Vision Encoder (ViT) CUDA Graphs

The CUDA Graphs infrastructure in vLLM primarily targets the decoder (language model) forward pass. vLLM also supports capturing the encoder (vision transformer) forward pass as CUDA Graphs, independently from the decoder. This is based on https://github.com/vllm-project/vllm/pull/35963.

For two-tower vision encoders (e.g., DeepSeek-OCR's SAM + CLIP with dynamic tiling), a dual-path graph mode captures two independent sets of CUDA graphs — one for the global image path and one for the local patch path — enabling independent budget selection and partial eager fallback per path. This is based on https://github.com/vllm-project/vllm/pull/43586.

!!! note Encoder CUDA Graphs are orthogonal to decoder CUDA Graphs — both can be enabled simultaneously. Encoder graphs capture the vision encoder execution (e.g., ViT in Qwen3-VL), while decoder graphs capture the language model execution as described in the CUDA Graphs design document.

Motivation

Vision encoder inference incurs CUDA kernel launch overhead on the host side. The overhead is more significant when the batch size is small or image size is small.

Encoder CUDA Graphs eliminate this overhead by pre-capturing the full encoder forward pass at multiple token budget levels during model initialization, then replaying the appropriate graph at runtime.

For two-tower vision encoders such as DeepSeek-OCR (SAM + CLIP with dynamic tiling), the global image path and local patch path have independent token profiles (272 tokens per global image vs. 100 tokens per local patch). Capturing a single monolithic graph for both paths would significantly reduce packing efficiency. The dual-path graph mode captures each path as a separate set of budgets, allowing the manager to pack and replay each path independently.

Design

The encoder CUDA Graph system uses a budget-based capture/replay strategy, managed by [EncoderCudaGraphManager][vllm.v1.worker.encoder_cudagraph.EncoderCudaGraphManager]. The system contains the following core components:

• [EncoderCudaGraphManager][vllm.v1.worker.encoder_cudagraph.EncoderCudaGraphManager]: orchestrates capture, replay, greedy packing, and data-parallel execution for encoder CUDA Graphs.
• [SupportsEncoderCudaGraph][vllm.model_executor.models.interfaces.SupportsEncoderCudaGraph]: a runtime-checkable protocol that models implement to opt-in to encoder CUDA Graphs.
• [EncoderItemSpec][vllm.v1.worker.encoder_cudagraph_defs.EncoderItemSpec]: describes a single encoder input item (image or video) with its input size and output token count.
• [BudgetGraphMetadata][vllm.v1.worker.encoder_cudagraph.BudgetGraphMetadata]: holds the captured CUDA Graph and its associated I/O buffers for a single token budget level.

Budget-based graph capture

Multiple CUDA Graphs are pre-captured at different token budget levels (e.g., [2048, 4096, 8192, 13824]). Each budget defines a fixed token capacity, and all budgets share the same maximum batch size (number of images). The BudgetGraphMetadata for each level stores the graph along with pre-allocated input, metadata, and output buffers:

@dataclass
class BudgetGraphMetadata:
    token_budget: int
    max_batch_size: int
    max_frames_per_batch: int
    graph: torch.cuda.CUDAGraph
    input_buffers: dict[str, torch.Tensor]  # e.g. pixel_values, embeddings, seq metadata
    output_buffer: torch.Tensor      # encoder hidden states

Budgets are auto-generated as power-of-2 levels from a model-provided range via get_encoder_cudagraph_budget_range(), with the maximum budget always included even if it does not fall on a power-of-2 boundary. Budgets can also be explicitly specified by the user via encoder_cudagraph_token_budgets in CompilationConfig.

When EncoderCudaGraphConfig.enable_dual_path_graph is True, the manager generates two independent budget lists — global_token_budgets (multiples of global_token_per_image) and local_token_budgets (multiples of local_token_per_patch) — and stores captured graphs under budget_graphs["global"] and budget_graphs["local"] respectively.

Greedy bin-packing at runtime

When a batch of images arrives, the manager sorts images by output token count (smallest first) and greedily packs as many images as possible into each sub-batch while staying within the largest token budget and the maximum batch size. Once a sub-batch is finalized (the next image would overflow either constraint), the manager finds the smallest budget that fits the sub-batch's total tokens and replays the corresponding CUDA Graph. This repeats until the batch is exhausted. Images that exceed all budgets fall back to eager execution.

For dual-path models, the manager routes to _execute_local_dual_path(), which constrains both global and local token budgets simultaneously during packing (see Dual-Path graph capture).

For each graph replay:

1. Call prepare_encoder_cudagraph_replay_buffers() to compute buffer values (including pixel_values and precomputed metadata) from actual batch inputs.
2. Zero the pre-allocated input_buffers, then slice-copy the replay values into them.
3. Replay the CUDA Graph.
4. Clone outputs from output_buffer (cloning is necessary since the buffer is reused across replays).

Dual-Path graph capture

For two-tower vision encoders (e.g., DeepSeek-OCR), the EncoderCudaGraphConfig sets enable_dual_path_graph=True and provides global_token_per_image / local_token_per_patch. The manager captures two independent sets of CUDA graphs — one for the global image path and one for the local patch path — stored under budget_graphs["global"] and budget_graphs["local"] respectively.

Budget generation. Two separate budget lists are generated:

• global_token_budgets — power-of-2 multiples of global_token_per_image (e.g., [272, 544, 1088, 2176, 4352, 8704, 13824] for DeepSeek-OCR).
• local_token_budgets — power-of-2 multiples of local_token_per_patch (e.g., [0, 100, 200, 400, 800, 1600, 3200, 6400, 12800] for DeepSeek-OCR). A budget of 0 is always included to handle images with no local patches (images ≤ 640×640 that produce only global features).

Both lists are capped at the same max_budget.

Dual-path greedy packing. Each EncoderItemSpec provides both global_output_tokens (constant per image) and local_output_tokens (proportional to the patch count). The dual-path packing algorithm constrains both budgets simultaneously:

• Sort images by total output tokens (global + local), smallest first.
• Greedily pack images: an image is added to the current sub-batch only if both the accumulated global tokens ≤ max_global_budget and the accumulated local tokens ≤ max_local_budget, with the image count ≤ max_batch_size.
• Once either constraint would overflow, finalize the sub-batch and find the smallest fitting budget independently for each path.
• Repeat until all images are packed.

Partial graph fallback. After packing, each sub-batch falls into one of four execution scenarios:

Note that the 0-budget graph is never actually replayed for local — it signals that local patch processing should be skipped entirely.

Buffer keys per path. Global and local paths use different buffer keys. For DeepSeek-OCR, the global path uses pixel_values (full images, shape [B, 3, 1280, 1280]) while the local path uses images_crop (patches, shape [P, 3, 1024, 1024]). The manager iterates over each captured graph's own input_buffers.keys() rather than a shared buffer_keys list, so both paths can use different buffers.

Post-processing. The postprocess_encoder_output method receives a local_output parameter (a tensor or None) containing the local-path encoder output. The model is responsible for assembling global and local features into the final per-image embedding. For DeepSeek-OCR, this means reshaping the global output into [B, 272, n_embed], the local output into [P, 100, n_embed], assembling patch grids with newline tokens, and concatenating [patches_grid, global, view_separator] for each image.

!!! note The dual-path design enables partial CUDA graph coverage — one path can hit while the other falls back to eager. This avoids wasted compute on zero-padded patch buffers for untiled images and avoids graph invalidation caused by variable crop_shape per image.

Data-parallel support

When mm_encoder_tp_mode="data", the manager distributes images across TP ranks using load-balanced assignment via get_load_balance_assignment, executes locally on each rank, then gathers results back in the original order via tensor_model_parallel_all_gather.

Video inference support

Following https://github.com/vllm-project/vllm/pull/35963 (ViT full CUDA graph support for image inference), https://github.com/vllm-project/vllm/pull/38061 extends the encoder CUDA graph framework to support video inference for Qwen3-VL. Previously, the CUDA graph capture/replay path only handled image inputs (pixel_values + image_grid_thw). Video inputs use different keys (pixel_values_videos + video_grid_thw) and require larger cu_seqlens buffers because each video item contributes multiple frames (T attention sequences). This PR generalizes the protocol and manager to handle both modalities through a single shared graph manager.

!!! note Video CUDA graphs are automatically disabled when EVS (Efficient Video Sampling) pruning is enabled, since EVS makes the token count data-dependent and incompatible with CUDA graph capture.

Mixed inputs (image+video) per prompt are also supported now.

