InternLM · CUHKSZzxy · Jun 22, 2026 · Jun 22, 2026 · Jun 22, 2026 · Jun 22, 2026
diff --git a/autotest/tools/pipeline/mllm_case.py b/autotest/tools/pipeline/mllm_case.py
@@ -9,8 +9,8 @@
 
 from lmdeploy import GenerationConfig, PytorchEngineConfig, TurbomindEngineConfig, pipeline
 from lmdeploy.messages import SpeculativeConfig
-from lmdeploy.vl import encode_image_base64, load_image, load_video
-from lmdeploy.vl.constants import IMAGE_TOKEN
+from lmdeploy.multimodal import encode_image_base64, load_image, load_video
+from lmdeploy.multimodal.constants import IMAGE_TOKEN
 
 gen_config = GenerationConfig(max_new_tokens=500, min_new_tokens=10)
 

diff --git a/docs/en/faq.md b/docs/en/faq.md
@@ -94,7 +94,7 @@ lmdeploy serve api_server internlm/internlm2_5-7b-chat --cache-max-entry-count 0
 ### Api Server Fetch Timeout
 
 The image URL fetch timeout for the API server can be configured via the environment variable `LMDEPLOY_FETCH_TIMEOUT`.
-By default, requests may take up to 10 seconds before timing out. See [lmdeploy/vl/utils.py](https://github.com/InternLM/lmdeploy/blob/7b6876eafcb842633e0efe8baabe5906d7beeeea/lmdeploy/vl/utils.py#L31) for usage.
+By default, requests may take up to 10 seconds before timing out. See [lmdeploy/multimodal/utils.py](https://github.com/InternLM/lmdeploy/blob/main/lmdeploy/multimodal/utils.py) for usage.
 
 ## Quantization
 

diff --git a/docs/en/get_started/ascend/get_started.md b/docs/en/get_started/ascend/get_started.md
@@ -50,7 +50,7 @@ Set `device_type="ascend"` in the `PytorchEngineConfig`:
 
 ```python
 from lmdeploy import pipeline, PytorchEngineConfig
-from lmdeploy.vl import load_image
+from lmdeploy.multimodal import load_image
 pipe = pipeline('OpenGVLab/InternVL2-2B',
         backend_config=PytorchEngineConfig(tp=1, device_type='ascend'))
 image = load_image('https://raw.githubusercontent.com/open-mmlab/mmdeploy/main/tests/data/tiger.jpeg')

diff --git a/docs/en/get_started/camb/get_started.md b/docs/en/get_started/camb/get_started.md
@@ -43,7 +43,7 @@ Set `device_type="camb"` in the `PytorchEngineConfig`:
 
 ```python
 from lmdeploy import pipeline, PytorchEngineConfig
-from lmdeploy.vl import load_image
+from lmdeploy.multimodal import load_image
 pipe = pipeline('OpenGVLab/InternVL2-2B',
         backend_config=PytorchEngineConfig(tp=1, device_type='camb'))
 image = load_image('https://raw.githubusercontent.com/open-mmlab/mmdeploy/main/tests/data/tiger.jpeg')

diff --git a/docs/en/get_started/get_started.md b/docs/en/get_started/get_started.md
@@ -83,7 +83,7 @@ For example, you can utilize the following code snippet to perform the inference
 
 ```python
 from lmdeploy import pipeline
-from lmdeploy.vl import load_image
+from lmdeploy.multimodal import load_image
 
 pipe = pipeline('OpenGVLab/InternVL2-8B')
 
@@ -96,7 +96,7 @@ In VLM pipeline, the default image processing batch size is 1. This can be adjus
 
 ```python
 from lmdeploy import pipeline, VisionConfig
-from lmdeploy.vl import load_image
+from lmdeploy.multimodal import load_image
 
 pipe = pipeline('OpenGVLab/InternVL2-8B',
                 vision_config=VisionConfig(

diff --git a/docs/en/get_started/maca/get_started.md b/docs/en/get_started/maca/get_started.md
@@ -33,7 +33,7 @@ Set `device_type="maca"` in the `PytorchEngineConfig`:
 
 ```python
 from lmdeploy import pipeline, PytorchEngineConfig
-from lmdeploy.vl import load_image
+from lmdeploy.multimodal import load_image
 pipe = pipeline('OpenGVLab/InternVL2-2B',
         backend_config=PytorchEngineConfig(tp=1, device_type='maca'))
 image = load_image('https://raw.githubusercontent.com/open-mmlab/mmdeploy/main/tests/data/tiger.jpeg')

diff --git a/docs/en/multi_modal/cogvlm.md b/docs/en/multi_modal/cogvlm.md
@@ -26,7 +26,7 @@ The following sample code shows the basic usage of VLM pipeline. For more exampl
 
