Feature GuidesMultimodal

SGLang Multimodal

View as Markdown

This document provides a comprehensive guide for multimodal inference using SGLang backend in Dynamo. SGLang multimodal supports native EPD and EP/D flows where the SGLang engine performs media encoding, plus explicit encode-worker E/PD and E/P/D flows with NIXL (RDMA) for zero-copy tensor transfer.

Support Matrix

ModalityInput FormatAggregatedDisaggregatedNotes
ImageHTTP/HTTPS URLYesYesVision encoder generates embeddings
ImageData URL (Base64)NoNo
VideoHTTP/HTTPS/file:// URLYesYesVision encoder generates embeddings
AudioHTTP/HTTPS URLNoNoNot supported in SGLang backend

MM-aware KV routing is available for SGLang via the Rust frontend — it substitutes per-image pad_value tokens in the routing-side view so SGLang’s RadixAttention prefix-cache key matches the router’s overlap calculation. The frontend auto-detects the backend from the worker’s ModelDeploymentCard (the SGLang worker advertises backend_framework="sglang"), so no deployer-side flag is required. See Multimodal KV Routing → SGLang section. That path is orthogonal to the encode-worker / EPD topologies documented below; it’s a frontend routing concern that works with the aggregated SGLang worker layout in examples/backends/sglang/launch/agg_multimodal_router.sh.

Supported URL Formats

FormatExampleDescription
HTTP/HTTPShttp://example.com/image.jpgRemote media files
file://file:///tmp/test.mp4Local files accessible to the backend

Deployment Patterns

SGLang supports EPD, EP/D, E/PD, and E/P/D patterns. See Multimodal Model Serving for detailed explanations.

PatternSupportedLaunch ScriptNotes
EPD (Simple Aggregated)✅agg_vision.shInternal encoding
EP/D or P/D (No Separate Encode Worker)✅disagg.sh, disagg_same_gpu.shNative SGLang P/D launchers; prefill performs encode + prefill, decode reprocesses raw media metadata for token layout
E/PD (Encode Separate)✅multimodal_epd.shVision encoder separate
E/P/D (Full Disaggregation)✅multimodal_disagg.shKV cache via bootstrap

Component Flags

ComponentFlagPurpose
Native multimodal worker--enable-multimodalAllow raw multimodal inputs. In EP/D or P/D this keeps the normal prefill/decode workers; both receive raw media metadata.
Encode Worker--enable-multimodal --disaggregation-mode encodeFrontend-facing, vision encoding, embeddings generation (Rust frontend tokenizes)
Internal PD Worker--enable-multimodal --dedicated-mm-encoder --disaggregation-mode pdPrefill + decode worker that consumes embeddings from the encode worker
Internal Decode Worker--enable-multimodal --dedicated-mm-encoder --disaggregation-mode decodeEntry point for E/P/D after the encode worker has produced embeddings
Internal Prefill Worker--enable-multimodal --dedicated-mm-encoder --disaggregation-mode prefillCalled by internal decode, bootstrap coordination with precomputed embeddings

--dedicated-mm-encoder is intentionally explicit. Do not infer the internal E/PD or E/P/D worker path from --enable-multimodal --disaggregation-mode prefill/decode. Native EP/D or P/D uses those same two disaggregation modes, but it stays on the normal SGLang handlers: prefill processes raw image/video inputs to build vision context, while decode reprocesses the same raw media metadata so token layout matches the transferred KV cache. If the dedicated encoder flag is removed or made implicit, native disaggregated deployments can register only internal topology workers and lose the public OpenAI chat/completions surface.

In SGLang E/P/D, keep this flag on both the decode and prefill workers. This differs from vLLM: SGLang’s encode worker delegates generation to backend.generate, which is the decode worker, and that decode worker forwards the precomputed multimodal payload to prefill. With this flag, the internal workers consume transferred embeddings instead of raw image/video URLs, avoiding the duplicate raw-media preprocessing used by native EP/D or P/D.

