LMDeploy review
Free, open-source runtime for optimized multi-GPU and multimodal model serving.
Reviewed by iTechGuides Editors · Editorial team · Updated Nov 2026
LMDeploy is an open-source toolkit for local inference and serving of large language models and vision-language models. It fits teams that need self-hosted or cloud deployment, from solo developers through enterprise groups, with particular value for Linux teams operating optimized multi-GPU workloads. Windows deployment is also documented, and the project supports multiple model checkpoint formats and both NVIDIA and AMD GPU setups.
The software is free and distributed under the Apache 2.0 license. Its core serving options combine TurboMind and PyTorch inference engines with weight-only and KV-cache quantization, continuous batching, blocked KV cache, and tensor, data, and context parallelism. These controls target throughput and memory management in demanding serving environments. LoRA adapter serving extends the runtime for adapted models, while vision-language serving accepts image and video inputs. The command-line interface covers model conversion, compression, deployment, interactive chat, and environment checks, giving operators a single toolkit for common model lifecycle tasks.
LMDeploy also provides an OpenAI-compatible HTTP API server and a distributed multi-model request server, which can simplify application integration and multi-model deployments. Documentation covers Linux and Windows operation, multimodal inputs, distributed serving, and formats including HF, AWQ, GPTQ, compressed-tensors, FP8, and MXFP4. Support is centered on documentation and community channels. Teams wanting focused control over inference engines, quantization, batching, and parallelism should consider LMDeploy; teams requiring wider platform coverage or a larger ecosystem presence may prefer an alternative runtime.
LMDeploy pros and cons
- Where it wins
- TurboMind and PyTorch engines with tensor, data, and context parallelism
- Continuous batching, blocked KV cache, and weight-only or KV-cache quantization
- OpenAI-compatible API plus multimodal, LoRA, and distributed serving
- Where it doesn't
- Platform coverage centers on Linux and Windows deployments
- Support channels are limited to documentation and community resources
- Its ecosystem presence is narrower than more broadly distributed runtimes
LMDeploy fact sheet, pricing and score →
Advertiser disclosure: iTechGuides is reader-supported. Vendors can pay for top positions in our rankings and for a place on other products' pages, and we may earn a commission when you click some links. Bought places stand at the number bought; this page and our disclosure say so. Paying vendors may get a fuller, more detailed review from our editors, and that review can change their score. How we rank.
Last updated · How we research and update