Additional ResourcesMultimodal Details

TensorRT-LLM Multimodal

View as Markdown

This document provides a comprehensive guide for multimodal inference using TensorRT-LLM backend in Dynamo.

You can provide multimodal inputs in the following ways:

  • By sending image URLs
  • By providing paths to pre-computed embedding files

You should provide either image URLs or embedding file paths in a single request.

Support Matrix

ModalityInput FormatAggregatedDisaggregatedNotes
ImageHTTP/HTTPS URLYesYesFull support for all image models
ImagePre-computed Embeddings (.pt, .pth, .bin)YesYesDirect embedding files
VideoHTTP/HTTPS URLNoNoNot implemented
AudioHTTP/HTTPS URLNoNoNot implemented

Supported URL Formats

FormatExampleDescription
HTTP/HTTPShttp://example.com/image.jpgRemote media files
Pre-computed Embeddings/path/to/embedding.ptLocal embedding files (.pt, .pth, .bin)

Deployment Patterns

TRT-LLM supports aggregated and traditional disaggregated patterns. See Architecture Patterns for detailed explanations.

PatternSupportedLaunch ScriptNotes
EPD (Simple Aggregated)✅agg.shEasiest setup
E/PD (Encode Separate)❌N/ANot supported
E/P/D (Full Disaggregation)🚧 WIPN/APR #4668 in progress
EP/D (Traditional Disaggregated)✅disagg_multimodal.shPrefill handles encoding

Component Flags

ComponentFlagPurpose
Worker--modality multimodalComplete pipeline (aggregated)
Prefill Worker--disaggregation-mode prefillImage processing + Prefill (multimodal tokenization happens here)
Decode Worker--disaggregation-mode decodeDecode only
Encode Worker (WIP)--disaggregation-mode encodeImage encoding (E/P/D flow)

Aggregated Serving

Quick steps to launch Llama-4 Maverick BF16 in aggregated mode:

$cd $DYNAMO_HOME
$
$export AGG_ENGINE_ARGS=./examples/backends/trtllm/engine_configs/llama4/multimodal/agg.yaml
$export SERVED_MODEL_NAME="meta-llama/Llama-4-Maverick-17B-128E-Instruct"
$export MODEL_PATH="meta-llama/Llama-4-Maverick-17B-128E-Instruct"
$./examples/backends/trtllm/launch/agg.sh

Client:

$curl localhost:8000/v1/chat/completions -H "Content-Type: application/json" -d '{
>    "model": "meta-llama/Llama-4-Maverick-17B-128E-Instruct",
>    "messages": [
>        {
>            "role": "user",
>            "content": [
>                {
>                    "type": "text",
>                    "text": "Describe the image"
>                },
>                {
>                    "type": "image_url",
>                    "image_url": {
>                        "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint.png"
>                    }
>                }
>            ]
>        }
>    ],
>    "stream": false,
>    "max_tokens": 160
>}'

Disaggregated Serving

Example using Qwen/Qwen2-VL-7B-Instruct:

$cd $DYNAMO_HOME
$
$export MODEL_PATH="Qwen/Qwen2-VL-7B-Instruct"
$export SERVED_MODEL_NAME="Qwen/Qwen2-VL-7B-Instruct"
$export PREFILL_ENGINE_ARGS="examples/backends/trtllm/engine_configs/qwen2-vl-7b-instruct/prefill.yaml"
$export DECODE_ENGINE_ARGS="examples/backends/trtllm/engine_configs/qwen2-vl-7b-instruct/decode.yaml"
$export MODALITY="multimodal"
$
$./examples/backends/trtllm/launch/disagg.sh
$curl 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 image"
>                },
>                {
>                    "type": "image_url",
>                    "image_url": {
>                        "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint.png"
>                    }
>                }
>            ]
>        }
>    ],
>    "stream": false,
>    "max_tokens": 160
>}'

For a large model like meta-llama/Llama-4-Maverick-17B-128E-Instruct, a multi-node setup is required for disaggregated serving (see Multi-node Deployment below), while aggregated serving can run on a single node. This is because the model with a disaggregated configuration is too large to fit on a single node’s GPUs. For instance, running this model in disaggregated mode requires 2 nodes with 8xH200 GPUs or 4 nodes with 4xGB200 GPUs.