SGLang-Specific Characteristics

  • Vision Encoder in Python: Encode worker uses SGLang’s MMEncoder for model-agnostic vision encoding
  • Token Expansion: Single <|image_pad|> token replaced with N tokens based on embedding shape
  • NIXL Transfer: Embeddings transferred from Encoder → PD Worker using NIXL
  • No Rust Processing: All tokenization and image handling happens in Python

Use the Latest Release

We recommend using the latest stable release of dynamo to avoid breaking changes:

GitHub Release

You can find the latest release and check out the corresponding branch with:

$git checkout $(git describe --tags $(git rev-list --tags --max-count=1))

EPD Serving (Simple Aggregated)

Components

Workflow

The DecodeWorkerHandler receives multimodal requests with image/video URLs and passes them directly to SGLang’s engine. SGLang’s internal mm_data_processor handles image/video fetching, loading, encoding, and token expansion.

Launch

$cd $DYNAMO_HOME/examples/backends/sglang
$./launch/agg_vision.sh --model-path Qwen/Qwen2-VL-7B-Instruct

Client:

$curl http://localhost:8000/v1/chat/completions \
>  -H "Content-Type: application/json" \
>  -d '{
>    "model": "Qwen/Qwen2.5-VL-7B-Instruct",
>    "messages": [
>      {
>        "role": "user",
>        "content": [
>          {
>            "type": "text",
>            "text": "Explain why Roger Federer is considered one of the greatest tennis players of all time"
>          },
>          {
>            "type": "image_url",
>            "image_url": {
>              "url": "http://images.cocodataset.org/test2017/000000155781.jpg"
>            }
>          }
>        ]
>      }
>    ],
>    "max_tokens": 50,
>    "stream": false
>  }' | jq

Video requests use the same aggregated path:

$curl http://localhost:8000/v1/chat/completions \
>  -H "Content-Type: application/json" \
>  -d '{
>    "model": "Qwen/Qwen2-VL-7B-Instruct",
>    "messages": [
>      {
>        "role": "user",
>        "content": [
>          {
>            "type": "text",
>            "text": "Describe the video in detail"
>          },
>          {
>            "type": "video_url",
>            "video_url": {
>              "url": "https://samplelib.com/mp4/sample-5s.mp4"
>            }
>          }
>        ]
>      }
>    ],
>    "max_tokens": 50,
>    "stream": false
>  }' | jq

EP/D or P/D Serving (No Separate Encode Worker)

Components

  • workers:
    • PrefillWorkerHandler receives raw multimodal metadata and lets SGLang perform media loading, encoding, token expansion, and KV production during prefill.
    • DecodeWorkerHandler receives matching multimodal metadata so token layout stays aligned with the transferred KV cache.

Workflow

The Rust frontend tokenizes the request and forwards image/video URLs as multi_modal_data. There is no encode worker. The prefill worker passes those URLs to SGLang’s normal multimodal engine path, so the vision context is produced inside the prefill worker. The decode worker also passes the same URLs to SGLang so the tokenizer manager can reproduce the multimodal token layout while consuming KV cache from prefill.

This native EP/D or P/D path intentionally trades simplicity for duplicated raw-media preprocessing: both prefill and decode call SGLang with image_data/video_data, so SGLang’s multimodal processor may fetch/load/preprocess the same media twice. Use the E/PD or E/P/D encode-worker topology when the deployment must preprocess media once and forward precomputed embeddings.

Launch

Native P/D and EP/D use the same launchers. The topology is selected by the model: text-only models run P/D, while VLMs run native EP/D where the prefill worker performs the media encode step. Neither path has a separate encode worker.

$cd $DYNAMO_HOME/examples/backends/sglang
$
$# P/D: text-only model, prefill on GPU 0 and decode on GPU 1.
$./launch/disagg.sh --model Qwen/Qwen3-0.6B
$
$# EP/D: VLM, same launcher; prefill performs media encoding and decode runs separately.
$./launch/disagg.sh --model Qwen/Qwen3-VL-4B-Instruct
$
$# Single-GPU smoke test: same EP/D behavior with prefill and decode packed together.
$./launch/disagg_same_gpu.sh --model Qwen/Qwen3-VL-4B-Instruct

These launchers pass --enable-multimodal to the prefill and decode workers but deliberately do not pass --dedicated-mm-encoder.

