apiVersion: serving.kserve.io/v1beta1
kind: InferenceService
metadata:
  annotations:
    serving.kserve.io/deploymentMode: RawDeployment
  labels:
    app.kubernetes.io/name: llm-reasoning
    app.kubernetes.io/part-of: llm-serving
  name: reasoning
  namespace: llm-serving
spec:
  predictor:
    containers:
    - args:
      - --model=unsloth/DeepSeek-R1-Distill-Qwen-32B-bnb-4bit
      - --served-model-name=reasoning
      - --quantization=bitsandbytes
      - --dtype=float16
      - --kv-cache-dtype=auto
      - --tensor-parallel-size=1
      - --max-model-len=16384
      - --gpu-memory-utilization=0.90
      - --max-num-seqs=4
      - --enable-chunked-prefill
      - --enable-prefix-caching
      - --reasoning-parser=deepseek_r1
      # Without these, any client sending tool_choice="auto" (pi does, for
      # Read/Bash/etc.) gets a 400: "auto" tool choice requires
      # --enable-auto-tool-choice and --tool-call-parser to be set.
      # deepseek_v3 (matching --reasoning-parser above) 400s here --
      # "DeepSeek-V3 Tool parser could not locate tool call start/end
      # tokens in the tokenizer" -- this checkpoint is a Qwen2.5-32B base
      # distilled on R1 traces, so its tokenizer never got DeepSeek-V3's
      # own special tool-call tokens registered even though it kept R1's
      # <think> reasoning format. hermes parses tool calls from plain text
      # patterns instead of special tokens, so it works against the
      # underlying Qwen tokenizer regardless. Verified live: deepseek_v3
      # 400s, hermes returns a real tool_calls response.
      - --enable-auto-tool-choice
      - --tool-call-parser=hermes
      # vLLM 0.11.0's native OffloadingConnector -- spills KV cache blocks
      # to CPU DRAM instead of discarding them on preemption (max-num-seqs=4
      # + max-model-len=16384 means concurrent long sequences compete for
      # the same GPU KV space). No extra dependency, built into vLLM core.
      # num_cpu_blocks is a starting point sized against the +4Gi/replica
      # headroom added below (worker-1 has ~18Gi of request headroom across
      # both replicas as of 2026-08-19) -- watch actual host memory usage
      # and adjust; block_size=128 tokens matches vLLM's own example.
      # Note: 0.11.0 ships the original (fragmented, small-transfer-block)
      # version of this connector -- 0.12.0 consolidates KV data into one
      # contiguous block per request and is reported an order of magnitude
      # faster for this specific feature, so this is a real but not yet
      # optimal implementation until the image gets bumped.
      - --kv-transfer-config={"kv_connector":"OffloadingConnector","kv_role":"kv_both","kv_connector_extra_config":{"num_cpu_blocks":2000,"block_size":128}}
      - --host=0.0.0.0
      - --port=8080
      env:
      - name: VLLM_USE_FLASHINFER_SAMPLER
        value: '0'
      - name: VLLM_ATTENTION_BACKEND
        value: TRITON_ATTN
      - name: HF_HOME
        value: /mnt/models
      image: vllm/vllm-openai:v0.11.0@sha256:014a95f21c9edf6abe0aea6b07353f96baa4ec291c427bb1176dc7c93a85845c
      name: kserve-container
      ports:
      - containerPort: 8080
        protocol: TCP
      readinessProbe:
        httpGet:
          path: /health
          port: 8080
        periodSeconds: 10
      resources:
        limits:
          cpu: '16'
          memory: 24Gi
          nvidia.com/gpu: '1'
        requests:
          cpu: '8'
          memory: 12Gi
          nvidia.com/gpu: '1'
      startupProbe:
        failureThreshold: 80
        httpGet:
          path: /health
          port: 8080
        periodSeconds: 15
      volumeMounts:
      - mountPath: /mnt/models
        name: models
      - mountPath: /dev/shm
        name: shm
    deploymentStrategy:
      type: Recreate
    maxReplicas: 2
    minReplicas: 2
    nodeSelector:
      kubernetes.io/hostname: worker-1
    runtimeClassName: nvidia
    volumes:
    - name: models
      persistentVolumeClaim:
        claimName: llm-models
    - emptyDir:
        medium: Memory
        sizeLimit: 2Gi
      name: shm