[UPDATE][sglangllmbasev3][1.2.27]

sglangllmbasev3/Chart.yaml

@@ -1,6 +1,6 @@
 apiVersion: v2
-appVersion: 0.5.14-cu130
+appVersion: 0.5.17-cu130
 description: Generic SGLang + llm-init base; the model is supplied at install time via env
 name: sglangllmbasev3
 type: application
-version: 1.2.25
+version: 1.2.27

sglangllmbasev3/OlaresManifest.yaml

@@ -7,7 +7,7 @@ metadata:
   description: "Generic SGLang engine base. Pick any HuggingFace Safetensors model at install via env."
   appid: sglangllmbasev3
   title: SGLang Engine Base
-  version: '1.2.25'
+  version: '1.2.27'
   categories:
     - AI
 sharedEntrances:
@@ -33,11 +33,20 @@ workloadReplicas:
   llminit: 1
 spec:
   onlyAdmin: true
-  versionName: '0.5.14-cu130'
+  versionName: '0.5.17-cu130'
   upgradeDescription: |
-    v1.2.25: Move `HF_TOKEN` into Kubernetes Secret (`secretKeyRef`); map HF env from `OLARES_SYSTEM_HUGGINGFACE_*`.
+    Upgrade SGLang from v0.5.14-cu130 to v0.5.17-cu130. Chart 1.2.27.
 
-    HF_ENDPOINT / HF_TOKEN from OLARES_SYSTEM_HUGGINGFACE_* only. Chart 1.2.24.
+    **What's Changed** (v0.5.14 → v0.5.17)
+    - New models: Kimi K3, MiniMax-H3, Inkling, GLM-5.2, LiquidAI LFM2.5, Kimi-K2.7-Code, LongCat 2.0, EmbeddingGemma, Hunyuan 3, Qwen3.6 NVFP4, and more
+    - DeepSeek-V4: GB300 day-0 serving, NVFP4 MoE, MegaMoE, sparse-prefill / indexer opts, Waterfill & LPLB MoE load balancing
+    - Speculative decoding: Spec V2 by default, DSpark, IndexShare MTP, ReplaySSM Ring Spec-Verify
+    - Parallelism & cache: DCP A2A backends, DWDP MoE prefill, decode CP for MLA, UnifiedRadixTree default, session-aware radix cache, faster engine recovery
+    - Kernels & runtime: Breakable CUDA Graph default, KDA/GDN linear-attention on Blackwell, initial Rust frontend, native Exa web_search
+    - Breaking: QServe/FBGEMM FP8 removed; NVFP4 GEMM requires FlashInfer
+
+    For detailed release notes, visit:
+    https://github.com/sgl-project/sglang/releases
   fullDescription: |
     **IMPORTANT NOTE**  
     This app is a template and cannot be used on its own. To use it, set the environment variables below to connect a specific model.

sglangllmbasev3/i18n/en-US/OlaresManifest.yaml

@@ -3,6 +3,19 @@ metadata:
   title: SGLang Engine Base
 
 spec:
+  upgradeDescription: |
+    Upgrade SGLang from v0.5.14-cu130 to v0.5.17-cu130. Chart 1.2.27.
+
+    **What's Changed** (v0.5.14 → v0.5.17)
+    - New models: Kimi K3, MiniMax-H3, Inkling, GLM-5.2, LiquidAI LFM2.5, Kimi-K2.7-Code, LongCat 2.0, EmbeddingGemma, Hunyuan 3, Qwen3.6 NVFP4, and more
+    - DeepSeek-V4: GB300 day-0 serving, NVFP4 MoE, MegaMoE, sparse-prefill / indexer opts, Waterfill & LPLB MoE load balancing
+    - Speculative decoding: Spec V2 by default, DSpark, IndexShare MTP, ReplaySSM Ring Spec-Verify
+    - Parallelism & cache: DCP A2A backends, DWDP MoE prefill, decode CP for MLA, UnifiedRadixTree default, session-aware radix cache, faster engine recovery
+    - Kernels & runtime: Breakable CUDA Graph default, KDA/GDN linear-attention on Blackwell, initial Rust frontend, native Exa web_search
+    - Breaking: QServe/FBGEMM FP8 removed; NVFP4 GEMM requires FlashInfer
+
+    For detailed release notes, visit:
+    https://github.com/sgl-project/sglang/releases
   fullDescription: |
     **IMPORTANT NOTE**  
     This app is a template and cannot be used on its own. To use it, set the environment variables below to connect a specific model.

sglangllmbasev3/i18n/zh-CN/OlaresManifest.yaml

@@ -3,6 +3,19 @@ metadata:
   title: SGLang 引擎基座
 
 spec:
+  upgradeDescription: |
+    将 SGLang 从 v0.5.14-cu130 升级至 v0.5.17-cu130。Chart 1.2.27。
+
+    **更新内容**（v0.5.14 → v0.5.17）
+    - 新增模型支持：Kimi K3、MiniMax-H3、Inkling、GLM-5.2、LiquidAI LFM2.5、Kimi-K2.7-Code、LongCat 2.0、EmbeddingGemma、Hunyuan 3、Qwen3.6 NVFP4 等
+    - DeepSeek-V4：GB300 首日支持、NVFP4 MoE、MegaMoE、sparse-prefill / indexer 优化、Waterfill 与 LPLB MoE 负载均衡
+    - 投机解码：默认 Spec V2、DSpark、IndexShare MTP、ReplaySSM Ring Spec-Verify
+    - 并行与缓存：DCP A2A 后端、DWDP MoE prefill、MLA decode CP、默认 UnifiedRadixTree、会话感知 radix cache、更快的引擎恢复
+    - 内核与运行时：默认 Breakable CUDA Graph、Blackwell 上 KDA/GDN 线性注意力、初始 Rust 前端、原生 Exa web_search
+    - 破坏性变更：移除 QServe/FBGEMM FP8；NVFP4 GEMM 需要 FlashInfer
+
+    详细发布说明请访问：
+    https://github.com/sgl-project/sglang/releases
   fullDescription: |
     **重要提示**  
     此应用是一个模板，不能单独使用。请按下方环境变量设置模型后，方可使用本应用提供模型服务。

sglangllmbasev3/templates/sglang.yaml

@@ -72,7 +72,7 @@ spec:
             defaultMode: 0555
       containers:
         - name: sglang
-          image: docker.io/lmsysorg/sglang:v0.5.14-cu130
+          image: docker.io/lmsysorg/sglang:v0.5.17-cu130
           imagePullPolicy: IfNotPresent
           command: ["/bin/sh", "/llm-init/wrappers/sglang.sh"]
           env: