- artifact type: library
- maturity: active
- function: llm-inference, gpu-computing, http-server, caching
- domain: large-language-models, machine-learning, gpu-computing, self-hosted
- platform: cpp, python
- tags: llm-serving, cuda-kernels, quantization, paged-attention, model-deployment, vllm-alternative, linux, gpu, docker

## Member repositories
- alibaba/rtp-llm (main) score 66

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:20.925423+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 2026-08-30T04:48:27.213721+00:00, confidence not recorded.
  - readme: https://github.com/alibaba/rtp-llm (fetched 2026-08-28T04:04:20.925423+00:00, sha 8211f86802d2)
- Data as of 2026-08-30T08:39:29.467469+00:00.