Skip to main content

Serving Multi-LoRA Models


In a world where the demand for highly specialized AI capabilities is surging, the ability to deploy multiple customized large language models (LLMs) without additional GPU resources represents a significant leap forward. The Friendli Engine introduces an innovative approach to this challenge through Multi-LoRA (Low-Rank Adaptation) serving, a method that allows for the simultaneous serving of multiple LLMs, optimized for specific tasks without the need for extensive retraining. This advancement opens new avenues for AI efficiency and adaptability, promising to revolutionize the deployment of AI solutions on constrained hardware. This article provides an overview of efficient serving Multi-LoRA models with the Friendli Engine.

Lora Serving


huggingface-cli should be installed in your local environment.

pip install "huggingface_hub[cli]"

Downloading Adapter Checkpoints

For each adapter model that you want to server, you have to download in your local storage.

export ADAPTER_MODEL1=""  # Hugging Face model name of the first adapter
export ADAPTER_MODEL2="" # Hugging Face model name of the second adapter
export ADAPTER_MODEL3="" # Hugging Face model name of the third adapter
export ADAPTER_DIR=/tmp/adapter

huggingface-cli download $ADAPTER_MODEL1 --include "adapter_model.safetensors" "adapter_config.json" --local-dir $ADAPTER_DIR/model1
huggingface-cli download $ADAPTER_MODEL2 --include "adapter_model.safetensors" "adapter_config.json" --local-dir $ADAPTER_DIR/model2
huggingface-cli download $ADAPTER_MODEL3 --include "adapter_model.safetensors" "adapter_config.json" --local-dir $ADAPTER_DIR/model3

If an adapter's Hugging Face repo does not contain adapter_model.safetensors checkpoint file, you have to manually convert adapter_model.bin into adapter_model.safetensors. You can use the official app or the python script for conversion.

Configuring Adapter

To run Multi-LoRA with the Friendli Engine, you need to prepare a file named adapter_attr.yaml that informs the Friendli Engine of the names and the checkpoint paths of adapters. The following is an example.

- name: FILL ME
ckpt-path: FILL ME
type: lora
  • The name field specifies the adapter's name (a user defined alias). It should be unique for each adapter.
  • The ckpt-path field specifies the absolute path to the adapter checkpoint directory. The directory should contain two files: adapter_model.safetensors and adapter_config.json.

In order to serve multiple adapters at once, configurations for all adapters should be listed.

Launch Friendli Engine in Container

When you have prepared adapter model checkpoints, now you can serve the Multi-LoRA model with Friendli Container. In addition to the command for running the base model, you have to add the --adapter-config-path argument.

  • --adapter-config-path: The path to the adapter_attr.yaml file.
# Fill the values of following variables.
export HF_BASE_MODEL_NAME="" # Hugging Face base model name (e.g., "meta-llama/Llama-2-7b-chat-hf")
export FRIENDLI_CONTAINER_SECRET="" # Friendli container secret
export FRIENDLI_CONTAINER_IMAGE="" # Friendli container image (e.g., "")
export GPU_ENUMERATION="" # GPUs (e.g., '"device=0,1"')

docker run \
-p 8000:8000 \
-v $ADAPTER_DIR:/adapter \
--hf-model-name $HF_BASE_MODEL_NAME \
--adapter-config-path /adapter/adapter_attr.yaml \

You can find available options for [LAUNCH_OPTIONS] at Running Friendli Container: Launch Options.

Example: Llama 2 7B Chat + LoRA Adapter

This is an example that runs meta-llama/Llama-2-7b-chat-hf with FinGPT/fingpt-forecaster_dow30_llama2-7b_lora adapter model.

export ADAPTER_DIR=/tmp/adapter

huggingface-cli download FinGPT/fingpt-forecaster_dow30_llama2-7b_lora --include "adapter_model.safetensors" "adapter_config.json" --local-dir $ADAPTER_DIR/model1

docker run \
--gpus '"device=0"' \
-p 8000:8000 \
-v $ADAPTER_DIR:/adapter \
--hf-model-name meta-llama/Llama-2-7b-chat-hf \
--adapter-config-path /adapter/adapter_attr.yaml

You have to manually create $ADAPTER_DIR/adapter_attr.yaml file with the following content:

- name: my-model
type: lora
ckpt-path: /adapter/model1

Sending Request to Specific Adapter

You can generate an inference result from a specific adapter model by specifying model in the body of an inference request. For example, assuming you set the name field of adapter_attr.yaml to "my-model", you can send a request to the adapter model as follows.

curl \
-H "Content-Type: application/json" \
-d '{
"prompt": "Python is a language",
"model": "my-model"
}' \

Sending Request to the Base Model

If you omit the model field in your request, the base model will be used for generating an inference request. You can send a request to the base model as shown below.

curl \
-H "Content-Type: application/json" \
-d '{
"prompt": "Python is a language",
}' \


  • We only support models compatible with peft.
  • Base model checkpoint and adapter model checkpoint should have the same datatype.
  • When serving multiple adapters simultaneously, each adapter model should have the same target modules. In Hugging Face, the target modules are listed at adapter_config.json.