Pre-computed Embeddings with E/P/D Flow

For high-performance multimodal inference, Dynamo supports pre-computed embeddings with an Encode-Prefill-Decode (E/P/D) flow using NIXL (RDMA) for zero-copy tensor transfer.

Supported File Types

  • .pt - PyTorch tensor files
  • .pth - PyTorch checkpoint files
  • .bin - Binary tensor files

Embedding File Formats

TRT-LLM supports two formats for embedding files:

1. Simple Tensor Format

Direct tensor saved as .pt file containing only the embedding tensor:

1embedding_tensor = torch.rand(1, 576, 4096)  # [batch, seq_len, hidden_dim]
2torch.save(embedding_tensor, "embedding.pt")

2. Dictionary Format with Auxiliary Data

Dictionary containing multiple keys, used by models like Llama-4 that require additional metadata:

1embedding_dict = {
2    "mm_embeddings": torch.rand(1, 576, 4096),
3    "special_tokens": [128256, 128257],
4    "image_token_offsets": [[0, 576]],
5    # ... other model-specific metadata
6}
7torch.save(embedding_dict, "llama4_embedding.pt")
  • Simple tensors: Loaded directly and passed to mm_embeddings parameter
  • Dictionary format: mm_embeddings key extracted as main tensor, other keys preserved as auxiliary data

How to Launch

$cd $DYNAMO_HOME/examples/backends/trtllm
$
$# Launch 3-worker E/P/D flow with NIXL
$./launch/epd_disagg.sh

This script is designed for 8-node H200 with Llama-4-Scout-17B-16E-Instruct model and assumes you have a model-specific embedding file ready.

Configuration

$# Encode endpoint for Prefill → Encode communication
$export ENCODE_ENDPOINT="dyn://dynamo.tensorrt_llm_encode.generate"
$
$# Security: Allowed directory for embedding files (default: /tmp)
$export ALLOWED_LOCAL_MEDIA_PATH="/tmp"
$
$# Security: Max file size to prevent DoS attacks (default: 50MB)
$export MAX_FILE_SIZE_MB=50

Example Request with Pre-computed Embeddings

$curl localhost:8000/v1/chat/completions -H "Content-Type: application/json" -d '{
>    "model": "meta-llama/Llama-4-Maverick-17B-128E-Instruct",
>    "messages": [
>        {
>            "role": "user",
>            "content": [
>                {"type": "text", "text": "Describe the image"},
>                {"type": "image_url", "image_url": {"url": "/path/to/embedding.pt"}}
>            ]
>        }
>    ],
>    "max_tokens": 160
>}'

E/P/D Architecture

The E/P/D flow implements a 3-worker architecture:

  • Encode Worker: Loads pre-computed embeddings, transfers via NIXL
  • Prefill Worker: Receives embeddings, handles context processing and KV-cache generation
  • Decode Worker: Performs streaming token generation

Multi-node Deployment (Slurm)

This section demonstrates how to deploy large multimodal models that require a multi-node setup using Slurm.

The scripts referenced in this section can be found in examples/basics/multinode/trtllm/.

Environment Setup

Assuming you have allocated your nodes via salloc and are inside an interactive shell:

$# Container image (build using docs/backends/trtllm/README.md#build-container)
$export IMAGE="<dynamo_trtllm_image>"
$
$# Host:container path pairs for mounting
$export MOUNTS="${PWD}/../../../../:/mnt"
$
$# Model configuration
$export MODEL_PATH="meta-llama/Llama-4-Maverick-17B-128E-Instruct"
$export SERVED_MODEL_NAME="meta-llama/Llama-4-Maverick-17B-128E-Instruct"
$export MODALITY=${MODALITY:-"multimodal"}

Multi-node Disaggregated Launch

For 4 4xGB200 nodes (2 for prefill, 2 for decode):

