docs: add vLLM-Omni serving guide

Document the compatible KRAFTON vLLM-Omni branch and validated revision, single-GPU server configuration, and TTS/STT HTTP examples in both English and Korean.

README.md

@@ -266,6 +266,103 @@ messages = [
 response = pipe.chat(messages)
 ```
 
+## Deployment with vLLM-Omni
+
+Use KRAFTON's official
+[vLLM-Omni repository](https://github.com/krafton-ai/vllm-omni) and the
+compatible `feat/axk2_raon_speech` branch. The validated revision is
+[`7bcaf21`](https://github.com/krafton-ai/vllm-omni/commit/7bcaf21df881bda4280cbbf9752fc356e753d3bb);
+you can optionally pin it to match the vLLM-Omni implementation revision used
+for validation.
+
+### Install and serve
+
+```bash
+git clone --branch feat/axk2_raon_speech --single-branch \
+    https://github.com/krafton-ai/vllm-omni.git
+cd vllm-omni
+
+# Optional: pin the revision validated with this checkpoint.
+git checkout 7bcaf21df881bda4280cbbf9752fc356e753d3bb
+
+python -m pip install "vllm==0.20.0"
+python -m pip install -e .
+```
+
+The bundled stage profile targets one GPU with approximately 96 GB of VRAM.
+It uses tensor parallelism 1 (`TP=1`) and sets both the maximum model length
+and maximum batched tokens to 4096. With vLLM 0.20,
+`VLLM_MLA_DISABLE=1` is required because this checkpoint has a 192-wide
+KV-cache head.
+
+From the cloned repository root, start the server with:
+
+```bash
+export VLLM_MLA_DISABLE=1
+
+vllm-omni serve KRAFTON/A.X-K2-Raon-Speech-21B-A3B \
+    --stage-configs-path examples/offline_inference/raon/raon_axk1_local.yaml \
+    --host 0.0.0.0 \
+    --port 8091 \
+    --trust-remote-code \
+    --omni
+```
+
+### Health and model listing
+
+```bash
+curl -fS http://localhost:8091/health
+curl -fS http://localhost:8091/v1/models
+```
+
+### TTS
+
+```bash
+curl -fS -X POST http://localhost:8091/v1/audio/speech \
+    -H "Content-Type: application/json" \
+    --data-binary @- \
+    --output axk2_tts.wav <<'EOF'
+{
+  "model": "KRAFTON/A.X-K2-Raon-Speech-21B-A3B",
+  "input": "Hello, this is A.X-K2 Raon Speech.",
+  "response_format": "wav"
+}
+EOF
+```
+
+### STT
+
+Encode the input audio as a portable single-line base64 data URL:
+
+```bash
+AUDIO_B64="$(base64 < audio.wav | tr -d '\n')"
+
+curl -fS -X POST http://localhost:8091/v1/chat/completions \
+    -H "Content-Type: application/json" \
+    --data-binary @- <<EOF
+{
+  "model": "KRAFTON/A.X-K2-Raon-Speech-21B-A3B",
+  "messages": [
+    {
+      "role": "user",
+      "content": [
+        {
+          "type": "audio_url",
+          "audio_url": {
+            "url": "data:audio/wav;base64,$AUDIO_B64"
+          }
+        },
+        {
+          "type": "text",
+          "text": "Transcribe the audio into text."
+        }
+      ]
+    }
+  ]
+}
+EOF
+```
+
 ## Training
 
 The released checkpoint is the final model following:
@@ -310,4 +407,4 @@ Republic of Korea (Grant No. 2026-AIData-WII01).
 This model is released under the
 [CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/) license. The
 bundled A.X K2 Light implementation code retains its Apache License 2.0
-notices, which apply to that code only.
+notices, which apply to that code only.

README_ko.md

@@ -267,6 +267,101 @@ messages = [
 response = pipe.chat(messages)
 ```
 
+## vLLM-Omni 배포
+
+KRAFTON의 공식
+[vLLM-Omni 저장소](https://github.com/krafton-ai/vllm-omni)와 호환 브랜치인
+`feat/axk2_raon_speech`를 사용합니다. 검증된 리비전은
+[`7bcaf21`](https://github.com/krafton-ai/vllm-omni/commit/7bcaf21df881bda4280cbbf9752fc356e753d3bb)이며,
+검증에 사용된 vLLM-Omni 구현 리비전과 일치시키려면 이 커밋으로 고정하세요.
+
+### 설치 및 실행
+
+```bash
+git clone --branch feat/axk2_raon_speech --single-branch \
+    https://github.com/krafton-ai/vllm-omni.git
+cd vllm-omni
+
+# Optional: pin the revision validated with this checkpoint.
+git checkout 7bcaf21df881bda4280cbbf9752fc356e753d3bb
+
+python -m pip install "vllm==0.20.0"
+python -m pip install -e .
+```
+
+함께 제공되는 단계 프로필은 약 96 GB VRAM을 갖춘 GPU 한 장을 대상으로
+합니다. 텐서 병렬화는 1(`TP=1`)이며, 최대 모델 길이와 최대 배치 토큰 수는
+모두 4096입니다. vLLM 0.20에서는 이 체크포인트의 KV 캐시 헤드 너비가
+192이므로 `VLLM_MLA_DISABLE=1` 설정이 필요합니다.
+
+복제한 저장소의 최상위 디렉터리에서 다음 명령으로 서버를 실행합니다.
+
+```bash
+export VLLM_MLA_DISABLE=1
+
+vllm-omni serve KRAFTON/A.X-K2-Raon-Speech-21B-A3B \
+    --stage-configs-path examples/offline_inference/raon/raon_axk1_local.yaml \
+    --host 0.0.0.0 \
+    --port 8091 \
+    --trust-remote-code \
+    --omni
+```
+
+### 상태 확인 및 모델 목록
+
+```bash
+curl -fS http://localhost:8091/health
+curl -fS http://localhost:8091/v1/models
+```
+
+### TTS
+
+```bash
+curl -fS -X POST http://localhost:8091/v1/audio/speech \
+    -H "Content-Type: application/json" \
+    --data-binary @- \
+    --output axk2_tts.wav <<'EOF'
+{
+  "model": "KRAFTON/A.X-K2-Raon-Speech-21B-A3B",
+  "input": "Hello, this is A.X-K2 Raon Speech.",
+  "response_format": "wav"
+}
+EOF
+```
+
+### STT
+
+입력 오디오를 이식성이 높은 한 줄짜리 base64 데이터 URL로 인코딩합니다.
+
+```bash
+AUDIO_B64="$(base64 < audio.wav | tr -d '\n')"
+
+curl -fS -X POST http://localhost:8091/v1/chat/completions \
+    -H "Content-Type: application/json" \
+    --data-binary @- <<EOF
+{
+  "model": "KRAFTON/A.X-K2-Raon-Speech-21B-A3B",
+  "messages": [
+    {
+      "role": "user",
+      "content": [
+        {
+          "type": "audio_url",
+          "audio_url": {
+            "url": "data:audio/wav;base64,$AUDIO_B64"
+          }
+        },
+        {
+          "type": "text",
+          "text": "Transcribe the audio into text."
+        }
+      ]
+    }
+  ]
+}
+EOF
+```
+
 ## Training
 
 공개된 checkpoint는 다음 과정을 거친 최종 모델입니다.