Files
sglang/docs/backend/openai_api_vision.ipynb
T

18 KiB

OpenAI-Compatible APIs - Vision

SGLang provides OpenAI-compatible APIs to enable a smooth transition from OpenAI services to self-hosted local models. A complete reference for the API is available in the OpenAI API Reference. This tutorial covers the vision APIs for vision language models.

SGLang supports vision language models such as Llama 3.2, LLaVA-OneVision, and QWen-VL2

Launch A Server

This code block is equivalent to executing

python3 -m sglang.launch_server --model-path meta-llama/Llama-3.2-11B-Vision-Instruct \
  --port 30010 --chat-template llama_3_vision

in your terminal and wait for the server to be ready.

Remember to add --chat-template llama_3_vision to specify the vision chat template, otherwise the server only supports text. We need to specify --chat-template for vision language models because the chat template provided in Hugging Face tokenizer only supports text.

In [13]:
from sglang.utils import (
    execute_shell_command,
    wait_for_server,
    terminate_process,
    print_highlight,
)

embedding_process = execute_shell_command(
"""
python3 -m sglang.launch_server --model-path meta-llama/Llama-3.2-11B-Vision-Instruct \
    --port=30010 --chat-template=llama_3_vision
"""
)

wait_for_server("http://localhost:30010")
2024-11-02 00:24:10.542705: E external/local_xla/xla/stream_executor/cuda/cuda_fft.cc:479] Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered
2024-11-02 00:24:10.554725: E external/local_xla/xla/stream_executor/cuda/cuda_dnn.cc:10575] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
2024-11-02 00:24:10.554758: E external/local_xla/xla/stream_executor/cuda/cuda_blas.cc:1442] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered
2024-11-02 00:24:11.063662: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Could not find TensorRT
[2024-11-02 00:24:19] server_args=ServerArgs(model_path='meta-llama/Llama-3.2-11B-Vision-Instruct', tokenizer_path='meta-llama/Llama-3.2-11B-Vision-Instruct', tokenizer_mode='auto', skip_tokenizer_init=False, load_format='auto', trust_remote_code=False, dtype='auto', kv_cache_dtype='auto', quantization=None, context_length=None, device='cuda', served_model_name='meta-llama/Llama-3.2-11B-Vision-Instruct', chat_template='llama_3_vision', is_embedding=False, host='127.0.0.1', port=30010, mem_fraction_static=0.88, max_running_requests=None, max_total_tokens=None, chunked_prefill_size=8192, max_prefill_tokens=16384, schedule_policy='lpm', schedule_conservativeness=1.0, tp_size=1, stream_interval=1, random_seed=553831757, constrained_json_whitespace_pattern=None, decode_log_interval=40, log_level='info', log_level_http=None, log_requests=False, show_time_cost=False, api_key=None, file_storage_pth='SGLang_storage', enable_cache_report=False, watchdog_timeout=600, dp_size=1, load_balance_method='round_robin', dist_init_addr=None, nnodes=1, node_rank=0, json_model_override_args='{}', enable_double_sparsity=False, ds_channel_config_path=None, ds_heavy_channel_num=32, ds_heavy_token_num=256, ds_heavy_channel_type='qk', ds_sparse_decode_threshold=4096, lora_paths=None, max_loras_per_batch=8, attention_backend='flashinfer', sampling_backend='flashinfer', grammar_backend='outlines', disable_flashinfer=False, disable_flashinfer_sampling=False, disable_radix_cache=False, disable_regex_jump_forward=False, disable_cuda_graph=False, disable_cuda_graph_padding=False, disable_disk_cache=False, disable_custom_all_reduce=False, disable_mla=False, disable_penalizer=False, disable_nan_detection=False, enable_overlap_schedule=False, enable_mixed_chunk=False, enable_torch_compile=False, torch_compile_max_bs=32, cuda_graph_max_bs=160, torchao_config='', enable_p2p_check=False, triton_attention_reduce_in_fp32=False, num_continuous_decode_steps=1)
[2024-11-02 00:24:20] Use chat template for the OpenAI-compatible API server: llama_3_vision
[2024-11-02 00:24:29 TP0] Automatically turn off --chunked-prefill-size and adjust --mem-fraction-static for multimodal models.
[2024-11-02 00:24:29 TP0] Init torch distributed begin.
[2024-11-02 00:24:32 TP0] Load weight begin. avail mem=76.83 GB
[2024-11-02 00:24:32 TP0] lm_eval is not installed, GPTQ may not be usable
INFO 11-02 00:24:32 weight_utils.py:243] Using model weights format ['*.safetensors']
Loading safetensors checkpoint shards:   0% Completed | 0/5 [00:00<?, ?it/s]
Loading safetensors checkpoint shards:  20% Completed | 1/5 [00:00<00:02,  1.61it/s]
Loading safetensors checkpoint shards:  40% Completed | 2/5 [00:02<00:04,  1.35s/it]
Loading safetensors checkpoint shards:  60% Completed | 3/5 [00:04<00:03,  1.58s/it]
Loading safetensors checkpoint shards:  80% Completed | 4/5 [00:06<00:01,  1.70s/it]
Loading safetensors checkpoint shards: 100% Completed | 5/5 [00:08<00:00,  1.76s/it]
Loading safetensors checkpoint shards: 100% Completed | 5/5 [00:08<00:00,  1.62s/it]