$# Customize parallelism to match your engine configs
$# export PREFILL_ENGINE_CONFIG="/mnt/examples/backends/trtllm/engine_configs/llama4/multimodal/prefill.yaml"
$# export DECODE_ENGINE_CONFIG="/mnt/examples/backends/trtllm/engine_configs/llama4/multimodal/decode.yaml"
$# export NUM_PREFILL_NODES=2
$# export NUM_DECODE_NODES=2
$# export NUM_GPUS_PER_NODE=4
$
$# Launches frontend + etcd/nats on head node, plus prefill and decode workers
$./srun_disaggregated.sh

Understanding the Output

  1. srun_disaggregated.sh launches three srun jobs: frontend, prefill worker, and decode worker
  2. The OpenAI frontend will dynamically discover workers as they register: 
    INFO dynamo_run::input::http: Watching for remote model at models
    INFO dynamo_llm::http::service::service_v2: Starting HTTP service on: 0.0.0.0:8000
  3. TRT-LLM workers output progress from each MPI rank while loading
  4. When ready, the frontend logs: 
    INFO dynamo_llm::discovery::watcher: added model model_name="meta-llama/Llama-4-Maverick-17B-128E-Instruct"

Cleanup

$pkill srun

NIXL Usage

Use CaseScriptNIXL Used?Data Transfer
EPD (Simple Aggregated)agg.shNoAll in one worker
EP/D (Traditional Disaggregated)disagg_multimodal.shOptionalPrefill → Decode (KV cache via UCX or NIXL)
E/P/D (pre-computed embeddings)epd_disagg.shYesEncoder → Prefill (embeddings via NIXL)
E/P/D (WIP)N/ANoEncoder → Prefill (handles via params), Prefill → Decode (KV cache)

NIXL for KV cache transfer is currently beta and only supported on AMD64 (x86_64) architecture.

ModelInput Types and Registration

TRT-LLM workers register with Dynamo using:

ModelInput TypePreprocessingUse Case
ModelInput.TokensRust frontend may tokenize, but multimodal flows re-tokenize and build inputs in the Python worker; Rust token_ids are ignoredAll TRT-LLM workers
1# TRT-LLM Worker - Register with Tokens
2await register_llm(
3    ModelInput.Tokens,      # Rust does minimal preprocessing
4    model_type,             # ModelType.Chat or ModelType.Prefill
5    generate_endpoint,
6    model_name,
7    ...
8)

Inter-Component Communication

Transfer StageMessageNIXL Transfer
Frontend → PrefillRequest with image URL or embedding pathNo
Encode → Prefill (pre-computed)NIXL metadataYes (Embeddings tensor)
Encode → Prefill (Image URL) (WIP)Disaggregated params with multimodal handlesNo
Prefill → DecodeDisaggregated paramsConfigurable (KV cache: NIXL default, UCX optional)

Known Limitations

  • No Data URL support - Only HTTP/HTTPS URLs supported; data:image/... base64 URLs not supported
  • No video support - No video encoder implementation
  • No audio support - No audio encoder implementation
  • Multimodal preprocessing/tokenization happens in Python - Rust may forward token_ids, but multimodal requests are parsed and re-tokenized in the Python worker
  • E/P/D mode is WIP - Full E/P/D with image URLs under development
  • Multi-node H100 limitation - Loading meta-llama/Llama-4-Maverick-17B-128E-Instruct with 8 nodes of H100 with TP=16 is not possible due to head count divisibility (num_attention_heads: 40 not divisible by tp_size: 16)

Supported Models

Multimodal models listed in TensorRT-LLM supported models are supported by Dynamo.

Common examples:

  • Llama 4 Vision models (Maverick, Scout)
  • Qwen2-VL models
  • Other vision-language models with TRT-LLM support

Key Files

FileDescription
components/src/dynamo/trtllm/main.pyWorker initialization and setup
components/src/dynamo/trtllm/utils/trtllm_utils.pyCommand-line argument parsing
components/src/dynamo/trtllm/multimodal_processor.pyMultimodal request processing
components/src/dynamo/trtllm/request_handlers/handlers.pyRequest handler factory
components/src/dynamo/trtllm/request_handlers/handler_base.pyBase handler and disaggregation modes