DeepSeek-V4-Flash
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 Parallel (default)Fits inside one node's interconnect. single_node_tep
• Data + Expert ParallelFits inside one node's interconnect. single_node_dep
• Multi-Node Data + Expert ParallelSpans nodes. multi_node_dep
• Prefill/Decode DisaggregationSpans nodes. pd_cluster
• Multi-Node Tensor ParallelSpans nodes. multi_node_tp

• B200 · 8 GPUs
• B300 · 8 GPUs
• DGX Spark (GB10) · 1 GPU
• DGX Station (GB300) · 1 GPU
• GB200 NVL4 · 4 GPUs
• GB300 NVL4 · 4 GPUs
• H200 · 8 GPUs
• RTX Pro 6000 8x · 8 GPUs
• MI300X · 8 GPUs
• MI325X · 8 GPUs
• MI355X · 8 GPUs
• Ascend 950DT 8x · 8 GPUs
• Ascend 950PR 8x · 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-V4-Flash-0731 \
  --trust-remote-code \
  --kv-cache-dtype fp8 \
  --block-size 256 \
  --tokenizer-mode deepseek_v4 \
  --tool-call-parser deepseek_v4 \
  --enable-auto-tool-choice \
  --reasoning-parser deepseek_v4 \
  --reasoning-config '{"reasoning_parser":"deepseek_v4","reasoning_start_str":"","reasoning_end_str":""}'

Features

• Tool callingOn by defaultEnable tool calling with DeepSeek V4 chat template support.
• ReasoningOn by defaultEnable reasoning/thinking mode with the DeepSeek V4 reasoning parser.
• Speculative decodingOpt-inSpeculative decoding — pick a drafting method. Modes: MTP, DSpark.

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-V4-Flash"

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. The one recipe default it applies, --max-num-seqs, is shown above.

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-V4-Flash need?The vLLM recipe publishes a minimum of 170 GB across the whole replica for its FP8 variant; FP8 (0731) 200 GB, NVFP4 170 GB, DSpark 200 GB. That is vLLM's own measurement, not a per-GPU figure.
• Which parallelism layouts does the DeepSeek-V4-Flash recipe support?Tensor Parallel, Tensor + Expert Parallel, Data + Expert Parallel, Multi-Node Data + Expert Parallel, Prefill/Decode Disaggregation and Multi-Node Tensor Parallel. The recipe's default is Tensor + Expert Parallel.
• Can Wingsmith serve DeepSeek-V4-Flash?Wingsmith serves models through NVIDIA Dynamo on clusters with NVIDIA GPUs. Register deepseek-ai/DeepSeek-V4-Flash 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.
• Does NVIDIA publish a Dynamo recipe for DeepSeek-V4-Flash?Yes: 10 deployment entries for vLLM and SGLang, aggregated and disaggregated. When a workload is placed, the plane defaults --max-num-seqs to the value the aggregated recipe publishes for that card, and names it on the draft.

From the vLLM recipe for deepseek-ai/DeepSeek-V4-Flash ↗ (snapshot at 4bcc189) and NVIDIA Dynamo's recipes at 165d854, Apache 2.0.