E/PD Serving (Encode Separate)

Components

Workflow

The Rust frontend tokenizes the request and extracts image URLs into multi_modal_data. The MultimodalEncodeWorker receives the pre-tokenized request, downloads and encodes the image, and passes the embeddings to the MultimodalWorker. The work complete event is sent via NATS, while the embeddings tensor is transferred via RDMA through the NIXL interface. The MultimodalWorker then prefills and decodes the prompt in the same engine, as in the LLM aggregated serving example. Only the encode worker is registered to the Dynamo frontend as an available endpoint. The PD worker does NOT register - it is an internal component and communicates via NATS.

Launch

$cd $DYNAMO_HOME/examples/backends/sglang
$./launch/multimodal_epd.sh

Client:

$curl http://localhost:8000/v1/chat/completions \
>  -H "Content-Type: application/json" \
>  -d '{
>    "model": "Qwen/Qwen2.5-VL-7B-Instruct",
>    "messages": [
>      {
>        "role": "user",
>        "content": [
>          {
>            "type": "text",
>            "text": "Explain why Roger Federer is considered one of the greatest tennis players of all time"
>          },
>          {
>            "type": "image_url",
>            "image_url": {
>              "url": "http://images.cocodataset.org/test2017/000000155781.jpg"
>            }
>          }
>        ]
>      }
>    ],
>    "max_tokens": 50,
>    "stream": false
>  }' | jq

E/P/D Serving (Full Disaggregation)

Components

Workflow

In models like Qwen2.5-VL, embeddings are only required during the prefill stage. The Rust frontend tokenizes and extracts image URLs. The MultimodalEncodeWorker receives the pre-tokenized request, encodes images, and transfers embeddings via NIXL to the Decode Worker (the entry point for disaggregation), which then coordinates with the Prefill Worker. The Prefill Worker processes the embeddings and forwards the KV cache back to the Decode Worker for token generation.

Launch

$cd $DYNAMO_HOME/examples/backends/sglang
$./launch/multimodal_disagg.sh

Client:

$curl http://localhost:8000/v1/chat/completions \
>  -H "Content-Type: application/json" \
>  -d '{
>    "model": "Qwen/Qwen2.5-VL-7B-Instruct",
>    "messages": [
>      {
>        "role": "user",
>        "content": [
>          {
>            "type": "text",
>            "text": "Explain why Roger Federer is considered one of the greatest tennis players of all time"
>          },
>          {
>            "type": "image_url",
>            "image_url": {
>              "url": "http://images.cocodataset.org/test2017/000000155781.jpg"
>            }
>          }
>        ]
>      }
>    ],
>    "max_tokens": 50,
>    "stream": false
>  }' | jq

Bootstrap Coordination

SGLang disaggregation uses a bootstrap mechanism for P->D coordination:

Request Flow (Important)

Client → Frontend → Processor → Encode → DECODE Worker → Prefill Worker
                                               ↑
                                    Entry point for disaggregation!

Bootstrap Process

  1. Decode Worker receives request from Encode Worker
  2. Decode Worker calls Prefill Worker via NATS to request bootstrap info
  3. Prefill Worker generates {host, port, room} and returns immediately
  4. Both workers connect to same “room” using bootstrap coordinates
  5. SGLang internally transfers KV cache state via bootstrap connection (not NIXL)

Key Difference from vLLM

  • vLLM: Frontend → Prefill → Decode (Prefill is entry point)
  • SGLang: Frontend → Processor → Encode → Decode → Prefill (Decode is entry point)

Inter-Component Communication

Control Flow (NATS)

All component-to-component communication happens via NATS:

E/PD Mode (Encode Separate)

Processor → Encode Worker → PD Worker
  (NATS)        (NATS + NIXL embeddings)

E/P/D Mode (Full Disaggregation)

Processor → Encode Worker → DECODE Worker → Prefill Worker
  (NATS)        (NATS)            (NATS)
                             ↓
                    Decode requests bootstrap
                             ↓
                    Prefill returns {host, port, room}
                             ↓
                    Both connect via bootstrap
                             ↓
                    SGLang internal KV cache transfer

Detailed Message Flow

Processor → Encode Worker:
  - NATS round_robin with SglangMultimodalRequest
  - Contains: tokenized input_ids, image URL, sampling params
Encode Worker → Decode/PD Worker:
  - NATS round_robin to "backend" component
  - Contains: expanded token_ids, NIXL metadata, embeddings shape
  - NIXL transfer: embeddings tensor
Decode Worker → Prefill Worker (disagg only):
  - NATS call to "prefill" component
  - Decode requests bootstrap coordinates
  - Prefill returns: {bootstrap_host, bootstrap_port, bootstrap_room}
Prefill ↔ Decode (via bootstrap):
  - SGLang internal connection (not NATS)
  - KV cache state shared via bootstrap mechanism

Data Transfer (NIXL)

NIXL is used only for embedding transfer:

1# Encode Worker
2descriptor = connect.Descriptor(precomputed_embeddings)
3with connector.create_readable(descriptor) as readable:
4    request.serialized_request = readable.metadata()
5    await pd_worker_client.round_robin(request)
6    await readable.wait_for_completion()
7
8# PD Worker
9embeddings = torch.empty(request.embeddings_shape, dtype=torch.float16)
10descriptor = connect.Descriptor(embeddings)
11read_op = await connector.begin_read(request.serialized_request, descriptor)
12await read_op.wait_for_completion()

Vision Encoding Details

Encode Worker Components

The encode worker uses SGLang’s MMEncoder for model-agnostic vision encoding. MMEncoder handles vision model loading, image preprocessing, and feature extraction internally:

1from sglang.srt.disaggregation.encode_server import MMEncoder
2
3self.encoder = MMEncoder(
4    server_args=config.server_args,
5    dist_init_method="tcp://127.0.0.1:0",
6    rank=0,
7)
8
9# At request time:
10image_grid_dim, mm_embedding = await self.encoder._encode([image_url])

Token Expansion Process

  1. Processor inserts single image token (e.g., <|image_pad|>)
  2. Encode worker generates embeddings: shape = (batch, num_patches, hidden_dim)
  3. Encode worker replaces single token with num_patches tokens
  4. Downstream worker receives expanded token sequence

Example:

1# Before: ["Hello", "<|image_pad|>", "world"]
2# After:  ["Hello", "<|image_pad|>", "<|image_pad|>", ...(576 tokens), "world"]

Chat Template Processing

SGLang uses its own chat template system:

1from sglang.srt.parser.conversation import chat_templates
2
3conv = chat_templates["qwen2-vl"].copy()
4conv.append_message(conv.roles[0], f"{conv.image_token} Describe this image")
5processed = tokenizer(text=conv.get_prompt(), return_tensors="pt")

Supported templates: qwen2-vl, llama-3, vicuna, etc.

NIXL Usage

Use CaseNIXL Used?Data TransferNotes
EPD (Simple Aggregated)NoN/AAll processing internal to SGLang
EP/D or P/D (No Separate Encode Worker)NoPrefill → Decode (KV cache via bootstrap)Prefill performs SGLang media encoding inline
E/PD (Encode Separate)YesEncoder → PD (embeddings)Vision encoder separate
E/P/D (Full Disaggregation)YesEncoder → Prefill (embeddings)KV cache via SGLang bootstrap

Key Difference: SGLang native EP/D or P/D uses bootstrap mechanism, not NIXL for KV cache like vLLM.

Environment Variables

SGLANG_ENCODER_MM_LOAD_WORKERS

Controls how many threads the encoder uses to fetch and load images concurrently. When a request contains multiple images (URLs, file paths, or base64 data), each image is loaded in a separate thread. Default is 4. Increase if image loading (network fetch or disk I/O) is the bottleneck rather than GPU compute. Has no effect if the vision encoder itself is the bottleneck, since encoding is sequential on GPU after all images are loaded.