Model integration via SupportsEncoderCudaGraph

Models opt-in to encoder CUDA Graphs by implementing the [SupportsEncoderCudaGraph][vllm.model_executor.models.interfaces.SupportsEncoderCudaGraph] protocol. This protocol encapsulates all model-specific logic so that the manager remains model-agnostic. The protocol defines the following methods:

• get_encoder_cudagraph_config() — returns static configuration (supported modalities, buffer keys, output hidden size, padding logics, max frames per video).
• get_encoder_cudagraph_budget_range(vllm_config) — returns (min_budget, max_budget) for auto-inference of token budgets.
• get_encoder_cudagraph_item_specs(mm_kwargs) — returns list[EncoderItemSpec] describing each item with its input size, total output token count (output_tokens), and optionally per-path token counts (global_output_tokens, local_output_tokens) for dual-path models.
• select_encoder_cudagraph_items(mm_kwargs, indices) — extracts a sub-batch of items by index, used during greedy packing and DP sharding.
• prepare_encoder_cudagraph_capture_inputs(..., path="default") — creates dummy inputs for graph capture. The path parameter ("global" or "local") tells the model which path to generate dummy inputs for. Returns EncoderCudaGraphCaptureInputs with a single values: dict[str, torch.Tensor] that contains all buffers to be recorded into the graph.
• prepare_encoder_cudagraph_replay_buffers(mm_kwargs, max_batch_size, max_frames_per_batch, path="default") — computes buffer values from actual batch inputs. The path parameter selects which modality keys to extract from mm_kwargs. Returns EncoderCudaGraphReplayBuffers with a values dict whose keys match the captured graph's input_buffers.keys().
• encoder_cudagraph_forward(inputs: dict[str, torch.Tensor], path="default") — forward pass accepting only fixed-shaped input tensors (the captured values dict). Called during both capture and replay. The path parameter dispatches to the correct encoder sub-module (e.g., global vs. local path for DeepSeek-OCR).
• encoder_eager_forward(mm_kwargs, path="default") — fallback eager forward when no graph fits. When path is "global" or "local", runs only that encoder path without graph capture.
• postprocess_encoder_output(..., local_output=None) — post-process encoder output. The local_output parameter receives the local-path encoder output tensor (or None), enabling dual-path models to assemble global and local features into the final per-image embedding.

!!! note The SupportsEncoderCudaGraph protocol is designed to be model-agnostic. New vision encoder models can opt-in by implementing the protocol methods without modifying the manager.

Supported models:

!!! note Encoder CUDA Graphs have currently been tested with --mm-encoder-attn-backend=FLASH_ATTN and --mm-encoder-attn-backend=FLASHINFER on Blackwell GPUs. For Qwen2-VL and Qwen2.5-VL only FA2 and FA3 has been tested.

Configuration

Three fields in CompilationConfig control encoder CUDA Graphs:

• cudagraph_mm_encoder (bool, default False) — enable CUDA Graph capture for multimodal encoder. When enabled, captures the full encoder forward as a CUDA Graph for each token budget level.
• encoder_cudagraph_token_budgets (list[int], default []) — token budget levels for capture. If empty (default), auto-inferred from model architecture as power-of-2 levels. User-provided values override auto-inference.
• encoder_cudagraph_max_vision_items_per_batch (int, default 0) — maximum number of images/videos per batch during capture. If 0 (default), auto-inferred as max_budget // min_budget.
• encoder_cudagraph_max_frames_per_batch (int, default None) — maximum number of video frames per batch during capture. If None (default), auto-inferred as encoder_cudagraph_max_vision_items_per_batch * max_frames_per_video (max_frames_per_video is a model-specific value from EncoderCudaGraphConfig, computed by get_max_frames_per_video() on the model). If we limit the video count per prompt to 0, it will also be set to 0 (i.e., fall back to image-only mode).

Dual-path mode is configured at the model level via EncoderCudaGraphConfig fields (enable_dual_path_graph, global_token_per_image, local_token_per_patch) — no additional user configuration is required. The manager automatically generates separate budget lists and routes to dual-path execution when the model opts in.

