export BUCKET="your-gcs-bucket"
export REGION="us-central1" # Supported RTX PRO 6000 regions: us-central1, europe-west4, asia-southeast1, asia-south2

gcloud storage buckets create "gs://${BUCKET}" --location="${REGION}"
hf download nvidia/diffusiongemma-26B-A4B-it-NVFP4 --local-dir /tmp/dgemma
gcloud storage cp -r /tmp/dgemma/* "gs://${BUCKET}/dgemma/"

gcloud beta run deploy djev-dgemma \
  --region="us-central1" \
  --image=ghcr.io/taeold/djev-run:latest \
  --gpu=1 --gpu-type=nvidia-rtx-pro-6000 --no-gpu-zonal-redundancy \
  --cpu=20 --memory=80Gi --no-cpu-throttling \
  --concurrency=32 --min-instances=0 --max-instances=1 \
  --port=8080 \
  --network=default --subnet=default --vpc-egress=all-traffic \
  --add-volume=name=weights,type=cloud-storage,bucket="${BUCKET}",readonly=false,mount-options=enable-buffered-read=true \
  --add-volume-mount=volume=weights,mount-path=/mnt/gcs \
  --startup-probe=httpGet.path=/health,httpGet.port=8080,initialDelaySeconds=5,periodSeconds=2,timeoutSeconds=2,failureThreshold=120

gcloud beta run deploy djev-dgemma \
  --region="us-central1" \
  --image=docker.io/vllm/vllm-openai:nightly \
  --gpu=1 --gpu-type=nvidia-rtx-pro-6000 --no-gpu-zonal-redundancy \
  --cpu=20 --memory=80Gi --no-cpu-throttling \
  --concurrency=32 --min-instances=0 --max-instances=1 \
  --port=8000 \
  --network=default --subnet=default --vpc-egress=all-traffic \
  --add-volume=name=weights,type=cloud-storage,bucket="${BUCKET}",readonly=false,mount-options=enable-buffered-read=true \
  --add-volume-mount=volume=weights,mount-path=/mnt/gcs \
  --startup-probe=httpGet.path=/health,httpGet.port=8000,initialDelaySeconds=5,periodSeconds=2,timeoutSeconds=2,failureThreshold=120 \
  --set-env-vars="VLLM_FLASHINFER_MOE_BACKEND=masked_gemm,VLLM_ENABLE_V1_MULTIPROCESSING=0" \
  --command="/bin/bash" \
  --args="-c","cp -r /mnt/gcs/dgemma /dev/shm/dgemma && exec vllm serve /dev/shm/dgemma --served-model-name djev-dgemma --allowed-origins '[\"*\"]' --trust-remote-code --enforce-eager --language-model-only --attention-backend TRITON_ATTN --kv-cache-memory 2G --max-num-seqs 32 --max-model-len 4096 --diffusion-config '{\"canvas_length\":128}' --override-generation-config '{\"max_new_tokens\":null}'"

# Cloud Run & Storage flags:
# --image=docker.io/vllm/vllm-openai:nightly: serves standard vLLM diffusion natively without a custom Dockerfile
# --no-gpu-zonal-redundancy: required for standard regional RTX PRO 6000 quota
# --no-cpu-throttling: keeps all 20 vCPUs active during weight loading and vLLM scheduling
# --network=default --subnet=default --vpc-egress=all-traffic: streams weights from GCS over Google internal networking (~1.05 GiB/s)
# mount-options=enable-buffered-read=true & cp -r to /dev/shm: prefetches 18 GB safetensors shards sequentially from GCS into RAM before vLLM starts
#
# vLLM cold-start & runtime flags (--set-env-vars / --args):
# VLLM_ENABLE_V1_MULTIPROCESSING=0: runs EngineCore in-process so Python/CUDA modules are not imported twice
# VLLM_FLASHINFER_MOE_BACKEND=masked_gemm: selects the low-latency FlashInfer MoE kernel on Blackwell SM120
# --enforce-eager: skips torch.compile and CUDA graph capture on startup
# --language-model-only: skips loading and profiling the unused SigLIP vision encoder
# --kv-cache-memory=2G: pre-allocates a fixed 2 GiB KV cache, skipping the startup memory-profiling forward pass
# --attention-backend=TRITON_ATTN: uses Triton bidirectional attention required by DiffusionGemma
# --diffusion-config='{"canvas_length":128}': configures the 128-token parallel diffusion canvas

# 1. Tokenize query
curl -s https://<your-cloud-run-url>/tokenize \
  -H "Content-Type: application/json" \
  -d '{"prompt": "Classify: Payment issues\nlabel:"}'

# 2. Diffusion Read
curl -s https://<your-cloud-run-url>/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "djev-dgemma",
    "messages": [
      {"role": "user", "content": "Classify: Payment issues\nlabel:"}
    ],
    "max_tokens": 8,
    "vllm_xargs": {
      "diffusion_seed_canvas": [1, 2, 3, 4],
      "diffusion_pinned": [0, 1, 2],
      "diffusion_max_steps": 1,
      "diffusion_read_only": true
    }
  }'

curl -s https://<your-cloud-run-url>/v1/systemone \
  -H "Content-Type: application/json" \
  -d '{
    "state": "Customer cannot complete checkout; card declined twice.",
    "questions": {
      "category": {
        "type": "choice",
        "instructions": "Classify the support ticket",
        "criteria": ["billing", "technical", "account"]
      }
    }
  }'