Skip to content

Refactor dynamic LoRA update to fix incorrect handling of variant weight shapes #7844

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
zhyncs merged 8 commits into from
Jul 14, 2025
Merged

Refactor dynamic LoRA update to fix incorrect handling of variant weight shapes #7844

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
bafcec0
Refactor dynamic LoRA update to fix incorrect handling of weight shap…
lifuhuang Jul 8, 2025
4c31ba3
Merge branch 'main' into lifuhuang/lora-buffer
lifuhuang Jul 8, 2025
d837319
Adopt Gemini suggestions.
lifuhuang Jul 8, 2025
0f6e4ee
Increase estimated test time.
lifuhuang Jul 8, 2025
497fd87
Merge branch 'main' into lifuhuang/lora-buffer
lifuhuang Jul 13, 2025
5d10d40
Add documents.
lifuhuang Jul 13, 2025
4606526
Improve doc wording.
lifuhuang Jul 13, 2025
26608ae
Merge branch 'main' into lifuhuang/lora-buffer
lifuhuang Jul 13, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
239 changes: 239 additions & 0 deletions docs/backend/lora.ipynb
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -33,6 +33,10 @@
"\n",
"* `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.\n",
"\n",
"* `max_lora_rank`: The maximum LoRA rank that should be supported. If not specified, it will be automatically inferred from the adapters provided in `--lora-paths`. This argument is needed when you expect to dynamically load adapters of larger LoRA rank after server startup.\n",
"\n",
"* `lora_target_modules`: The union set of all target modules where LoRA should be applied (e.g., `q_proj`, `k_proj`, `gate_proj`). If not specified, it will be automatically inferred from the adapters provided in `--lora-paths`. This argument is needed when you expect to dynamically load adapters of different target modules after server startup.\n",
"\n",
"* `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](https://arxiv.org/pdf/2311.03285) paper.\n",
"\n",
"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."
Expand Down Expand Up @@ -176,6 +180,241 @@
"terminate_process(server_process)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Dynamic LoRA loading"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Basic Usage\n",
"\n",
"Instead of specifying all adapters during server startup via `--lora-paths`. You can also load & unload LoRA adapters dynamically via the `/load_lora_adapter` and `/unload_lora_adapter` API.\n",
"\n",
"(Please note that, currently we still require you to specify at least one adapter in `--lora-paths` to enable the LoRA feature, this limitation will be lifted soon.)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"server_process, port = launch_server_cmd(\n",
" \"\"\"\n",
" python3 -m sglang.launch_server --model-path meta-llama/Meta-Llama-3.1-8B-Instruct \\\n",
" --lora-paths lora0=philschmid/code-llama-3-1-8b-text-to-sql-lora \\\n",
" --cuda-graph-max-bs 2 \\\n",
" --max-loras-per-batch 2 --lora-backend triton \\\n",
" --disable-radix-cache\n",
" \"\"\"\n",
")\n",
"\n",
"url = f\"http://127.0.0.1:{port}\"\n",
"wait_for_server(url)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"response = requests.post(\n",
" url + \"/load_lora_adapter\",\n",
" json={\n",
" \"lora_name\": \"lora1\",\n",
" \"lora_path\": \"Nutanix/Meta-Llama-3.1-8B-Instruct_lora_4_alpha_16\",\n",
" },\n",
")\n",
"\n",
"if response.status_code == 200:\n",
" print(\"LoRA adapter loaded successfully.\", response.json())\n",
"else:\n",
" print(\"Failed to load LoRA adapter.\", response.json())"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"response = requests.post(\n",
" url + \"/generate\",\n",
" json={\n",
" \"text\": [\n",
" \"List 3 countries and their capitals.\",\n",
" \"List 3 countries and their capitals.\",\n",
" ],\n",
" \"sampling_params\": {\"max_new_tokens\": 32, \"temperature\": 0},\n",
" \"lora_path\": [\"lora0\", \"lora1\"],\n",
" },\n",
")\n",
"print(f\"Output from lora0: {response.json()[0]['text']}\")\n",
"print(f\"Output from lora1: {response.json()[1]['text']}\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"response = requests.post(\n",
" url + \"/unload_lora_adapter\",\n",
" json={\n",
" \"lora_name\": \"lora0\",\n",
" },\n",
")"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"response = requests.post(\n",
" url + \"/load_lora_adapter\",\n",
" json={\n",
" \"lora_name\": \"lora2\",\n",
" \"lora_path\": \"pbevan11/llama-3.1-8b-ocr-correction\",\n",
" },\n",
")\n",
"\n",
"if response.status_code == 200:\n",
" print(\"LoRA adapter loaded successfully.\", response.json())\n",
"else:\n",
" print(\"Failed to load LoRA adapter.\", response.json())"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"response = requests.post(\n",
" url + \"/generate\",\n",
" json={\n",
" \"text\": [\n",
" \"List 3 countries and their capitals.\",\n",
" \"List 3 countries and their capitals.\",\n",
" ],\n",
" \"sampling_params\": {\"max_new_tokens\": 32, \"temperature\": 0},\n",
" \"lora_path\": [\"lora1\", \"lora2\"],\n",
" },\n",
")\n",
"print(f\"Output from lora1: {response.json()[0]['text']}\")\n",
"print(f\"Output from lora2: {response.json()[1]['text']}\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"terminate_process(server_process)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Advanced: hosting adapters of different shapes\n",
"\n",
"In some cases, you may want to load LoRA adapters with different ranks or target modules (e.g., `q_proj`, `k_proj`) simultaneously. To ensure the server can accommodate all expected LoRA shapes, it's recommended to explicitly specify `--max-lora-rank` and/or `--lora-target-modules` at startup.\n",
"\n",
"For backward compatibility, SGLang will infer these values from `--lora-paths` if they are not explicitly provided. This means it's safe to omit them **only if** all dynamically loaded adapters share the same shape (rank and target modules) as those in the initial `--lora-paths` or are strictly \"smaller\"."
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"lora0 = \"Nutanix/Meta-Llama-3.1-8B-Instruct_lora_4_alpha_16\" # rank - 4, target modules - q_proj, k_proj, v_proj, o_proj, gate_proj\n",
"lora1 = \"algoprog/fact-generation-llama-3.1-8b-instruct-lora\" # rank - 64, target modules - q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj\n",
"\n",
"\n",
"# The `--target-lora-modules` param below is technically not needed, as the server will infer it from lora0 which already has all the target modules specified.\n",
"# We are adding it here just to demonstrate usage.\n",
"server_process, port = launch_server_cmd(\n",
" f\"\"\"\n",
" python3 -m sglang.launch_server --model-path meta-llama/Meta-Llama-3.1-8B-Instruct \\\n",
" --lora-paths lora0={lora0} \\\n",
" --cuda-graph-max-bs 2 \\\n",
" --max-loras-per-batch 2 --lora-backend triton \\\n",
" --disable-radix-cache\n",
" --max-lora-rank 64\n",
" --lora-target-modules q_proj k_proj v_proj o_proj down_proj up_proj gate_proj\n",
" \"\"\"\n",
")\n",
"\n",
"url = f\"http://127.0.0.1:{port}\"\n",
"wait_for_server(url)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"response = requests.post(\n",
" url + \"/load_lora_adapter\",\n",
" json={\n",
" \"lora_name\": \"lora1\",\n",
" \"lora_path\": lora1,\n",
" },\n",
")\n",
"\n",
"if response.status_code == 200:\n",
" print(\"LoRA adapter loaded successfully.\", response.json())\n",
"else:\n",
" print(\"Failed to load LoRA adapter.\", response.json())"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"url = f\"http://127.0.0.1:{port}\"\n",
"json_data = {\n",
" \"text\": [\n",
" \"List 3 countries and their capitals.\",\n",
" \"AI is a field of computer science focused on\",\n",
" ],\n",
" \"sampling_params\": {\"max_new_tokens\": 32, \"temperature\": 0},\n",
" # The first input uses lora0, and the second input uses lora1\n",
" \"lora_path\": [\"lora0\", \"lora1\"],\n",
"}\n",
"response = requests.post(\n",
" url + \"/generate\",\n",
" json=json_data,\n",
")\n",
"print(f\"Output from lora0: {response.json()[0]['text']}\")\n",
"print(f\"Output from lora1: {response.json()[1]['text']}\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"terminate_process(server_process)"
]
},
{
"cell_type": "markdown",
"metadata": {},
Expand Down
2 changes: 2 additions & 0 deletions docs/backend/server_arguments.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -167,6 +167,8 @@ Please consult the documentation below and [server_args.py](https://github.com/s
| `--lora-paths` | The list of LoRA adapters. You can provide a list of either path in str or renamed path in the format {name}={path}. | None |
| `--max-loras-per-batch` | Maximum number of adapters for a running batch, include base-only request. | 8 |
| `--lora-backend` | Choose the kernel backend for multi-LoRA serving. | triton |
| `--max-lora-rank` | The maximum LoRA rank that should be supported. If not specified, it will be automatically inferred from the adapters provided in `--lora-paths`. This argument is needed when you expect to dynamically load adapters of larger LoRA rank after server startup. | None |
| `--lora-target-modules` | The union set of all target modules where LoRA should be applied (e.g., `q_proj`, `k_proj`, `gate_proj`). If not specified, it will be automatically inferred from the adapters provided in `--lora-paths`. This argument is needed when you expect to dynamically load adapters of different target modules after server startup. | None |

## Kernel backend

Expand Down
Loading
Loading