 ```python
 from lmdeploy import pipeline
-from lmdeploy.vl import load_image
+from lmdeploy.multimodal import load_image
 
 
 if __name__ == "__main__":

diff --git a/docs/en/multi_modal/deepseek_vl2.md b/docs/en/multi_modal/deepseek_vl2.md
@@ -30,7 +30,7 @@ To construct valid DeepSeek-VL2 prompts with image inputs, users should insert `
 
 ```python
 from lmdeploy import pipeline
-from lmdeploy.vl import load_image
+from lmdeploy.multimodal import load_image
 
 
 if __name__ == "__main__":

diff --git a/docs/en/multi_modal/gemma3.md b/docs/en/multi_modal/gemma3.md
@@ -18,7 +18,7 @@ The following sample code shows the basic usage of VLM pipeline. For more exampl
 
 ```python
 from lmdeploy import pipeline
-from lmdeploy.vl import load_image
+from lmdeploy.multimodal import load_image
 
 
 if __name__ == "__main__":

diff --git a/docs/en/multi_modal/index.rst b/docs/en/multi_modal/index.rst
@@ -1,6 +1,9 @@
-Vision-Language Models
+Multimodal Models
 =================================
 
+Use ``lmdeploy.multimodal`` for multimodal helper APIs such as media loading
+and local-file encoding.
+
 .. toctree::
    :maxdepth: 2
    :caption: Examples

diff --git a/docs/en/multi_modal/internvl.md b/docs/en/multi_modal/internvl.md
@@ -42,7 +42,7 @@ The following sample code shows the basic usage of VLM pipeline. For detailed in
 
 ```python
 from lmdeploy import pipeline
-from lmdeploy.vl import load_image
+from lmdeploy.multimodal import load_image
 
 pipe = pipeline('OpenGVLab/InternVL2-8B')
 
@@ -60,7 +60,7 @@ More examples are listed below:
 
 ```python
 from lmdeploy import pipeline, GenerationConfig
-from lmdeploy.vl.constants import IMAGE_TOKEN
+from lmdeploy.multimodal.constants import IMAGE_TOKEN
 
 pipe = pipeline('OpenGVLab/InternVL2-8B', log_level='INFO')
 messages = [
@@ -86,7 +86,7 @@ out = pipe(messages, gen_config=GenerationConfig(top_k=1))
 
 ```python
 from lmdeploy import pipeline, GenerationConfig
-from lmdeploy.vl.constants import IMAGE_TOKEN
+from lmdeploy.multimodal.constants import IMAGE_TOKEN
 
 pipe = pipeline('OpenGVLab/InternVL2-8B', log_level='INFO')
 messages = [
@@ -114,8 +114,8 @@ out = pipe(messages, gen_config=GenerationConfig(top_k=1))
 import numpy as np
 from lmdeploy import pipeline, GenerationConfig
 from decord import VideoReader, cpu
-from lmdeploy.vl.constants import IMAGE_TOKEN
-from lmdeploy.vl import encode_image_base64
+from lmdeploy.multimodal.constants import IMAGE_TOKEN
+from lmdeploy.multimodal import encode_image_base64
 from PIL import Image
 pipe = pipeline('OpenGVLab/InternVL2-8B', log_level='INFO')
 

diff --git a/docs/en/multi_modal/llava.md b/docs/en/multi_modal/llava.md
@@ -33,7 +33,7 @@ The following sample code shows the basic usage of VLM pipeline. For detailed in
 
 ```python
 from lmdeploy import GenerationConfig, TurbomindEngineConfig, pipeline
-from lmdeploy.vl import load_image
+from lmdeploy.multimodal import load_image
 
 
 pipe = pipeline("llava-hf/llava-interleave-qwen-7b-hf", backend_config=TurbomindEngineConfig(cache_max_entry_count=0.5),

diff --git a/docs/en/multi_modal/minicpmv.md b/docs/en/multi_modal/minicpmv.md
@@ -19,7 +19,7 @@ The following sample code shows the basic usage of VLM pipeline. For detailed in
 
 ```python
 from lmdeploy import pipeline
-from lmdeploy.vl import load_image
+from lmdeploy.multimodal import load_image
 
 pipe = pipeline('openbmb/MiniCPM-V-2_6')
 
@@ -97,7 +97,7 @@ print(out.text)
 
 ```python
 from lmdeploy import pipeline, GenerationConfig
-from lmdeploy.vl import encode_image_base64
+from lmdeploy.multimodal import encode_image_base64
 import torch
 from PIL import Image
 from transformers import AutoModel, AutoTokenizer

diff --git a/docs/en/multi_modal/molmo.md b/docs/en/multi_modal/molmo.md
@@ -19,7 +19,7 @@ The following sample code shows the basic usage of VLM pipeline. For detailed in
 
 ```python
 from lmdeploy import pipeline
-from lmdeploy.vl import load_image
+from lmdeploy.multimodal import load_image
 
 pipe = pipeline('allenai/Molmo-7B-D-0924')
 

diff --git a/docs/en/multi_modal/multimodal_inputs.md b/docs/en/multi_modal/multimodal_inputs.md
@@ -398,7 +398,7 @@ In addition to HTTP URLs, lmdeploy accepts:
 - **Local file paths** via `file://` scheme: `file:///absolute/path/to/file.jpg`
 - **Base64-encoded data** via data URLs: `data:<mime>;base64,<encoded_data>`
 
-Use the helpers in `lmdeploy.vl.utils` to encode local files:
+Use the helpers in `lmdeploy.multimodal.utils` to encode local files:
 
 <details>
 <summary>Local file path example</summary>
@@ -434,7 +434,7 @@ print(response.choices[0].message.content)
 
 ```python
 from openai import OpenAI
-from lmdeploy.vl.utils import encode_image_base64
+from lmdeploy.multimodal.utils import encode_image_base64
 
 client = OpenAI(api_key='EMPTY', base_url='http://localhost:23333/v1')
 model_name = client.models.list().data[0].id
@@ -465,7 +465,7 @@ print(response.choices[0].message.content)
 
 ```python
 from openai import OpenAI
-from lmdeploy.vl.utils import encode_video_base64
+from lmdeploy.multimodal.utils import encode_video_base64
 
 client = OpenAI(api_key='EMPTY', base_url='http://localhost:23333/v1')
 model_name = client.models.list().data[0].id
@@ -497,7 +497,7 @@ print(response.choices[0].message.content)
 
 ```python
 from openai import OpenAI
-from lmdeploy.vl.utils import encode_audio_base64
+from lmdeploy.multimodal.utils import encode_audio_base64
 
 client = OpenAI(api_key='EMPTY', base_url='http://localhost:23333/v1')
 model_name = client.models.list().data[0].id
@@ -528,7 +528,7 @@ print(response.choices[0].message.content)
 
 ```python
 from openai import OpenAI
-from lmdeploy.vl.utils import encode_time_series_base64
+from lmdeploy.multimodal.utils import encode_time_series_base64
 
 client = OpenAI(api_key='EMPTY', base_url='http://localhost:23333/v1')
 model_name = client.models.list().data[0].id

diff --git a/docs/en/multi_modal/phi3.md b/docs/en/multi_modal/phi3.md
@@ -26,7 +26,7 @@ The following sample code shows the basic usage of VLM pipeline. For more exampl
 
 ```python
 from lmdeploy import pipeline
-from lmdeploy.vl import load_image
+from lmdeploy.multimodal import load_image
 
 pipe = pipeline('microsoft/Phi-3.5-vision-instruct')
 