[2024-11-02 00:24:41 TP0] Load weight end. type=MllamaForConditionalGeneration, dtype=torch.bfloat16, avail mem=56.75 GB
[2024-11-02 00:24:41 TP0] Memory pool end. avail mem=11.53 GB
[2024-11-02 00:24:42 TP0] Capture cuda graph begin. This can take up to several minutes.
[2024-11-02 00:24:52 TP0] max_total_num_tokens=289349, max_prefill_tokens=16384, max_running_requests=2049, context_len=131072
[2024-11-02 00:24:52] INFO:     Started server process [108249]
[2024-11-02 00:24:52] INFO:     Waiting for application startup.
[2024-11-02 00:24:52] INFO:     Application startup complete.
[2024-11-02 00:24:52] INFO:     Uvicorn running on http://127.0.0.1:30010 (Press CTRL+C to quit)
[2024-11-02 00:24:53] INFO:     127.0.0.1:43056 - "GET /v1/models HTTP/1.1" 200 OK
[2024-11-02 00:24:53] INFO:     127.0.0.1:43072 - "GET /get_model_info HTTP/1.1" 200 OK
[2024-11-02 00:24:53 TP0] Prefill batch. #new-seq: 1, #new-token: 7, #cached-token: 0, cache hit rate: 0.00%, token usage: 0.00, #running-req: 0, #queue-req: 0
[2024-11-02 00:24:53] INFO:     127.0.0.1:43086 - "POST /generate HTTP/1.1" 200 OK
[2024-11-02 00:24:53] The server is fired up and ready to roll!


NOTE: Typically, the server runs in a separate terminal.
In this notebook, we run the server and notebook code together, so their outputs are combined.
To improve clarity, the server logs are displayed in the original black color, while the notebook outputs are highlighted in blue.

Using cURL

Once the server is up, you can send test requests using curl.

In [15]:
import subprocess

curl_command = """
curl http://localhost:30010/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer None" \
  -d '{
    "model": "meta-llama/Llama-3.2-11B-Vision-Instruct",
    "messages": [
      {
        "role": "user",
        "content": [
          {
            "type": "text",
            "text": "What’s in this image?"
          },
          {
            "type": "image_url",
            "image_url": {
              "url": "https://github.com/sgl-project/sglang/blob/main/test/lang/example_image.png?raw=true"
            }
          }
        ]
      }
    ],
    "max_tokens": 300
  }'
"""

response = subprocess.check_output(curl_command, shell=True).decode()
print_highlight(response)
  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                 Dload  Upload   Total   Spent    Left  Speed
100   485    0     0  100   485      0   2420 --:--:-- --:--:-- --:--:--  2412
[2024-11-02 00:26:23 TP0] Prefill batch. #new-seq: 1, #new-token: 1, #cached-token: 6462, cache hit rate: 49.97%, token usage: 0.02, #running-req: 0, #queue-req: 0
[2024-11-02 00:26:24] INFO:     127.0.0.1:39828 - "POST /v1/chat/completions HTTP/1.1" 200 OK
100   965  100   480  100   485    789    797 --:--:-- --:--:-- --:--:--  1584
{"id":"5e9e1c80809f492a926a2634c3d162d0","object":"chat.completion","created":1730507184,"model":"meta-llama/Llama-3.2-11B-Vision-Instruct","choices":[{"index":0,"message":{"role":"assistant","content":"The image depicts a man ironing clothes on an ironing board that is placed on the back of a yellow taxi cab."},"logprobs":null,"finish_reason":"stop","matched_stop":128009}],"usage":{"prompt_tokens":6463,"total_tokens":6489,"completion_tokens":26,"prompt_tokens_details":null}}

Using OpenAI Python Client

You can use the OpenAI Python API library to send requests.

In [16]:
from openai import OpenAI

client = OpenAI(base_url="http://localhost:30010/v1", api_key="None")

response = client.chat.completions.create(
    model="meta-llama/Llama-3.2-11B-Vision-Instruct",
    messages=[
        {
            "role": "user",
            "content": [
                {
                    "type": "text",
                    "text": "What is in this image?",
                },
                {
                    "type": "image_url",
                    "image_url": {"url": "https://github.com/sgl-project/sglang/blob/main/test/lang/example_image.png?raw=true"},
                },
            ],
        }
    ],
    max_tokens=300,
)

print_highlight(response.choices[0].message.content)
[2024-11-02 00:26:33 TP0] Prefill batch. #new-seq: 1, #new-token: 11, #cached-token: 6452, cache hit rate: 66.58%, token usage: 0.02, #running-req: 0, #queue-req: 0
[2024-11-02 00:26:34 TP0] Decode batch. #running-req: 1, #token: 6477, token usage: 0.02, gen throughput (token/s): 0.77, #queue-req: 0
[2024-11-02 00:26:34] INFO:     127.0.0.1:43258 - "POST /v1/chat/completions HTTP/1.1" 200 OK
The image shows a man ironing clothes on the back of a yellow taxi cab.

Multiple-Image Inputs

The server also supports multiple images and interleaved text and images if the model supports it.

In [11]:
from openai import OpenAI

client = OpenAI(base_url="http://localhost:30010/v1", api_key="None")

response = client.chat.completions.create(
    model="meta-llama/Llama-3.2-11B-Vision-Instruct",
    messages=[
        {
            "role": "user",
            "content": [
                {
                    "type": "image_url",
                    "image_url": {
                        "url": "https://github.com/sgl-project/sglang/blob/main/test/lang/example_image.png?raw=true",
                    },
                },
                {
                    "type": "image_url",
                    "image_url": {
                        "url": "https://raw.githubusercontent.com/sgl-project/sglang/main/assets/logo.png",
                    },
                },
                {
                    "type": "text",
                    "text": "I have two very different images. They are not related at all. "
                            "Please describe the first image in one sentence, and then describe the second image in another sentence.",
                },
            ],
        }
    ],
    temperature=0,
)

print_highlight(response.choices[0].message.content)
[2024-11-02 00:20:30 TP0] Prefill batch. #new-seq: 1, #new-token: 1, #cached-token: 12894, cache hit rate: 83.27%, token usage: 0.04, #running-req: 0, #queue-req: 0
[2024-11-02 00:20:30 TP0] Decode batch. #running-req: 1, #token: 12903, token usage: 0.04, gen throughput (token/s): 2.02, #queue-req: 0
[2024-11-02 00:20:30 TP0] Decode batch. #running-req: 1, #token: 12943, token usage: 0.04, gen throughput (token/s): 105.52, #queue-req: 0
[2024-11-02 00:20:30] INFO:     127.0.0.1:41386 - "POST /v1/chat/completions HTTP/1.1" 200 OK
The first image shows a man in a yellow shirt ironing a shirt on the back of a yellow taxi cab, with a red line connecting the two objects. The second image shows a large orange "S" and "G" on a white background, with a red line connecting them.
In [17]:
terminate_process(embedding_process)

Chat Template

As mentioned before, if you do not specify a vision model's --chat-template, the server uses Hugging Face's default template, which only supports text.

We list popular vision models with their chat templates: