DeepSeek-V3.2
serving recipe.

Floors are the minimum the vLLM recipe publishes across the whole replica, from vLLM's own runs. The control plane compares them with its own sizing from the model config.

• Tensor ParallelFits inside one node's interconnect. single_node_tp
• Tensor + Expert ParallelFits inside one node's interconnect. single_node_tep
• Data + Expert ParallelFits inside one node's interconnect. single_node_dep
• Multi-Node Tensor ParallelSpans nodes. multi_node_tp
• Multi-Node TP + Pipeline ParallelSpans nodes. multi_node_tp_pp
• Multi-Node Data + Expert ParallelSpans nodes. multi_node_dep
• Multi-Node Tensor + Expert ParallelSpans nodes. multi_node_tep
• Prefill/Decode DisaggregationSpans nodes. pd_cluster

• H100 · 8 GPUs
• H200 · 8 GPUs

Wingsmith runs models through NVIDIA Dynamo, so it serves on NVIDIA GPUs. Other vendors are listed as the recipe states them.

vllm serve deepseek-ai/DeepSeek-V3.2 \
  --trust-remote-code \
  --kernel-config.enable_flashinfer_autotune=False \
  --tokenizer-mode deepseek_v32 \
  --tool-call-parser deepseek_v32 \
  --enable-auto-tool-choice \
  --reasoning-parser deepseek_v3

Features

• Tool callingOn by defaultEnable tool calling with DeepSeek V3.2 chat template support.
• ReasoningOn by defaultEnable reasoning/thinking mode with the DeepSeek V3 reasoning parser.
• Speculative decodingOpt-inMulti-Token Prediction speculative decoding with 3 speculative tokens.

Dependencies

• DeepGEMM pinned build for MQA logits (FP8 MoE kernels)
• Set VLLM_USE_DEEP_GEMM=0 to skip DeepGEMM for the MoE path (recommended on H20) (optional)

1. Register the weights

   In the console, or through the API. Resolving a slug proposes an artifact and writes nothing.

   curl "http://localhost:8000/v1/artifacts/resolve?source=hf:deepseek-ai/DeepSeek-V3.2"

2. Read the draft

   The draft carries this recipe as upstream evidence: the tested flags, the precision and memory floor, the parallelism layouts, and the hardware vLLM verified, with links back to the recipe. Nothing from it is applied without you, and the draft warns when the plane's own sizing and the published floor differ by a quarter or more.

3. Create the workload

   Choose the engine, device shape and scale. The plane picks the clusters, and each cluster's operator renders the deployment as a Dynamo graph.

4. Give it an endpoint

   One model name for clients, routed by the gateway to deployments on every cluster serving it.

Run the control plane locally →

• How much GPU memory does DeepSeek-V3.2 need?The vLLM recipe publishes a minimum of 403 GB across the whole replica for its NVFP4 variant; FP8 805 GB. That is vLLM's own measurement, not a per-GPU figure.
• Which parallelism layouts does the DeepSeek-V3.2 recipe support?Tensor Parallel, Tensor + Expert Parallel, Data + Expert Parallel, Multi-Node Tensor Parallel, Multi-Node TP + Pipeline Parallel, Multi-Node Data + Expert Parallel, Multi-Node Tensor + Expert Parallel and Prefill/Decode Disaggregation.
• Can Wingsmith serve DeepSeek-V3.2?Wingsmith serves models through NVIDIA Dynamo on clusters with NVIDIA GPUs. Register deepseek-ai/DeepSeek-V3.2 and the plane attaches this vLLM recipe to the draft as evidence: its tested flags, precision and memory floor, for you to review rather than applied on your behalf.

From the vLLM recipe for deepseek-ai/DeepSeek-V3.2 ↗ (snapshot at 4bcc189), Apache 2.0.