# thu-pacman/chitu

High-performance inference framework for large language models, focusing on efficiency, flexibility, and availability.

Repository: https://github.com/thu-pacman/chitu
Canonical: https://ross.abutalabs.com/products/chitu
Homepage: https://chitu.ai/
Language: Python
License: Apache-2.0
License Family: permissive
Topics: deepseek, gpu, llm, pytorch, llm-serving, model-serving
Last push: 2026-08-26T01:40:40+00:00

## Health v2 (maintenance only)
Score: 84/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 99, release rhythm 91, longevity 39
- inputs: {"age_days": 559, "days_push": 8, "days_rel": 62, "gap_med": 13.0, "n_releases_24m": 29}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2995, forks 256 (observed 2026-08-28T04:07:36.721498+00:00)

## What it is
Chitu (赤兔) is a high-performance inference framework for large language models, positioned as a production-grade LLM serving engine. It supports diverse hardware including NVIDIA GPUs, Huawei Ascend, and other domestic chips, with scalable deployment from CPU-only to large clusters.

## Use cases
- serve deepseek-r1 671b on a single gpu
- deploy llm inference on huawei ascend 910b
- run quantized fp8 or fp4 large language models
- set up multi-node llm serving cluster with pd disaggregation
- serve qwen or glm models in production
- run llm inference on domestic chinese gpu hardware

## When to choose
- you need production-grade LLM serving on non-NVIDIA or domestic Chinese accelerators
- you want to run very large MoE models like DeepSeek-R1 with limited GPU memory via CPU+GPU hybrid inference
- you need scalable deployment from single GPU to multi-node clusters

## When to avoid
- you need the broadest ecosystem and community plugins, where vLLM or SGLang may be safer
- you only need lightweight local inference on consumer hardware, where llama.cpp fits better
- you require guaranteed commercial support without a service agreement

## Facets
- artifact type: library
- maturity: active
- function: llm-inference, gpu-computing, machine-learning
- domain: large-language-models, deep-learning, artificial-intelligence
- platform: python, cloud
- tags: llm-serving, model-serving, deepseek, quantization, moe, inference-engine, domestic-gpu-support, cpu-gpu-hybrid, linux, gpu, docker

## Member repositories
- thu-pacman/chitu (main) score 84

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:07:36.721498+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 2026-08-30T07:30:50.814156+00:00, confidence not recorded.
  - readme: https://github.com/thu-pacman/chitu (fetched 2026-08-28T04:07:36.721498+00:00, sha 2b6351d6e585)
  - homepage: https://chitu.ai/ (fetched 2026-08-29T09:46:05.152526+00:00, sha 77fd50b36aec)
  - registry_pypi: https://pypi.org/pypi/chitu/json (fetched 2026-08-29T09:46:05.155408+00:00, sha 7097fc0b1856)
- Data as of 2026-08-30T08:39:29.467469+00:00.