$# Example: allow up to 16 concurrent image loads per encoder
$export SGLANG_ENCODER_MM_LOAD_WORKERS=16

Only applies to the EPD encode worker (which uses SGLang’s MMEncoder internally).

Profiling

Dynamo’s SGLang multimodal workers include NVTX markers for nsys profiling. They are disabled by default (zero overhead) and enabled by setting DYN_NVTX=1.

$cd $DYNAMO_HOME/examples/backends/sglang
$DYN_NVTX=1 nsys profile --trace=cuda,nvtx -o profile.nsys-rep \
>  bash launch/multimodal_epd.sh ...
ENV VariableDefaultDescription
DYN_NVTX0Set to 1 to enable NVTX range/mark annotations in multimodal encode/prefill/decode worker paths for nsys profiling

Key NVTX ranges emitted:

RangeWorkerDescription
mm:enc:generateEncodeFull encode request lifetime
mm:enc:vision_encodeEncodeVision encode call (MMEncoder._encode)
mm:enc:embedding_transferEncodeEmbedding handoff to downstream worker
mm:nixl:begin_readPD (agg) / PrefillBegin NIXL read operation for embeddings
mm:nixl:wait_completionPD (agg) / PrefillWait for NIXL embedding transfer completion
mm:pd:generateAggregated worker / Decode worker (MultimodalWorkerHandler)Full worker-side request lifetime
mm:pd:generate_aggPD (agg)Aggregated generation path
mm:pd:load_multimodalPD (agg)Build multimodal items from transferred embeddings
mm:pd:generate_disaggDecode worker (disagg entrypoint)Disaggregated generation path
mm:prefill:bootstrapPrefill (disagg)Bootstrap coordination path before returning {bootstrap_host, bootstrap_port, bootstrap_room}
mm:prefill:load_multimodalPrefill (disagg)Build multimodal items from transferred embeddings in the prefill worker
mm:prefill:engine_async_generatePrefill (disagg)SGLang prefill engine invocation (engine.async_generate)
mm:pd:ttftAggregated worker / Decode worker (MultimodalWorkerHandler)Worker-entry TTFT: from request arrival at this worker to first output token (excludes client->frontend->worker network transit)
mm:dec:first_tokenAggregated worker / Decode worker (MultimodalWorkerHandler)Decode-stage first-token range (starts when decode stream is launched; not worker-entry TTFT)

Known Limitations

  • No Data URL support - Only HTTP/HTTPS URLs supported; data:image/... base64 URLs not supported
  • No pre-computed embeddings - Cannot use .pt, .pth, .bin embedding files; vision encoder runs for every request
  • No audio support - No audio encoder implementation
  • Only Processor registers with Dynamo - Workers are internal components, frontend routes to Processor only
  • Disaggregated routing - Decode Worker is the entry point (calls Prefill), cannot route directly to Prefill workers
  • Limited model generalization - Token expansion logic is model-specific; adding new models may require implementation updates

Supported Models

SGLang multimodal only supports image-based vision-language models:

  • Qwen2-VL / Qwen2.5-VL - Qwen/Qwen2.5-VL-7B-Instruct
  • Qwen3-VL - Qwen/Qwen3-VL-30B-A3B-Instruct
  • Models supported by SGLang’s MMEncoder

Key Files

FileDescription
components/src/dynamo/sglang/main.pyComponent initialization, Encode Worker registers
components/src/dynamo/sglang/request_handlers/multimodal/encode_worker_handler.pyFrontend-facing: vision encoding, embeddings generation (receives pre-tokenized input)
components/src/dynamo/sglang/request_handlers/multimodal/worker_handler.pyPD/Prefill/Decode workers, NIXL read
components/src/dynamo/sglang/protocol.pyRequest/response data structures
components/src/dynamo/sglang/register.pyRegistration logic (called for Encode Worker)