# vllm 0.33.0 -> sglang 0.5.23  (input read as: kubernetes)

apiVersion: apps/v1
kind: Deployment
metadata:
  name: qwen3
spec:
  replicas: 1
  selector:
    matchLabels:
      app: qwen3
  template:
    metadata:
      labels:
        app: qwen3
    spec:
      containers:
      - name: vllm
        image: <sglang-image>
        command:
        - python3
        - -m
        - sglang.launch_server
        args:
        - --model-path
        - Qwen/Qwen3-32B
        - --context-length
        - '32768'
        - --tp-size
        - '2'
        - --mem-fraction-static
        - '0.90'
        - --disable-radix-cache
        - --lora-paths
        - sql=/adapters/sql
        - support=/adapters/support
        - --enable-lora
        - --port
        - '8000'
        env:
        - name: HF_TOKEN
          valueFrom:
            secretKeyRef:
              name: hf
              key: token
        ports:
        - containerPort: 8000
        resources:
          limits:
            nvidia.com/gpu: 2

## Report: 6 exact, 2 approximate, 0 no equivalent, 1 unknown

UNKNOWN        VLLM_LOGGING_LEVEL=INFO
               VLLM_LOGGING_LEVEL is a known vllm 0.33.0 option, but no mapping row covers it yet.
               source: vllm@v0.33.0:vllm/envs.py:861
EXACT          --model Qwen/Qwen3-32B  ->  --model-path Qwen/Qwen3-32B
               source: vllm@v0.33.0:vllm/engine/arg_utils.py:969
               target: sglang@v0.5.23:python/sglang/srt/arg_groups/fields/model.py:51
EXACT          --max-model-len 32768  ->  --context-length 32768
               source: vllm@v0.33.0:vllm/engine/arg_utils.py:991
               target: sglang@v0.5.23:python/sglang/srt/arg_groups/fields/model.py:110
EXACT          --tensor-parallel-size 2  ->  --tp-size 2
               source: vllm@v0.33.0:vllm/engine/arg_utils.py:1208
               target: sglang@v0.5.23:python/sglang/srt/arg_groups/fields/parallel.py:45
APPROXIMATE    --gpu-memory-utilization 0.90  ->  --mem-fraction-static 0.90
               vllm: Fraction (0-1) of GPU memory for the model executor of this instance. sglang:
               Fraction of memory used for static allocation (model weights and KV cache pool).
               source: vllm@v0.33.0:vllm/engine/arg_utils.py:1388
               target: sglang@v0.5.23:python/sglang/srt/arg_groups/fields/schedule.py:27
EXACT          --disable-prefix-caching  ->  --disable-radix-cache
               source: vllm@v0.33.0:vllm/engine/arg_utils.py:1401
               target: sglang@v0.5.23:python/sglang/srt/arg_groups/fields/memory.py:64
APPROXIMATE    --lora-modules sql=/adapters/sql support=/adapters/support  ->  --lora-paths sql=/adapters/sql support=/adapters/support
               vllm: Adapters as name=path items or JSON objects with "name" and "path". sglang:
               Adapters as PATH, NAME=PATH or JSON objects with "lora_name" and "lora_path".
               note: Adapter weight formats are not interchangeable between engines (for example
               llama.cpp needs GGUF adapters).
               source: vllm@v0.33.0:vllm/entrypoints/common/cli_args.py:81
               target: sglang@v0.5.23:python/sglang/srt/arg_groups/fields/lora.py:54
EXACT          --enable-lora  ->  --enable-lora
               source: vllm@v0.33.0:vllm/engine/arg_utils.py:1609
               target: sglang@v0.5.23:python/sglang/srt/arg_groups/fields/lora.py:34
EXACT          --port 8000  ->  --port 8000
               source: vllm@v0.33.0:vllm/entrypoints/common/cli_args.py:295
               target: sglang@v0.5.23:python/sglang/srt/arg_groups/fields/serving.py:66

## Also check

- The container image is not translated: replace <sglang-image>; probes, ports and resources are
  copied unchanged and may need the new port.
- Listen address: not set, and the defaults differ - vllm 0.33.0 --host defaults to None, sglang
  0.5.23 --host defaults to '127.0.0.1'.
- Prometheus metrics: sglang 0.5.23 has a switch for this (--enable-metrics); vllm has none, so
  decide whether to add it.
