AI Tools
LoRAX
LoRAX è un server di inferenza open source per servire più modelli LoRA sottoposti a fine-tuning tramite un’infrastruttura GPU condivisa.

In inglese
Key facts
- Price
- Free
- Free plan
- Not stated on the maker’s pages
- Platforms
- Linux · CLI · API
- API
- Yes
- Open source
- Yes · Apache-2.0 Read on github.com
- Latest version
- lorax-0.4.0
- Prices checked
- 19 Nov 2026
LoRAX serves requests for different LoRA adapters on a shared base model. It dynamically loads adapters, batches requests across adapters, and can move adapters between GPU and CPU memory. The project is aimed at developers and ML teams deploying fine-tuned language models.
It provides REST and OpenAI-compatible APIs, plus Docker-based deployment. It is Apache-2.0 licensed; its GitHub repository was archived and made read-only on October 12, 2026. The former Predibase pricing address now redirects to a Rubrik page that does not list LoRAX prices.
- Category
- Developer Tools
- Read from the maker’s pages
- 2026-11-18
Company
Features
- Dynamically loads LoRA adapters from Hugging Face, Predibase, or a filesystem
- Batches requests for different adapters together
- Merges adapters per request
- Moves adapters asynchronously between GPU and CPU memory
- Supports tensor parallelism, CUDA kernels, quantization, and token streaming
- Provides REST and OpenAI-compatible APIs
- Apache-2.0 license
Use cases
- Serve multiple fine-tuned model variants from shared GPU infrastructure
- Route requests to different LoRA adapters in a multi-tenant service
- Test or compare adapter variants behind one inference server
- Deploy a model-serving endpoint with Docker
- Connect an application using an OpenAI-compatible API
Pricing
Free
- Prices checked
- 2026-11-19
- Read
- by web search
Open Source
free
- Apache 2.0 license
- Multi-LoRA inference server for LLMs
- Serve 100s-1000s of fine-tuned adapters on one GPU
- Dynamic adapter loading, OpenAI-compatible API
- Docker/Helm/K8s ready, production features (metrics, tracing)
Capabilities
- Choice of models — “Just specify any adapter as the model parameter.” source
- Self-hosted — “We recommend starting with our pre-built Docker image to avoid compiling custom CUDA kernels and other dependencies.” source
- API — “LoRAX supports multi-turn chat conversations combined with dynamic adapter loading through an OpenAI compatible API.” source
- Official SDKs — “For a full tutorial including token streaming and the Python client, see Getting Started - Docker .” source
- Runs models for you — “LoRAX (LoRA eXchange) is a framework that allows users to serve thousands of fine-tuned models on a single GPU, dramatically reducing the cost of serving without compromising on throughput or latency.” source
- Traces and evaluates — “prebuilt Docker images, Helm charts for Kubernetes, Prometheus metrics, and distributed tracing with Open Telemetry.” source
Where it works
Linux · CLI · API
Alternatives to LoRAX
- Text Generation Inference
Text Generation Inference is an open-source toolkit for deploying and serving large language models through an API.
- vLLM
vLLM runs and serves large language models on supported hardware, with APIs for applications to send inference requests.
- Baseten
Baseten provides infrastructure to deploy and run AI models, train models, and build agent workflows using managed inference, APIs, training tools, and sandboxes.
- llama.cpp
Run language and vision-language models on CPU or GPU with a C/C++ inference runtime and optional API server.
- Fireworks AI
Fireworks AI hosts, routes, and trains open and custom models for applications, teams, and research.
- CoreWeave Cloud
Cloud infrastructure for training and serving AI, with GPU and CPU compute, storage, networking, and managed Kubernetes.
More from Predibase
- Ludwig
Open-source framework for training, fine-tuning, evaluating, and serving AI models with YAML configurations.
Questions about LoRAX
How much does LoRAX cost?
The plans on the maker’s pricing page: Open Source (free). These prices were read by web search, not on the maker’s own page: check its pricing page before relying on them. Prices checked on 19 Nov 2026.
Does LoRAX have an API?
Yes. The maker’s pages state that LoRAX has an API. Read on the maker’s pages on 18 Nov 2026.
Which platforms does LoRAX run on?
LoRAX runs on Linux, CLI and API, as its maker’s pages state. Read on the maker’s pages on 18 Nov 2026.
Is LoRAX open source?
Yes. The code of LoRAX is published at github.com/predibase/lorax under the Apache-2.0 licence. Read on the maker’s pages on 18 Nov 2026.
Is LoRAX still available?
Yes, as of 18 Nov 2026: its maker’s pages were read that day and carried no notice that LoRAX has ended.
Sources
- https://www.rubrik.com/products/rubrik-agent-cloud — The former Predibase pricing address redirects here; the page is for Rubrik Agent Cloud and does not show LoRAX pricing.
- https://loraexchange.ai/ — LoRAX documentation describes adapter loading, batching, adapter exchange, and inference capabilities.
- https://github.com/predibase/lorax — Project repository identifies the Apache-2.0 license and archived, read-only status.
- https://github.com/predibase/lorax/releases — Release page marks lorax-0.4.0 as the latest release and lists v0.12.1.
Not confirmed
- Prices
- the former pricing address redirects to an unrelated Rubrik product page, not a LoRAX pricing page
- no LoRAX tiers or prices were available on the redirected page
- no current LoRAX paid price could be read
- Free plan
- no current pricing page was available to confirm a free tier