Usage guide

Image inference

Enable encoder CUDA Graphs via compilation_config:

vllm serve Qwen/Qwen3-VL-32B \
  --compilation-config '{"cudagraph_mm_encoder": true}'

For Llama 4 (image only):

vllm serve meta-llama/Llama-4-Scout-17B-16E-Instruct \
  --limit-mm-per-prompt '{"image": 1}' \
  --compilation-config '{"cudagraph_mm_encoder": true}'

With explicit budgets:

vllm serve Qwen/Qwen3-VL-32B \
  --compilation-config '{"cudagraph_mm_encoder": true, "encoder_cudagraph_token_budgets": [2048, 4096, 8192, 13824], "encoder_cudagraph_max_vision_items_per_batch": 8}'

Python example:

import vllm

compilation_config = {
    "cudagraph_mm_encoder": True,
    # Optional: override auto-inferred budgets
    # "encoder_cudagraph_token_budgets": [2048, 4096, 8192, 13824],
    # "encoder_cudagraph_max_vision_items_per_batch": 8,
}

model = vllm.LLM(
    model="Qwen/Qwen3-VL-32B",
    compilation_config=compilation_config,
)

The manager tracks hit/miss statistics and logs them periodically. A "hit" means an image was processed via CUDA Graph replay; a "miss" means eager fallback (image exceeded all budgets).

Video inference

Enable encoder CUDA Graphs via compilation_config:

vllm serve Qwen/Qwen3-VL-32B \
  --compilation-config '{"cudagraph_mm_encoder": true}'

With explicit budgets:

vllm serve Qwen/Qwen3-VL-32B \
  --compilation-config '{"cudagraph_mm_encoder": true, "encoder_cudagraph_token_budgets": [2048, 4096, 8192, 13824], "encoder_cudagraph_max_vision_items_per_batch": 8, "encoder_cudagraph_max_frames_per_batch": 64}'

Python example:

import vllm

compilation_config = {
    "cudagraph_mm_encoder": True,
    # Optional: override auto-inferred budgets
    # "encoder_cudagraph_token_budgets": [2048, 4096, 8192, 13824],
    # "encoder_cudagraph_max_vision_items_per_batch": 8,
    # "encoder_cudagraph_max_frames_per_batch": 64,
}

model = vllm.LLM(
    model="Qwen/Qwen3-VL-32B",
    compilation_config=compilation_config,
)

About the Performance

The following benchmarks were run on Blackwell GPUs (GB200) using vllm bench mm-processor. See #35963 for full details.

Single GPU (1x GB200)

Model: Qwen/Qwen3-VL-30B-A3B-Instruct, dataset: lmarena-ai/VisionArena-Chat (3000 prompts, 300 warmup), max_model_len=32768.

To reproduce:

vllm bench mm-processor \
  --model Qwen/Qwen3-VL-30B-A3B-Instruct \
  --dataset-name hf --dataset-path lmarena-ai/VisionArena-Chat \
  --num-prompts 3000 --num-warmups 300 \
  --max-model-len 32768 --seed 42 \
  --mm-encoder-attn-backend FLASH_ATTN \
  --compilation-config '{"cudagraph_mm_encoder": true, "encoder_cudagraph_token_budgets": [512, 1024, 1536, 2048, 2560, 3072, 3584, 4096, 4864], "encoder_cudagraph_max_vision_items_per_batch": 8}'

Multi-GPU (4x GB200, TP=4, DP=4)

Model: Qwen/Qwen3-VL-32B-Instruct, dataset: random-mm (1000 prompts, 200 warmup, 20 images/request at 336x336), max_model_len=8192.

To reproduce:

vllm bench mm-processor \
  --model Qwen/Qwen3-VL-32B-Instruct \
  --dataset-name random-mm \
  --random-mm-base-items-per-request 20 \
  --random-mm-num-mm-items-range-ratio 0.0 \
  --random-mm-bucket-config '{"(336,336,1)": 1.0}' \
  --num-prompts 1000 --num-warmups 200 \
  --max-model-len 8192 --seed 42 \
  --mm-encoder-attn-backend FLASHINFER \
  --tensor-parallel-size 4 --mm-encoder-tp-mode data \
  --compilation-config '{"cudagraph_mm_encoder": true, "encoder_cudagraph_token_budgets": [512, 1024, 1536, 2048, 2560, 3072, 3584, 4096, 4864], "encoder_cudagraph_max_vision_items_per_batch": 8}'

!!! note Find more details about benchmarks on GPUs (A100) for video inference at #38061.