diff --git a/docs/en/multi_modal/qwen2_5_vl.md b/docs/en/multi_modal/qwen2_5_vl.md
@@ -25,7 +25,7 @@ The following sample code shows the basic usage of the VLM pipeline. For detaile
 
 ```python
 from lmdeploy import pipeline
-from lmdeploy.vl import load_image
+from lmdeploy.multimodal import load_image
 
 pipe = pipeline('Qwen/Qwen2.5-VL-7B-Instruct')
 
@@ -98,8 +98,8 @@ out = pipe(messages, gen_config=GenerationConfig(top_k=1))
 import numpy as np
 from lmdeploy import pipeline, GenerationConfig
 from decord import VideoReader, cpu
-from lmdeploy.vl.constants import IMAGE_TOKEN
-from lmdeploy.vl import encode_image_base64
+from lmdeploy.multimodal.constants import IMAGE_TOKEN
+from lmdeploy.multimodal import encode_image_base64
 from PIL import Image
 pipe = pipeline('Qwen/Qwen2.5-VL-7B-Instruct', log_level='INFO')
 

diff --git a/docs/en/multi_modal/qwen2_vl.md b/docs/en/multi_modal/qwen2_vl.md
@@ -36,7 +36,7 @@ The following sample code shows the basic usage of VLM pipeline. For detailed in
 
 ```python
 from lmdeploy import pipeline
-from lmdeploy.vl import load_image
+from lmdeploy.multimodal import load_image
 
 pipe = pipeline('Qwen/Qwen2-VL-2B-Instruct')
 

diff --git a/docs/en/multi_modal/vl_pipeline.md b/docs/en/multi_modal/vl_pipeline.md
@@ -16,7 +16,7 @@ Using the pipeline interface to infer other VLM models is similar, with the main
 
 ```python
 from lmdeploy import pipeline
-from lmdeploy.vl import load_image
+from lmdeploy.multimodal import load_image
 
 pipe = pipeline('OpenGVLab/InternVL2_5-8B')
 
@@ -53,7 +53,7 @@ Tensor paramllelism can be activated by setting the engine parameter `tp`
 
 ```python
 from lmdeploy import pipeline, TurbomindEngineConfig
-from lmdeploy.vl import load_image
+from lmdeploy.multimodal import load_image
 
 pipe = pipeline('OpenGVLab/InternVL2_5-8B',
                 backend_config=TurbomindEngineConfig(tp=2))
@@ -69,7 +69,7 @@ When creating the pipeline, you can customize the size of the context window by
 
 ```python
 from lmdeploy import pipeline, TurbomindEngineConfig
-from lmdeploy.vl import load_image
+from lmdeploy.multimodal import load_image
 
 pipe = pipeline('OpenGVLab/InternVL2_5-8B',
                 backend_config=TurbomindEngineConfig(session_len=8192))
@@ -85,7 +85,7 @@ You can change the default sampling parameters of pipeline by passing `Generatio
 
 ```python
 from lmdeploy import pipeline, GenerationConfig, TurbomindEngineConfig
-from lmdeploy.vl import load_image
+from lmdeploy.multimodal import load_image
 
 pipe = pipeline('OpenGVLab/InternVL2_5-8B',
                 backend_config=TurbomindEngineConfig(tp=2, session_len=8192))
@@ -101,8 +101,8 @@ By default, LMDeploy inserts the special image token into the user prompt follow
 
 ```python
 from lmdeploy import pipeline
-from lmdeploy.vl import load_image
-from lmdeploy.vl.constants import IMAGE_TOKEN
+from lmdeploy.multimodal import load_image
+from lmdeploy.multimodal.constants import IMAGE_TOKEN
 
 pipe = pipeline('deepseek-ai/deepseek-vl-1.3b-chat')
 
@@ -117,7 +117,7 @@ While performing inference, LMDeploy identifies an appropriate chat template fro
 
 ```python
 from lmdeploy import pipeline, ChatTemplateConfig
-from lmdeploy.vl import load_image
+from lmdeploy.multimodal import load_image
 pipe = pipeline('local_model_folder',
                 chat_template_config=ChatTemplateConfig(model_name='llava-v1'))
 image = load_image('https://raw.githubusercontent.com/open-mmlab/mmdeploy/main/tests/data/tiger.jpeg')
@@ -133,7 +133,7 @@ The default parameters of the visual model can be modified by setting `VisionCon
 
 ```python
 from lmdeploy import pipeline, VisionConfig
-from lmdeploy.vl import load_image
+from lmdeploy.multimodal import load_image
 vision_config=VisionConfig(max_batch_size=16)
 pipe = pipeline('liuhaotian/llava-v1.5-7b', vision_config=vision_config)
 image = load_image('https://raw.githubusercontent.com/open-mmlab/mmdeploy/main/tests/data/tiger.jpeg')
@@ -145,7 +145,7 @@ print(response)
 
