Gemma 4 26B-A4B IT
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 Tensor + Expert ParallelSpans nodes. multi_node_tep
• Multi-Node Data + Expert ParallelSpans nodes. multi_node_dep

• DGX Spark (GB10) · 1 GPU
• DGX Station (GB300) · 1 GPU
• H100 · 8 GPUs
• H200 · 8 GPUs
• RTX 5090 · 1 GPU
• RTX Pro 6000 · 1 GPU
• MI300X · 8 GPUs
• MI325X · 8 GPUs
• MI355X · 8 GPUs
• TPU v6e (Trillium) · 8 GPUs
• TPU v7 (Ironwood) · 1 GPU
• Xeon 6 · 4 GPUs

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

vllm serve Google/gemma-4-26B-A4B-it \
  --enable-auto-tool-choice \
  --tool-call-parser gemma4 \
  --chat-template examples/tool_chat_template_gemma4.jinja \
  --reasoning-parser gemma4

Features

• Tool callingOn by defaultEnable automatic tool choice with Gemma 4 parser and chat template
• ReasoningOn by defaultEnable structured thinking/reasoning output
• Text-only modeOpt-inSkip loading the vision encoder for text-only workloads — frees VRAM for KV cache. Mutually exclusive with encoder_parallel.
• Speculative decodingOpt-inMTP speculative decoding for accelerated inference Modes: MTP, Eagle3.

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:Google/gemma-4-26B-A4B-it"

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 Gemma 4 26B-A4B IT need?The vLLM recipe publishes a minimum of 16 GB across the whole replica for its NVFP4 variant; BF16 64 GB, FP8 32 GB, NVFP4 16 GB. That is vLLM's own measurement, not a per-GPU figure.
• Which parallelism layouts does the Gemma 4 26B-A4B IT recipe support?Tensor Parallel, Tensor + Expert Parallel, Data + Expert Parallel, Multi-Node Tensor Parallel, Multi-Node Tensor + Expert Parallel and Multi-Node Data + Expert Parallel.
• Can Wingsmith serve Gemma 4 26B-A4B IT?Wingsmith serves models through NVIDIA Dynamo on clusters with NVIDIA GPUs. Register Google/gemma-4-26B-A4B-it 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 Google/gemma-4-26B-A4B-it ↗ (snapshot at 4bcc189), Apache 2.0.