# deepspeedai/DeepSpeed-MII

MII makes low-latency and high-throughput inference possible, powered by DeepSpeed.

Repository: https://github.com/deepspeedai/DeepSpeed-MII
Canonical: https://ross.abutalabs.com/products/deepspeed-mii
Language: Python
License: Apache-2.0
License Family: permissive
Topics: deep-learning, inference, pytorch
Last push: 2025-06-30T16:21:45+00:00

## Health v2 (maintenance only)
Score: 43/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 29, release rhythm 28, longevity 100
- inputs: {"age_days": 1624, "days_push": 429, "days_rel": 526, "gap_med": 72.5, "n_releases_24m": 3}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2111, forks 192 (observed 2026-08-28T04:06:14.306550+00:00)

## What it is
DeepSpeed-MII is a Python library for high-throughput, low-latency large language model inference, built on DeepSpeed. It provides blocked KV-caching, continuous batching, Dynamic SplitFuse, and tensor parallelism with optimized CUDA kernels for serving LLMs like Llama, Mixtral, and Phi.

## Use cases
- serve llama-70b with high throughput text generation
- run low-latency llm inference on gpus
- deploy mixtral moe model for inference
- benchmark llm serving throughput against vllm
- generate text from phi-2 efficiently
- reduce cost of llm inference serving

## When to choose
- you need high-throughput batch text generation with DeepSpeed-optimized kernels
- you want continuous batching and SplitFuse for serving LLMs on NVIDIA GPUs
- you are already in the DeepSpeed ecosystem

## When to avoid
- you need multi-vendor GPU or CPU-only inference
- you want the broadest model coverage or largest community (vLLM may fit better)
- you need a fully managed inference service rather than a library

## Facets
- artifact type: library
- maturity: active
- function: llm-inference, machine-learning, gpu-computing
- domain: large-language-models, deep-learning, machine-learning
- platform: python
- tags: deepspeed, text-generation, continuous-batching, tensor-parallelism, kv-cache, pytorch, gpu, linux

## Member repositories
- deepspeedai/DeepSpeed-MII (main) score 43

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:06:14.306550+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 2026-08-30T02:53:43.436369+00:00, confidence not recorded.
  - readme: https://github.com/deepspeedai/DeepSpeed-MII (fetched 2026-08-28T04:06:14.306550+00:00, sha 4eba83b881f0)
  - registry_pypi: https://pypi.org/pypi/deepspeed-mii/json (fetched 2026-08-29T10:33:45.046319+00:00, sha d21c4f5a8937)
- Data as of 2026-08-30T08:39:29.467469+00:00.