sglang/docs/backend/lora.ipynb

LoRA Serving

SGLang enables the use of LoRA adapters with a base model. By incorporating techniques from S-LoRA and Punica, SGLang can efficiently support multiple LoRA adapters for different sequences within a single batch of inputs.

Arguments for LoRA Serving

The following server arguments are relevant for multi-LoRA serving:

• lora_paths: A mapping from each adaptor's name to its path, in the form of {name}={path} {name}={path}.

• max_loras_per_batch: Maximum number of adaptors used by each batch. This argument can affect the amount of GPU memory reserved for multi-LoRA serving, so it should be set to a smaller value when memory is scarce. Defaults to be 8.

• lora_backend: The backend of running GEMM kernels for Lora modules. It can be one of triton or flashinfer, and set to triton by default. For better performance and stability, we recommend using the Triton LoRA backend. In the future, faster backend built upon Cutlass or Cuda kernels will be added.

• tp_size: LoRA serving along with Tensor Parallelism is supported by SGLang. tp_size controls the number of GPUs for tensor parallelism. More details on the tensor sharding strategy can be found in S-Lora paper.

From client side, the user needs to provide a list of strings as input batch, and a list of adaptor names that each input sequence corresponds to.

Usage

Serving Single Adaptor

In [ ]:

from sglang.test.test_utils import is_in_ci

if is_in_ci():
    from patch import launch_server_cmd
else:
    from sglang.utils import launch_server_cmd

from sglang.utils import wait_for_server, terminate_process

import json
import requests

In [ ]:

server_process, port = launch_server_cmd(
    """
python3 -m sglang.launch_server --model-path meta-llama/Meta-Llama-3.1-8B-Instruct \
    --lora-paths lora0=algoprog/fact-generation-llama-3.1-8b-instruct-lora \
    --max-loras-per-batch 1 --lora-backend triton \
    --disable-cuda-graph --disable-radix-cache
"""
)

wait_for_server(f"http://localhost:{port}")

In [ ]:

url = f"http://127.0.0.1:{port}"
json_data = {
    "text": [
        "List 3 countries and their capitals.",
        "AI is a field of computer science focused on",
    ],
    "sampling_params": {"max_new_tokens": 32, "temperature": 0},
    # The first input uses lora0, and the second input uses the base model
    "lora_path": ["lora0", None],
}
response = requests.post(
    url + "/generate",
    json=json_data,
)
print(f"Output 0: {response.json()[0]['text']}")
print(f"Output 1: {response.json()[1]['text']}")

In [ ]:

terminate_process(server_process)

Serving Multiple Adaptors

In [ ]:

server_process, port = launch_server_cmd(
    """
python3 -m sglang.launch_server --model-path meta-llama/Meta-Llama-3.1-8B-Instruct \
    --lora-paths lora0=algoprog/fact-generation-llama-3.1-8b-instruct-lora \
    lora1=Nutanix/Meta-Llama-3.1-8B-Instruct_lora_4_alpha_16 \
    --max-loras-per-batch 2 --lora-backend triton \
    --disable-cuda-graph --disable-radix-cache
"""
)

wait_for_server(f"http://localhost:{port}")

In [ ]:

url = f"http://127.0.0.1:{port}"
json_data = {
    "text": [
        "List 3 countries and their capitals.",
        "AI is a field of computer science focused on",
    ],
    "sampling_params": {"max_new_tokens": 32, "temperature": 0},
    # The first input uses lora0, and the second input uses lora1
    "lora_path": ["lora0", "lora1"],
}
response = requests.post(
    url + "/generate",
    json=json_data,
)
print(f"Output 0: {response.json()[0]['text']}")
print(f"Output 1: {response.json()[1]['text']}")

In [ ]:

terminate_process(server_process)

Future Works

The development roadmap for LoRA-related features can be found in this issue. Currently Cuda graph and radix attention are not incompatible with LoRA and must be manually disabled. Other features, including Unified Paging, Cutlass backend, and dynamic loading/unloadingm, are still under development.