 ```python
 from lmdeploy import pipeline, GenerationConfig
-from lmdeploy.vl import load_image
+from lmdeploy.multimodal import load_image
 pipe = pipeline('OpenGVLab/InternVL2_5-8B')
 
 image = load_image('https://raw.githubusercontent.com/open-mmlab/mmdeploy/main/tests/data/tiger.jpeg')
@@ -162,7 +162,7 @@ When dealing with multiple images, you can put them all in one list. Keep in min
 
 ```python
 from lmdeploy import pipeline, TurbomindEngineConfig
-from lmdeploy.vl import load_image
+from lmdeploy.multimodal import load_image
 
 pipe = pipeline('OpenGVLab/InternVL2_5-8B',
                 backend_config=TurbomindEngineConfig(session_len=8192))
@@ -183,7 +183,7 @@ Conducting inference with batch prompts is quite straightforward; just place the
 
 ```python
 from lmdeploy import pipeline, TurbomindEngineConfig
-from lmdeploy.vl import load_image
+from lmdeploy.multimodal import load_image
 
 pipe = pipeline('OpenGVLab/InternVL2_5-8B',
                 backend_config=TurbomindEngineConfig(session_len=8192))
@@ -203,7 +203,7 @@ There are two ways to do the multi-turn conversations with the pipeline. One is
 
 ```python
 from lmdeploy import pipeline, TurbomindEngineConfig, GenerationConfig
-from lmdeploy.vl import load_image
+from lmdeploy.multimodal import load_image
 
 pipe = pipeline('OpenGVLab/InternVL2_5-8B',
                 backend_config=TurbomindEngineConfig(session_len=8192))
@@ -224,7 +224,7 @@ You can release the pipeline explicitly by calling its `close()` method, or alte
 from lmdeploy import pipeline
 
 from lmdeploy import pipeline
-from lmdeploy.vl import load_image
+from lmdeploy.multimodal import load_image
 
 with pipeline('OpenGVLab/InternVL2_5-8B') as pipe:
     image = load_image('https://raw.githubusercontent.com/open-mmlab/mmdeploy/main/tests/data/tiger.jpeg')

diff --git a/docs/zh_cn/faq.md b/docs/zh_cn/faq.md
@@ -94,7 +94,7 @@ lmdeploy serve api_server internlm/internlm2_5-7b-chat --cache-max-entry-count 0
 
 API 服务器的图像 URL 获取超时可通过环境变量 `LMDEPLOY_FETCH_TIMEOUT` 进行配置。默认情况下，请求可能需要长达 10 秒才会超时。
 
-请参阅 [lmdeploy/vl/utils.py](https://github.com/InternLM/lmdeploy/blob/7b6876eafcb842633e0efe8baabe5906d7beeeea/lmdeploy/vl/utils.py#L31) 了解用法。
+请参阅 [lmdeploy/multimodal/utils.py](https://github.com/InternLM/lmdeploy/blob/main/lmdeploy/multimodal/utils.py) 了解用法。
 
 ## 量化
 

diff --git a/docs/zh_cn/get_started/ascend/get_started.md b/docs/zh_cn/get_started/ascend/get_started.md
@@ -43,7 +43,7 @@ print(response)
 
 ```python
 from lmdeploy import pipeline, PytorchEngineConfig
-from lmdeploy.vl import load_image
+from lmdeploy.multimodal import load_image
 pipe = pipeline('OpenGVLab/InternVL2-2B',
      backend_config=PytorchEngineConfig(tp=1, device_type='ascend'))
 image = load_image('https://raw.githubusercontent.com/open-mmlab/mmdeploy/main/tests/data/tiger.jpeg')

diff --git a/docs/zh_cn/get_started/camb/get_started.md b/docs/zh_cn/get_started/camb/get_started.md
@@ -40,7 +40,7 @@ print(response)
 
 ```python
 from lmdeploy import pipeline, PytorchEngineConfig
-from lmdeploy.vl import load_image
+from lmdeploy.multimodal import load_image
 pipe = pipeline('OpenGVLab/InternVL2-2B',
      backend_config=PytorchEngineConfig(tp=1, device_type='camb'))
 image = load_image('https://raw.githubusercontent.com/open-mmlab/mmdeploy/main/tests/data/tiger.jpeg')