diff --git a/deploy/CURRENT.md b/deploy/CURRENT.md
index 7877a00..0065fe7 100644
--- a/deploy/CURRENT.md
+++ b/deploy/CURRENT.md
@@ -11,10 +11,10 @@
 | 60.2 (6000D-2) | `glm53-nvfp4` 实验容器（09-08 晚 TP1PP8 phase，run_phase.sh 实验链进行中） | **他人实验进行中，勿动**（方案 F PD 链已拆除；GPU7 曾有外部裸金属任务）。动卡前仍先核实归属 | `profiles/pro6000/glm53_nvfp4_pro6000_pd_{prefill→decode 侧}.env` |
 | 60.3 | 无容器，但 8 卡被外部裸金属训练占用（`/data/mas/larm`，09-08 晚实测） | 外部任务，勿动（此前台账漏记） | — |
 | 60.4 | `glm53-nvfp4`（Up，:30000，restart=unless-stopped，09-09 部署） | **TP2PP4 GLM-5.3-NVFP4 在役**（60.1 方案 D 生产配方复刻：TP2/PP4/mem0.85/MRR48/chunk16384/radix-off/无投机/SM120 三件套/disable-custom-all-reduce/index_topk_freq=4，镜像 nightly-dev-20260828-daf63171；health 200、生成冒烟通过）。09-09 经授权清退外部 vllm 评测流水线（tmux `mas` 的 run_multiseed.sh 链，带 --resume 可续跑）后部署；启动脚本 `/root/deploy_s2_test_604.sh`（基于 exp602/deploy_s2_test.sh，唯一改动 restart=unless-stopped）。旧 `dsv4_scan` 容器仍在盘（Exited） | `profiles/pro6000/glm53_nvfp4_pro6000_sglang_tp2pp4.env`（同 60.1 D 口径） |
-| 60.5 | `glm53-nvfp4`（Up 29h） | **NVFP4 团队生产**（deploy_glm53_605.sh，md5 fcd9109b）。生产机铁律：不实验、不重启、不覆盖脚本 | `profiles/pro6000/glm53_nvfp4_pro6000_sglang_tp8eagle.env`（方案 A 口径） |
+| 60.5 | `glm53-nvfp4`（Up 29h） | **NVFP4 团队生产**（deploy_glm53_605.sh，md5 fcd9109b）。生产机铁律：不实验、不重启、不覆盖脚本。**09-09 已交付容量扩容 v2 脚本（TP2PP4-hicache 优胜配置，池 909,632/c6/单条 909k，与 60.8 现役同款），未执行，择维护窗口跑 `/root/deploy_glm53_605_v2.sh`；回滚=原脚本** | `profiles/pro6000/glm53_nvfp4_pro6000_sglang_tp8eagle.env`（方案 A 口径）；待切 `glm53_nvfp4_pro6000_sglang_tp2pp4_hicache.env` |
 | 60.6 | 无容器，但 8 卡被外部裸金属实验占用（`/data/hzy/sparse-opd-*`，09-08 晚实测） | 外部任务，勿动（此前台账漏记） | — |
 | 60.7 | 基本空（4 卡仍有 `/home/user/dirA_exp` 外部小任务，09-08 晚实测） | 09-08 已拆除清空（方案 F 前身单机实验 + 场景一深优资产留盘），不再恢复 | — |
-| 60.8 | `glm53-nvfp4`（Up，:30000，restart=unless-stopped） | **r37 PP+MTP verify-graph 口径在役**（09-08 深夜取代 A16：TP4PP2 + EAGLE 3/1/4 + verify CUDA graph + draft eager + de-GLOO + SYNC_MASK=127，9 挂载文件，启动 `bash /root/deploy_ppmtp_r37.sh '--tp 4 --pp-size 2 --disable-overlap-schedule --max-prefill-tokens 16384' mtp 8192 0.88 1 1 0 0`）。质量门 7/7 ×2、conc_test×3、killer×6 零崩溃；killer cc16 双种子 **63.8/64.7s vs A16 105.6/96.0s（-40%）**、TPOT 41.5/39.3ms、TTFT 33s vs 58-61s（近乎减半）。修复根因=pre-planned 早退路径缺 pp_proxy 补拷（详见 `experiments/pro6000/glm53_ppmtp_r37_verify_graph/`）。A16 切回脚本 deploy_glm53_605.sh 留盘；sglang_patch2/eagle_worker_common.py 已升级 bisect 版（"1"/"0" 语义兼容） | `experiments/pro6000/glm53_ppmtp_r37_verify_graph/`（patches+scripts+results 全量）；前史 `experiments/pro6000/glm53_ppmtp_r36_degloo/` |
+| 60.8 | `glm53-nvfp4`（Up，:30000，restart=unless-stopped） | **TP2PP4-nomtp 容量扩容口径在役**（09-09 按用户决定取代 r37：TP2PP4 + radix/hicache 开 + ctx 1048576 + cu13 栈 9 挂载 + de-GLOO，memfrac 0.85/chunk 8192/MRR 16，KV 池 **909,632**，启动 `bash /root/deploy_ppmtp_r37.sh '--tp 2 --pp-size 4 --disable-overlap-schedule --max-prefill-tokens 16384 --disable-custom-all-reduce --context-length 1048576' nomtp 8192 0.85`）。四臂对拍优胜（vs A-mirror/r37/B'，i128k/o512 c1-4）：c4 输入/输出 **5,907/23.1 tok/s**、TTFT p50 43.4s、并发上限 c6、单条 ~909k（900k 实跑）、512k 单条 TTFT 88.4s、90% 命中 c4 输入 17,625 tok/s；质量门 7/7。r37 的 MTP 优势区间是 i16k/cc≤16，切回 = deploy_ppmtp_r37.sh mtp 模式（详见 `experiments/pro6000/glm53_nvfp4_128k_capacity_topology/`） | `profiles/pro6000/glm53_nvfp4_pro6000_sglang_tp2pp4_hicache.env`；实验全量 `experiments/pro6000/glm53_nvfp4_128k_capacity_topology/`；前史 r37 `experiments/pro6000/glm53_ppmtp_r37_verify_graph/` |
 
 ## 方案 A-F 一览（GLM-5.3-NVFP4 @ pro6000，2026-09-08 双场景报告口径）
 
diff --git a/deploy/profiles/pro6000/glm53_nvfp4_pro6000_sglang_tp2pp4_hicache.env b/deploy/profiles/pro6000/glm53_nvfp4_pro6000_sglang_tp2pp4_hicache.env
new file mode 100644
index 0000000..fad79de
--- /dev/null
+++ b/deploy/profiles/pro6000/glm53_nvfp4_pro6000_sglang_tp2pp4_hicache.env
@@ -0,0 +1,44 @@
+# GLM-5.3-NVFP4 SGLang TP=2 PP=4 + hicache deployment profile (single RTX 6000D node, 8 GPUs).
+# 2026-09-09 128k 低并发容量扩容实验优胜配置（60.8 现役；60.5 交付 deploy_glm53_605_v2.sh 待执行）。
+# 可执行部署脚本：experiments/pro6000/glm53_nvfp4_128k_capacity_topology/scripts/deploy_glm53_605_v2.sh
+#
+# 与方案 D（glm53_nvfp4_pro6000_sglang_tp2pp4.env）的关键差异（勿混淆）：
+#   - radix/hicache 保持开启（60.5 真实流量命中率 90%+，关 radix 不可接受）；D 为 16k 独立输入场景关了 radix
+#   - cu13 镜像 + 9 补丁只读挂载（de-GLOO request_receiver / decode_cuda_graph_runner_fix 等 r37 栈遗产，
+#     nomtp 下 spec 相关补丁为惰性，de-GLOO 为 PP 通用修复）；D 用旧镜像 20260828 无挂载
+#   - --reasoning-parser glm45 --tool-call-parser glm47 齐备（质量门 7/7）；D 当时 6/7
+#   - --context-length 1048576（模型原生 1M；D 未设）；chunk 8192（D 16384）；MRR 16（D 48）
+#   - 实测（60.8，i128k/o512 冷缓存）：KV 池 909,632 token（A 的 3.29×）、并发上限 c6、
+#     单条上限 ~909k（900k 实跑通过）、c4 输入/输出 5,907/23.1 tok/s、TTFT p50 43.4s、
+#     512k 单条 TTFT 88.4s、90% 命中 c4 输入 17,625 tok/s
+#   - memfrac 0.85 为验证档；PP0 stage 空闲 22GB 提示 0.88 有余量（未验证，改动须重跑质量门+容量冒烟）
+PLATFORM=pro6000
+EXPERIMENT=glm53_nvfp4_128k_capacity_topology
+MODEL_NAME=GLM-5.3-NVFP4
+ENGINE=sglang
+RUNTIME=docker
+DOCKER_IMAGE=lmsysorg/sglang:nightly-dev-cu13-20260901-07c8f729
+CONTAINER_NAME=glm53-nvfp4
+MODEL_PATH=/data/hf_models/GLM-5.3-NVFP4
+SERVED_MODEL_NAME=/data/hf_models/GLM-5.3-NVFP4
+PORT=30000
+HEALTH_PATH=/health
+HEALTH_WAIT_S=1800
+CONTAINER_PYTHON=python3
+
+TP=2
+PP=4
+MEM_FRACTION_STATIC=0.85
+MAX_RUNNING_REQUESTS=16
+CHUNKED_PREFILL_SIZE=8192
+CONTEXT_LENGTH=1048576
+
+DEVICE_VARS="CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7"
+ENGINE_ENV="PYTHONUNBUFFERED=1 SGLANG_PP_DEGLOO=1 SGLANG_PP_SPEC_FORCE_EAGER_DRAFT=1 SGLANG_PP_FORCE_EAGER_VERIFY=0 SGLANG_PP_SPEC_DEBUG=0"
+
+DOCKER_FLAGS="--gpus all --shm-size 64g --ipc=host --cap-add SYS_PTRACE -p ${PORT}:${PORT}"
+VOLUMES="/data/hf_models:/data/hf_models + 9 patch ro-mounts (full list in scripts/deploy_glm53_605_v2.sh; files live in /root + /root/sglang_patch2 on the host)"
+
+BOOTSTRAP="python3 -m sglang.launch_server ${LAUNCH_ARGS}"
+
+LAUNCH_ARGS="--model-path ${MODEL_PATH} --tp-size ${TP} --pp-size ${PP} --mem-fraction-static ${MEM_FRACTION_STATIC} --max-running-requests ${MAX_RUNNING_REQUESTS} --chunked-prefill-size ${CHUNKED_PREFILL_SIZE} --disable-shared-experts-fusion --moe-runner-backend flashinfer_cutlass --disable-flashinfer-autotune --reasoning-parser glm45 --tool-call-parser glm47 --enable-hierarchical-cache --hicache-ratio 3 --disable-overlap-schedule --max-prefill-tokens 16384 --disable-custom-all-reduce --context-length ${CONTEXT_LENGTH} --host 0.0.0.0 --port ${PORT}"
diff --git a/experiments/pro6000/glm53_nvfp4_128k_capacity_topology/README.md b/experiments/pro6000/glm53_nvfp4_128k_capacity_topology/README.md
new file mode 100644
index 0000000..4bc0e1f
--- /dev/null
+++ b/experiments/pro6000/glm53_nvfp4_128k_capacity_topology/README.md
@@ -0,0 +1,135 @@
+# GLM-5.3-NVFP4 128k 低并发容量扩容与拓扑选型实验报告 — 2026-09-09 @ 174.1.60.8
+
+## 一、结论速览
+
+**优胜配置 = TP2PP4 nomtp**（radix/hicache 保持开启、`--context-length 1048576`、cu13 栈 9 挂载、memfrac 0.85、chunk 8192）。
+
+| 维度 | 60.5 现役（TP8+EAGLE） | 优胜配置（TP2PP4 nomtp） | 提升 |
+|---|---|---|---|
+| KV 池（token） | 276,864 | **909,632** | 3.29× |
+| i128k 并发上限 | c2（c3 起排队） | **c6**（c7 起排队，实测） | 3× |
+| 单条请求上限 | 264k（ctx 270,336 参数所限） | **~909k**（900k 实跑通过） | 3.4× |
+| c4 输入吞吐 | 3,228 tok/s | **5,907 tok/s** | 1.83× |
+| c4 输出吞吐 | 12.6 tok/s | **23.1 tok/s** | 1.83× |
+| c4 TTFT p50 | 119.6s | **43.4s** | −64% |
+| c4 e2e p50 | 161.3s | **88.8s** | −45% |
+| 90% 命中真实流量形态（c4） | — | 输入 **17,625 tok/s**、TTFT 7.3s | — |
+
+- 质量门 7/7 通过（GSM8K×5 + 中文推理 + tool call，PP4+radix+hicache+parser 新组合无 correctness 问题）。
+- **60.8 已按用户决定部署优胜配置留役**（restart=unless-stopped，替代 r37）。
+- **60.5 交付 `deploy_glm53_605_v2.sh`（只交未执行）**，原 `deploy_glm53_605.sh` 原样留盘即回滚路径。
+
+## 二、问题诊断（60.5 生产日志实证）
+
+现役容器 `glm53-nvfp4`（TP8 + EAGLE 4/1/5，memfrac 0.90，fp8 KV，hicache-ratio 3，`--context-length 270336`）：
+
+1. **"单条最高 256k" 不是模型限制**：模型原生 `max_position_embeddings = 1,048,576`（1M）。上限来自部署参数 `--context-length 270336`，且 KV 池 276,864 也刚好只装得下一条 26.4 万请求。
+2. **排队是纯容量算术**：单条 16~22 万 token 请求占池 0.58~0.90；近期一万行日志 ~13% 的 decode 步有 `#queue-req>0`，09-08 有一次池满 retract。日志实例：一条 173k（95% 命中）请求在 160k 请求 decode 期间排队无法准入。
+3. **命中不省并发容量**：radix/hicache 只加速重复前缀的 prefill；不同文档的并发请求各自全额驻留 KV，照样排队。60.5 真实流量 12万~22万 token、命中率 90%+，仍然撞容量墙。
+4. **TP8 调参无解（结构性）**：每 token KV 57.3KB（MLA 44.9 + DSA indexer ~10），fp8_e4m3 已是断言封死的硬顶（FP4 启动即 AssertionError）；TP8 全系调参上限 314,944（60.6 实测 0.89+hicache4）< c3×128k 需求 394,752。
+
+## 三、机理：为什么只有 PP 拓扑能解
+
+GLM-5.3 是 MLA/DSA 家族，**KV 在 TP 组内每卡全量复制、只按 PP 分层切**：TP8 每卡存全部 78 层 KV（15.85GB→276,864 token）；PP2 每卡存半数层（池×2）；PP4 每卡存 1/4 层（池×4，实际因 hicache/图开销打折）。**KV 池容量 ≈ PP 度数倍增，与 TP 度数无关**。这就是 TP4PP2≈2.37×、TP2PP4≈3.29×（实测）的来源。另：MTP/EAGLE 的 draft KV + verify 图额外吃池（r37 同 memfrac 下 589,696→384,960，−35%）。
+
+## 四、实验方法
+
+- **平台**：60.8（8×RTX 6000D 85.6GB，与 60.5 同型）；四配置臂各部署后跑同窗口。
+- **场景**：i131072 / o512 / cc∈{1,2,3,4}，nreq=8/点，PG19 真实语料（bench_corpus.py，input_ids 直发 /generate，temp 0，ignore_eos，stream）。
+- **公平性协议**：每点独立语料窗口（cc1→base 10M，cc2→11,048,576，cc3→12,097,152，cc4→13,145,728），**所有臂用同一映射=同文本**；每点前 `POST /flush_cache`（实测连 host 层一起清，复跑命中率 0.0）；每臂先做一次不计分的 131k 形状预热（吸收内核首触成本，实测不做会污染首点：16k 首跑 1653 vs 复跑 4526 tok/s）。
+- **指标口径**（沿用 rev20 双场景协议）：输入吞吐 = 总输入 token/墙钟；输出吞吐 = 服务端 completion_tokens 总和/墙钟；TTFT = 发出→首个流式响应（含排队）；命中率从 TP0 Prefill 日志核算。queue-req 日志用于排队取证。
+- **配置臂**：
+  - **arm0 r37**：60.8 在役原样（TP4PP2+EAGLE 3/1/4+verify 图，memfrac 0.88，池 384,960）
+  - **arm1 A-mirror**：60.5 现役逐参数复刻（旧镜像 20260828，池 276,864，ctx 270,336）
+  - **arm2 B'**：TP4PP2 nomtp，memfrac 0.88，ctx 524,288（池 589,696）
+  - **arm3 TP2PP4**：TP2PP4 nomtp，memfrac 0.85，ctx 1048576，radix/hicache 开，含 glm45/glm47 parser（池 909,632）
+
+## 五、主扫数据（i128k/o512，冷缓存）
+
+**输入吞吐（tok/s）**
+
+| cc | A-mirror（池276,864） | r37（池384,960） | B'（池589,696） | TP2PP4（池909,632） |
+|---|---|---|---|---|
+| 1 | 3,026 | **4,110** | 3,737 | 3,262 |
+| 2 | 3,142 | **4,754** | 4,619 | 4,635 |
+| 3 | 3,214 | 5,016 | 4,843 | **5,017** |
+| 4 | 3,228 | 5,061 | 5,256 | **5,907** |
+
+**输出吞吐（tok/s）**
+
+| cc | A-mirror | r37 | B' | TP2PP4 |
+|---|---|---|---|---|
+| 1 | 11.8 | **16.1** | 14.6 | 12.7 |
+| 2 | 12.3 | **18.6** | 18.0 | 18.1 |
+| 3 | 12.6 | 19.6 | 18.9 | **19.6** |
+| 4 | 12.6 | 19.8 | 20.5 | **23.1** |
+
+**TTFT p50 / max（秒）**（排队签名 = p50 跳升整请求时长倍数）
+
+| cc | A-mirror | r37 | B' | TP2PP4 |
+|---|---|---|---|---|
+| 1 | 37.8 | 21.1 | 21.0 | **15.5** |
+| 2 | 37.8 / 76.9 | 21.1 / 41.7 | 41.3 | 29.0 |
+| 3 | **77.2** / 120.3 | 43.6 / 73.8 | 41.3 / 61.6 | **29.6** / 43.4 |
+| 4 | **119.6** / 156.4 | 72.2 / 93.9 | 61.6 / 81.9 | **43.4** / 57.3 |
+
+**e2e p50 / max（秒）**
+
+| cc | A-mirror | r37 | B' | TP2PP4 |
+|---|---|---|---|---|
+| 1 | 43.8 | **31.7** | 35.0 | 40.2 |
+| 2 | 83.7 / 121.6 | **55.3** / 76.6 | 56.8 | 56.5 |
+| 3 | 121.7 / 163.3 | **75.2** / 105.1 | 80.0 | 75.8 |
+| 4 | 161.3 / 201.8 | 103.3 / 125.2 | 99.7 | **88.8** |
+
+**排队证据与机理注解**：
+- A-mirror：池只容 2 条并发，c3/c4 排队（TTFT p50 77→120s）；c4 输入吞吐被排队锁死在 3,228（prefill 带宽根本没用满）。
+- r37：池 384,960 同样只容 2 条 + 1 条错峰补位，c3/c4 排队（TTFT p50 43.6→72.2s；日志 518 行 queue-req>0）。**60.8 原在役配置在本场景也不合格。**
+- B'：c3/c4 全并发准入、零容量排队（稳态 usage 精确停在 0.22/0.45/0.67/0.89；e2e p50≈max 波内同步）。日志中的 queue-req 行是 chunk 调度/波间 radix 逐出瞬态（秒级），非容量排队——判据：TTFT 缩放=纯 prefill 带宽分摊（cc2 p50=max=41.3≈2×21.0）。
+- TP2PP4：同上零容量排队，且 TTFT 全场最优（4 级流水 prefill 重叠最深：cc1 TTFT 15.5s、串行 prefill 等效 ~8.5k tok/s）；代价是 TPOT 最慢（cc1 48.4ms vs B' 27.3 vs r37 20.7 vs A 12.0），c1 短输出场景 e2e 吃亏（40.2s），c2 起被并发摊平、c3/c4 反超。
+- MTP/EAGLE 观察：accept 随并发上升（r37 2.91→3.64，A-mirror 3.31→4.26，verify batch 越大接受越高）；但 MTP 的 35% 池代价在本容量场景不划算——r37 全程吞吐被 B'/TP2PP4 压制或打平。
+
+## 六、优胜者（TP2PP4）附加验证
+
+| 验证项 | 结果 |
+|---|---|
+| 质量门 | **7/7**（GSM8K 72/3/60/63/10 + 鸡兔同笼 23 + tool call get_weather北京） |
+| 512k 单条（523,776+512 真实语料） | ✅ TTFT 88.4s，e2e 114.4s，TPOT 50.9ms（旧配置直接拒绝） |
+| 900k 单条（900,000+512） | ✅ TTFT 210.6s，e2e 237.8s —— **单条上限 ~909k 实证**（池减 512 后的余量） |
+| cc5 | ✅ in 5,747 / out 22.5，e2e p50≈max 107s，零容量排队 |
+| cc6 | ✅ in 5,849 / out 22.9，e2e p50≈max 122.7s，零容量排队 —— **并发上限 = c6** |
+| cc7 | ⚠️ 第 7 条排队（TTFT max 138.7s、e2e max 179.7s）—— 边界与池算术吻合（7×131,584=921,088 > 909,632） |
+| 90% 命中 c4（60.5 真实流量形态，实测命中 0.8999） | in **17,625** / out 68.9 tok/s，TTFT p50 **7.3s**，e2e 29.9s —— radix 去重+hicache 对重复查询流量再放大 ~3× |
+
+## 七、方案对比与推荐
+
+| 方案 | 池 | c4 in/out | 容量定位 | 判定 |
+|---|---|---|---|---|
+| A（TP8+EAGLE，60.5 现役） | 276,864 | 3,228/12.6 | c2、单条 264k | 本场景被全面支配，淘汰 |
+| r37（TP4PP2+MTP，60.8 昨日在役） | 384,960 | 5,061/19.8 | c2、单条 ~380k | c3 起排队，容量不合格；仅 c1 最优 |
+| B'（TP4PP2 nomtp） | 589,696 | 5,256/20.5 | c4、单条 ~588k（512k 单条会占 89% 池、堵死并发） | 强力备选：16k 短文本场景历史成绩好 |
+| **TP2PP4 nomtp（优胜）** | **909,632** | **5,907/23.1** | **c6、单条 ~909k、512k 单条+2 并发共存** | 用户需求（512k 单条+长上下文为主）下的正解 |
+
+**推荐**：60.5 采用 TP2PP4 nomtp（`deploy_glm53_605_v2.sh`，参数与 60.8 现役完全一致）。决策依据：c3/c4/c5/c6 吞吐全场第一 + 唯一满足"512k 单条与并发共存" + 质量门 7/7 + TTFT 全档最优。若 60.5 未来 16k 短文本流量占比显著上升，再评估切 B'（其 s1 短文本历史成绩比 TP2PP4 好 27~48%）。
+
+**留观旋钮**（未验证，报告只记录不推荐）：memfrac 0.85→0.88 或可再抬池（PP0 空闲 22GB 最大的 stage 不均匀提示有空间）；TP2PP4+MTP 可修 c1 短输出短板（池约降至 ~59 万，恰为 B' 水平）；hicache-ratio 3→4 扩 host 前缀池（60.6 在 TP8 验证过）。
+
+**r37 角色变化说明**：r37 的 MTP 优势区间是 i16k/cc≤16（09-09 判决 +6%@cc16），本次让位给容量优先的 TP2PP4 是按用户明确选择执行；若 60.8 未来主要服务短文本低并发，可用 deploy_ppmtp_r37.sh mtp 模式一键切回。
+
+## 八、已证伪 / 排除项（本轮+引用前判）
+
+- TP8 任何调参：fp8 KV 断言封顶 + 上限 314,944 < c3 需求 394,752（60.6 数据）。
+- r37 现役直接顶上：池 384,960，c3 差 1 万 token 仍排队（本轮实测）。
+- MTP/EAGLE 换 KV 池：draft+verify 图 −35% 池，本场景不划算。
+- HiSparse 超池驻留：本 nightly 自旋不可用（60.6 前判）。
+- TP1PP8：池 2M 但吞吐 −12~33%（60.2 前判），仅当需要 >c6 且能接受慢时再议。
+- PD 分离/DP/EP：前判劣化或不可用，与本问题正交。
+
+## 九、资产与复现
+
+- **60.8**：容器 glm53-nvfp4:30000 = 优胜配置，restart=unless-stopped。启动命令 = `bash /root/deploy_ppmtp_r37.sh '--tp 2 --pp-size 4 --disable-overlap-schedule --max-prefill-tokens 16384 --disable-custom-all-reduce --context-length 1048576' nomtp 8192 0.85`。实验日志全量：`/root/bench_logs/128kexp/`（26 点 SUMMARY 已提取为 all_summaries.json）。
+- **60.5**：交付 `deploy_glm53_605_v2.sh`（内嵌完整 docker run+前置检查+显存归零等待+健康门+回滚提示）；回滚 = `bash /root/deploy_glm53_605.sh`。执行前置检查会列出缺失的补丁/镜像清单（60.8 /root 均有）。
+- **压测复现**：`bash /root/arm_runner.sh <arm>`（c1-4 四点+形状预热）、`bash /root/val_runner.sh <arm>`（512k/900k/cc5-7/hit90）；语料窗口映射与 flush 协议见第四节；质量门 `bash /root/quality_gate_605.sh`。
+- **仓库归档**：`experiments/pro6000/glm53_nvfp4_128k_capacity_topology/`（README=本报告、scripts、results/20260909）；profile `deploy/profiles/pro6000/glm53_nvfp4_pro6000_sglang_tp2pp4_hicache.env`；CURRENT.md 已更新 60.8 行。
+
+*报告与数据：ZCode 实验 2026-09-09；压测窗口 run-id 9501-9507/9601-9604/9990-9991；全部冷缓存口径。*
diff --git a/experiments/pro6000/glm53_nvfp4_128k_capacity_topology/results/20260909/all_summaries.json b/experiments/pro6000/glm53_nvfp4_128k_capacity_topology/results/20260909/all_summaries.json
new file mode 100644
index 0000000..7d74ddc
--- /dev/null
+++ b/experiments/pro6000/glm53_nvfp4_128k_capacity_topology/results/20260909/all_summaries.json
@@ -0,0 +1,1484 @@
+{
+ "arm0_r37_cc1": {
+  "concurrency": 1,
+  "num_requests": 8,
+  "run_id": 9501,
+  "corpus_window": {
+   "start": 10000000,
+   "end": 11048576
+  },
+  "ok": 8,
+  "failed": 0,
+  "wall_s": 255.13,
+  "input_len": 131072,
+  "shared_len": 0,
+  "unique_len": 131072,
+  "output_len": 512,
+  "output_tokens_total": 4096,
+  "output_throughput_tok_s": 16.05,
+  "input_throughput_tok_s": 4109.96,
+  "ttft_s": {
+   "mean": 21.1436,
+   "p50": 21.1351,
+   "max": 21.1968,
+   "min": 21.1318
+  },
+  "tpot_s": {
+   "mean": 0.021,
+   "p50": 0.0207,
+   "max": 0.0255,
+   "min": 0.0176
+  },
+  "e2e_s": {
+   "mean": 31.8909,
+   "p50": 31.6992,
+   "max": 34.1398,
+   "min": 30.1727
+  },
+  "per_req_out_tok_s_e2e": {
+   "mean": 16.0805,
+   "p50": 16.3007,
+   "max": 16.969,
+   "min": 14.9972
+  },
+  "per_req_decode_tok_s": {
+   "mean": 48.3191,
+   "p50": 49.8219,
+   "max": 57.0435,
+   "min": 39.3633
+  },
+  "spec_accept_length_mean": 2.912,
+  "retractions_total": 0,
+  "cache_hit_from_logs": {
+   "prefill_batches": 256,
+   "new_tokens": 2097152,
+   "cached_tokens": 0,
+   "hit_rate": 0.0
+  }
+ },
+ "arm0_r37_cc2": {
+  "concurrency": 2,
+  "num_requests": 8,
+  "run_id": 9502,
+  "corpus_window": {
+   "start": 11048576,
+   "end": 12097152
+  },
+  "ok": 8,
+  "failed": 0,
+  "wall_s": 220.58,
+  "input_len": 131072,
+  "shared_len": 0,
+  "unique_len": 131072,
+  "output_len": 512,
+  "output_tokens_total": 4096,
+  "output_throughput_tok_s": 18.57,
+  "input_throughput_tok_s": 4753.73,
+  "ttft_s": {
+   "mean": 23.7214,
+   "p50": 21.1489,
+   "max": 41.6861,
+   "min": 21.1264
+  },
+  "tpot_s": {
+   "mean": 0.0613,
+   "p50": 0.0669,
+   "max": 0.0689,
+   "min": 0.0242
+  },
+  "e2e_s": {
+   "mean": 55.0699,
+   "p50": 55.3137,
+   "max": 76.6223,
+   "min": 33.5372
+  },
+  "per_req_out_tok_s_e2e": {
+   "mean": 9.719,
+   "p50": 9.3973,
+   "max": 15.2666,
+   "min": 6.6821
+  },
+  "per_req_decode_tok_s": {
+   "mean": 18.3311,
+   "p50": 15.3492,
+   "max": 41.3304,
+   "min": 14.5499
+  },
+  "spec_accept_length_mean": 2.672,
+  "retractions_total": 0,
+  "cache_hit_from_logs": {
+   "prefill_batches": 256,
+   "new_tokens": 2097152,
+   "cached_tokens": 0,
+   "hit_rate": 0.0
+  }
+ },
+ "arm0_r37_cc3": {
+  "concurrency": 3,
+  "num_requests": 8,
+  "run_id": 9503,
+  "corpus_window": {
+   "start": 12097152,
+   "end": 13145728
+  },
+  "ok": 8,
+  "failed": 0,
+  "wall_s": 209.05,
+  "input_len": 131072,
+  "shared_len": 0,
+  "unique_len": 131072,
+  "output_len": 512,
+  "output_tokens_total": 4096,
+  "output_throughput_tok_s": 19.59,
+  "input_throughput_tok_s": 5016.0,
+  "ttft_s": {
+   "mean": 45.9651,
+   "p50": 43.5894,
+   "max": 73.7557,
+   "min": 21.2775
+  },
+  "tpot_s": {
+   "mean": 0.0509,
+   "p50": 0.0613,
+   "max": 0.0655,
+   "min": 0.0182
+  },
+  "e2e_s": {
+   "mean": 71.9541,
+   "p50": 75.2035,
+   "max": 105.0954,
+   "min": 51.2337
+  },
+  "per_req_out_tok_s_e2e": {
+   "mean": 7.5561,
+   "p50": 6.8525,
+   "max": 9.9934,
+   "min": 4.8718
+  },
+  "per_req_decode_tok_s": {
+   "mean": 25.7245,
+   "p50": 16.4485,
+   "max": 55.2037,
+   "min": 15.3031
+  },
+  "spec_accept_length_mean": 3.493,
+  "retractions_total": 0,
+  "cache_hit_from_logs": {
+   "prefill_batches": 256,
+   "new_tokens": 2097152,
+   "cached_tokens": 0,
+   "hit_rate": 0.0
+  }
+ },
+ "arm0_r37_cc4": {
+  "concurrency": 4,
+  "num_requests": 8,
+  "run_id": 9504,
+  "corpus_window": {
+   "start": 13145728,
+   "end": 14194304
+  },
+  "ok": 8,
+  "failed": 0,
+  "wall_s": 207.18,
+  "input_len": 131072,
+  "shared_len": 0,
+  "unique_len": 131072,
+  "output_len": 512,
+  "output_tokens_total": 4096,
+  "output_throughput_tok_s": 19.77,
+  "input_throughput_tok_s": 5061.26,
+  "ttft_s": {
+   "mean": 67.6095,
+   "p50": 72.2484,
+   "max": 93.9173,
+   "min": 21.2909
+  },
+  "tpot_s": {
+   "mean": 0.0498,
+   "p50": 0.0608,
+   "max": 0.1012,
+   "min": 0.0183
+  },
+  "e2e_s": {
+   "mean": 93.0788,
+   "p50": 103.3347,
+   "max": 125.1566,
+   "min": 51.1124
+  },
+  "per_req_out_tok_s_e2e": {
+   "mean": 5.8997,
+   "p50": 4.9549,
+   "max": 10.0171,
+   "min": 4.0909
+  },
+  "per_req_decode_tok_s": {
+   "mean": 33.4842,
+   "p50": 52.2793,
+   "max": 54.7641,
+   "min": 9.898
+  },
+  "spec_accept_length_mean": 3.635,
+  "retractions_total": 0,
+  "cache_hit_from_logs": {
+   "prefill_batches": 256,
+   "new_tokens": 2097152,
+   "cached_tokens": 0,
+   "hit_rate": 0.0
+  }
+ },
+ "arm0_r37_warmup": {
+  "concurrency": 1,
+  "num_requests": 1,
+  "run_id": 9991,
+  "corpus_window": {
+   "start": 17000000,
+   "end": 17131072
+  },
+  "ok": 1,
+  "failed": 0,
+  "wall_s": 23.66,
+  "input_len": 131072,
+  "shared_len": 0,
+  "unique_len": 131072,
+  "output_len": 64,
+  "output_tokens_total": 64,
+  "output_throughput_tok_s": 2.7,
+  "input_throughput_tok_s": 5539.35,
+  "ttft_s": {
+   "mean": 21.2651,
+   "p50": 21.2651,
+   "max": 21.2651,
+   "min": 21.2651
+  },
+  "tpot_s": {
+   "mean": 0.038,
+   "p50": 0.038,
+   "max": 0.038,
+   "min": 0.038
+  },
+  "e2e_s": {
+   "mean": 23.6611,
+   "p50": 23.6611,
+   "max": 23.6611,
+   "min": 23.6611
+  },
+  "per_req_out_tok_s_e2e": {
+   "mean": 2.7049,
+   "p50": 2.7049,
+   "max": 2.7049,
+   "min": 2.7049
+  },
+  "per_req_decode_tok_s": {
+   "mean": 26.7145,
+   "p50": 26.7145,
+   "max": 26.7145,
+   "min": 26.7145
+  },
+  "spec_accept_length_mean": 2.133,
+  "retractions_total": 0,
+  "cache_hit_from_logs": {
+   "prefill_batches": 32,
+   "new_tokens": 262144,
+   "cached_tokens": 0,
+   "hit_rate": 0.0
+  }
+ },
+ "arm1_Amirror_cc1": {
+  "concurrency": 1,
+  "num_requests": 8,
+  "run_id": 9501,
+  "corpus_window": {
+   "start": 10000000,
+   "end": 11048576
+  },
+  "ok": 8,
+  "failed": 0,
+  "wall_s": 346.56,
+  "input_len": 131072,
+  "shared_len": 0,
+  "unique_len": 131072,
+  "output_len": 512,
+  "output_tokens_total": 4096,
+  "output_throughput_tok_s": 11.82,
+  "input_throughput_tok_s": 3025.64,
+  "ttft_s": {
+   "mean": 37.7456,
+   "p50": 37.7503,
+   "max": 37.7573,
+   "min": 37.7043
+  },
+  "tpot_s": {
+   "mean": 0.0109,
+   "p50": 0.012,
+   "max": 0.0132,
+   "min": 0.0083
+  },
+  "e2e_s": {
+   "mean": 43.32,
+   "p50": 43.8174,
+   "max": 44.4887,
+   "min": 41.9741
+  },
+  "per_req_out_tok_s_e2e": {
+   "mean": 11.8244,
+   "p50": 11.7971,
+   "max": 12.198,
+   "min": 11.5085
+  },
+  "per_req_decode_tok_s": {
+   "mean": 94.6188,
+   "p50": 90.5656,
+   "max": 121.3763,
+   "min": 76.065
+  },
+  "spec_accept_length_mean": 3.314,
+  "retractions_total": 0,
+  "cache_hit_from_logs": {
+   "prefill_batches": 128,
+   "new_tokens": 1048576,
+   "cached_tokens": 0,
+   "hit_rate": 0.0
+  }
+ },
+ "arm1_Amirror_cc2": {
+  "concurrency": 2,
+  "num_requests": 8,
+  "run_id": 9502,
+  "corpus_window": {
+   "start": 11048576,
+   "end": 12097152
+  },
+  "ok": 8,
+  "failed": 0,
+  "wall_s": 333.76,
+  "input_len": 131072,
+  "shared_len": 0,
+  "unique_len": 131072,
+  "output_len": 512,
+  "output_tokens_total": 4096,
+  "output_throughput_tok_s": 12.27,
+  "input_throughput_tok_s": 3141.72,
+  "ttft_s": {
+   "mean": 42.8869,
+   "p50": 37.7865,
+   "max": 76.9116,
+   "min": 37.7742
+  },
+  "tpot_s": {
+   "mean": 0.0793,
+   "p50": 0.0892,
+   "max": 0.1617,
+   "min": 0.0132
+  },
+  "e2e_s": {
+   "mean": 83.4182,
+   "p50": 83.749,
+   "max": 121.6486,
+   "min": 44.5522
+  },
+  "per_req_out_tok_s_e2e": {
+   "mean": 6.5603,
+   "p50": 6.1412,
+   "max": 11.4921,
+   "min": 4.2088
+  },
+  "per_req_decode_tok_s": {
+   "mean": 26.5566,
+   "p50": 11.3508,
+   "max": 75.6796,
+   "min": 6.1948
+  },
+  "spec_accept_length_mean": 2.922,
+  "retractions_total": 0,
+  "cache_hit_from_logs": {
+   "prefill_batches": 128,
+   "new_tokens": 1048576,
+   "cached_tokens": 0,
+   "hit_rate": 0.0
+  }
+ },
+ "arm1_Amirror_cc3": {
+  "concurrency": 3,
+  "num_requests": 8,
+  "run_id": 9503,
+  "corpus_window": {
+   "start": 12097152,
+   "end": 13145728
+  },
+  "ok": 8,
+  "failed": 0,
+  "wall_s": 326.28,
+  "input_len": 131072,
+  "shared_len": 0,
+  "unique_len": 131072,
+  "output_len": 512,
+  "output_tokens_total": 4096,
+  "output_throughput_tok_s": 12.55,
+  "input_throughput_tok_s": 3213.76,
+  "ttft_s": {
+   "mean": 78.0816,
+   "p50": 77.2216,
+   "max": 120.3096,
+   "min": 38.9908
+  },
+  "tpot_s": {
+   "mean": 0.076,
+   "p50": 0.0846,
+   "max": 0.1619,
+   "min": 0.0095
+  },
+  "e2e_s": {
+   "mean": 116.8932,
+   "p50": 121.7292,
+   "max": 163.3489,
+   "min": 82.0943
+  },
+  "per_req_out_tok_s_e2e": {
+   "mean": 4.5817,
+   "p50": 4.2567,
+   "max": 6.2367,
+   "min": 3.1344
+  },
+  "per_req_decode_tok_s": {
+   "mean": 30.6847,
+   "p50": 11.8916,
+   "max": 105.0827,
+   "min": 6.1882
+  },
+  "spec_accept_length_mean": 4.019,
+  "retractions_total": 0,
+  "cache_hit_from_logs": {
+   "prefill_batches": 128,
+   "new_tokens": 1048576,
+   "cached_tokens": 0,
+   "hit_rate": 0.0
+  }
+ },
+ "arm1_Amirror_cc4": {
+  "concurrency": 4,
+  "num_requests": 8,
+  "run_id": 9504,
+  "corpus_window": {
+   "start": 13145728,
+   "end": 14194304
+  },
+  "ok": 8,
+  "failed": 0,
+  "wall_s": 324.86,
+  "input_len": 131072,
+  "shared_len": 0,
+  "unique_len": 131072,
+  "output_len": 512,
+  "output_tokens_total": 4096,
+  "output_throughput_tok_s": 12.61,
+  "input_throughput_tok_s": 3227.8,
+  "ttft_s": {
+   "mean": 112.9849,
+   "p50": 119.5745,
+   "max": 156.4269,
+   "min": 38.9673
+  },
+  "tpot_s": {
+   "mean": 0.0659,
+   "p50": 0.0817,
+   "max": 0.1632,
+   "min": 0.0091
+  },
+  "e2e_s": {
+   "mean": 146.6423,
+   "p50": 161.3281,
+   "max": 201.7932,
+   "min": 80.7246
+  },
+  "per_req_out_tok_s_e2e": {
+   "mean": 3.9136,
+   "p50": 3.1898,
+   "max": 6.3426,
+   "min": 2.5373
+  },
+  "per_req_decode_tok_s": {
+   "mean": 43.9614,
+   "p50": 12.2625,
+   "max": 110.4456,
+   "min": 6.1379
+  },
+  "spec_accept_length_mean": 4.257,
+  "retractions_total": 0,
+  "cache_hit_from_logs": {
+   "prefill_batches": 128,
+   "new_tokens": 1048576,
+   "cached_tokens": 0,
+   "hit_rate": 0.0
+  }
+ },
+ "arm1_Amirror_warmup": {
+  "concurrency": 1,
+  "num_requests": 1,
+  "run_id": 9991,
+  "corpus_window": {
+   "start": 17000000,
+   "end": 17131072
+  },
+  "ok": 1,
+  "failed": 0,
+  "wall_s": 52.85,
+  "input_len": 131072,
+  "shared_len": 0,
+  "unique_len": 131072,
+  "output_len": 64,
+  "output_tokens_total": 64,
+  "output_throughput_tok_s": 1.21,
+  "input_throughput_tok_s": 2480.16,
+  "ttft_s": {
+   "mean": 51.3163,
+   "p50": 51.3163,
+   "max": 51.3163,
+   "min": 51.3163
+  },
+  "tpot_s": {
+   "mean": 0.0243,
+   "p50": 0.0243,
+   "max": 0.0243,
+   "min": 0.0243
+  },
+  "e2e_s": {
+   "mean": 52.8471,
+   "p50": 52.8471,
+   "max": 52.8471,
+   "min": 52.8471
+  },
+  "per_req_out_tok_s_e2e": {
+   "mean": 1.211,
+   "p50": 1.211,
+   "max": 1.211,
+   "min": 1.211
+  },
+  "per_req_decode_tok_s": {
+   "mean": 41.8164,
+   "p50": 41.8164,
+   "max": 41.8164,
+   "min": 41.8164
+  },
+  "spec_accept_length_mean": 2.065,
+  "retractions_total": 0,
+  "cache_hit_from_logs": {
+   "prefill_batches": 16,
+   "new_tokens": 131072,
+   "cached_tokens": 0,
+   "hit_rate": 0.0
+  }
+ },
+ "arm2_Bp_cc1": {
+  "concurrency": 1,
+  "num_requests": 8,
+  "run_id": 9501,
+  "corpus_window": {
+   "start": 10000000,
+   "end": 11048576
+  },
+  "ok": 8,
+  "failed": 0,
+  "wall_s": 280.57,
+  "input_len": 131072,
+  "shared_len": 0,
+  "unique_len": 131072,
+  "output_len": 512,
+  "output_tokens_total": 4096,
+  "output_throughput_tok_s": 14.6,
+  "input_throughput_tok_s": 3737.27,
+  "ttft_s": {
+   "mean": 21.0555,
+   "p50": 21.0457,
+   "max": 21.1298,
+   "min": 21.0325
+  },
+  "tpot_s": {
+   "mean": 0.0274,
+   "p50": 0.0273,
+   "max": 0.0282,
+   "min": 0.0271
+  },
+  "e2e_s": {
+   "mean": 35.0713,
+   "p50": 35.0062,
+   "max": 35.544,
+   "min": 34.8595
+  },
+  "per_req_out_tok_s_e2e": {
+   "mean": 14.5994,
+   "p50": 14.6312,
+   "max": 14.6875,
+   "min": 14.4047
+  },
+  "per_req_decode_tok_s": {
+   "mean": 36.5374,
+   "p50": 36.704,
+   "max": 37.0298,
+   "min": 35.5211
+  },
+  "spec_accept_length_mean": null,
+  "retractions_total": 0,
+  "cache_hit_from_logs": {
+   "prefill_batches": 256,
+   "new_tokens": 2097152,
+   "cached_tokens": 0,
+   "hit_rate": 0.0
+  }
+ },
+ "arm2_Bp_cc2": {
+  "concurrency": 2,
+  "num_requests": 8,
+  "run_id": 9502,
+  "corpus_window": {
+   "start": 11048576,
+   "end": 12097152
+  },
+  "ok": 8,
+  "failed": 0,
+  "wall_s": 226.99,
+  "input_len": 131072,
+  "shared_len": 0,
+  "unique_len": 131072,
+  "output_len": 512,
+  "output_tokens_total": 4096,
+  "output_throughput_tok_s": 18.04,
+  "input_throughput_tok_s": 4619.49,
+  "ttft_s": {
+   "mean": 31.2138,
+   "p50": 41.3247,
+   "max": 41.3388,
+   "min": 21.0898
+  },
+  "tpot_s": {
+   "mean": 0.05,
+   "p50": 0.0695,
+   "max": 0.0701,
+   "min": 0.0298
+  },
+  "e2e_s": {
+   "mean": 56.7421,
+   "p50": 56.7513,
+   "max": 56.9058,
+   "min": 56.5568
+  },
+  "per_req_out_tok_s_e2e": {
+   "mean": 9.0233,
+   "p50": 9.0273,
+   "max": 9.0529,
+   "min": 8.9973
+  },
+  "per_req_decode_tok_s": {
+   "mean": 23.8132,
+   "p50": 32.8787,
+   "max": 33.6254,
+   "min": 14.2991
+  },
+  "spec_accept_length_mean": null,
+  "retractions_total": 0,
+  "cache_hit_from_logs": {
+   "prefill_batches": 256,
+   "new_tokens": 2097152,
+   "cached_tokens": 0,
+   "hit_rate": 0.0
+  }
+ },
+ "arm2_Bp_cc3": {
+  "concurrency": 3,
+  "num_requests": 8,
+  "run_id": 9503,
+  "corpus_window": {
+   "start": 12097152,
+   "end": 13145728
+  },
+  "ok": 8,
+  "failed": 0,
+  "wall_s": 216.51,
+  "input_len": 131072,
+  "shared_len": 0,
+  "unique_len": 131072,
+  "output_len": 512,
+  "output_tokens_total": 4096,
+  "output_throughput_tok_s": 18.92,
+  "input_throughput_tok_s": 4843.07,
+  "ttft_s": {
+   "mean": 38.8062,
+   "p50": 41.3294,
+   "max": 61.5934,
+   "min": 21.0703
+  },
+  "tpot_s": {
+   "mean": 0.0691,
+   "p50": 0.0755,
+   "max": 0.1156,
+   "min": 0.0294
+  },
+  "e2e_s": {
+   "mean": 74.1346,
+   "p50": 79.9715,
+   "max": 80.2009,
+   "min": 56.3659
+  },
+  "per_req_out_tok_s_e2e": {
+   "mean": 7.0671,
+   "p50": 6.4036,
+   "max": 9.0835,
+   "min": 6.384
+  },
+  "per_req_decode_tok_s": {
+   "mean": 18.4756,
+   "p50": 14.4943,
+   "max": 34.0308,
+   "min": 8.6643
+  },
+  "spec_accept_length_mean": null,
+  "retractions_total": 0,
+  "cache_hit_from_logs": {
+   "prefill_batches": 256,
+   "new_tokens": 2097152,
+   "cached_tokens": 0,
+   "hit_rate": 0.0
+  }
+ },
+ "arm2_Bp_cc4": {
+  "concurrency": 4,
+  "num_requests": 8,
+  "run_id": 9504,
+  "corpus_window": {
+   "start": 13145728,
+   "end": 14194304
+  },
+  "ok": 8,
+  "failed": 0,
+  "wall_s": 199.49,
+  "input_len": 131072,
+  "shared_len": 0,
+  "unique_len": 131072,
+  "output_len": 512,
+  "output_tokens_total": 4096,
+  "output_throughput_tok_s": 20.53,
+  "input_throughput_tok_s": 5256.26,
+  "ttft_s": {
+   "mean": 51.477,
+   "p50": 61.5611,
+   "max": 81.8537,
+   "min": 21.0928
+  },
+  "tpot_s": {
+   "mean": 0.0944,
+   "p50": 0.1142,
+   "max": 0.1539,
+   "min": 0.0349
+  },
+  "e2e_s": {
+   "mean": 99.7182,
+   "p50": 99.7397,
+   "max": 99.7733,
+   "min": 99.6468
+  },
+  "per_req_out_tok_s_e2e": {
+   "mean": 5.1345,
+   "p50": 5.1347,
+   "max": 5.1381,
+   "min": 5.1316
+  },
+  "per_req_decode_tok_s": {
+   "mean": 14.3454,
+   "p50": 13.4435,
+   "max": 28.7267,
+   "min": 6.5103
+  },
+  "spec_accept_length_mean": null,
+  "retractions_total": 0,
+  "cache_hit_from_logs": {
+   "prefill_batches": 256,
+   "new_tokens": 2097152,
+   "cached_tokens": 0,
+   "hit_rate": 0.0
+  }
+ },
+ "arm2_Bp_warmup": {
+  "concurrency": 1,
+  "num_requests": 1,
+  "run_id": 9991,
+  "corpus_window": {
+   "start": 17000000,
+   "end": 17131072
+  },
+  "ok": 1,
+  "failed": 0,
+  "wall_s": 35.22,
+  "input_len": 131072,
+  "shared_len": 0,
+  "unique_len": 131072,
+  "output_len": 64,
+  "output_tokens_total": 64,
+  "output_throughput_tok_s": 1.82,
+  "input_throughput_tok_s": 3721.64,
+  "ttft_s": {
+   "mean": 32.9797,
+   "p50": 32.9797,
+   "max": 32.9797,
+   "min": 32.9797
+  },
+  "tpot_s": {
+   "mean": 0.0355,
+   "p50": 0.0355,
+   "max": 0.0355,
+   "min": 0.0355
+  },
+  "e2e_s": {
+   "mean": 35.2179,
+   "p50": 35.2179,
+   "max": 35.2179,
+   "min": 35.2179
+  },
+  "per_req_out_tok_s_e2e": {
+   "mean": 1.8173,
+   "p50": 1.8173,
+   "max": 1.8173,
+   "min": 1.8173
+  },
+  "per_req_decode_tok_s": {
+   "mean": 28.5963,
+   "p50": 28.5963,
+   "max": 28.5963,
+   "min": 28.5963
+  },
+  "spec_accept_length_mean": null,
+  "retractions_total": 0,
+  "cache_hit_from_logs": {
+   "prefill_batches": 32,
+   "new_tokens": 262144,
+   "cached_tokens": 0,
+   "hit_rate": 0.0
+  }
+ },
+ "arm3_TP2PP4_cc1": {
+  "concurrency": 1,
+  "num_requests": 8,
+  "run_id": 9501,
+  "corpus_window": {
+   "start": 10000000,
+   "end": 11048576
+  },
+  "ok": 8,
+  "failed": 0,
+  "wall_s": 321.42,
+  "input_len": 131072,
+  "shared_len": 0,
+  "unique_len": 131072,
+  "output_len": 512,
+  "output_tokens_total": 4096,
+  "output_throughput_tok_s": 12.74,
+  "input_throughput_tok_s": 3262.35,
+  "ttft_s": {
+   "mean": 15.4876,
+   "p50": 15.4928,
+   "max": 15.5353,
+   "min": 15.4459
+  },
+  "tpot_s": {
+   "mean": 0.0483,
+   "p50": 0.0484,
+   "max": 0.0486,
+   "min": 0.0481
+  },
+  "e2e_s": {
+   "mean": 40.1768,
+   "p50": 40.2302,
+   "max": 40.3235,
+   "min": 40.0206
+  },
+  "per_req_out_tok_s_e2e": {
+   "mean": 12.7438,
+   "p50": 12.7615,
+   "max": 12.7934,
+   "min": 12.6973
+  },
+  "per_req_decode_tok_s": {
+   "mean": 20.7383,
+   "p50": 20.8009,
+   "max": 20.8346,
+   "min": 20.6009
+  },
+  "spec_accept_length_mean": null,
+  "retractions_total": 0,
+  "cache_hit_from_logs": {
+   "prefill_batches": 512,
+   "new_tokens": 4194304,
+   "cached_tokens": 0,
+   "hit_rate": 0.0
+  }
+ },
+ "arm3_TP2PP4_cc2": {
+  "concurrency": 2,
+  "num_requests": 8,
+  "run_id": 9502,
+  "corpus_window": {
+   "start": 11048576,
+   "end": 12097152
+  },
+  "ok": 8,
+  "failed": 0,
+  "wall_s": 226.22,
+  "input_len": 131072,
+  "shared_len": 0,
+  "unique_len": 131072,
+  "output_len": 512,
+  "output_tokens_total": 4096,
+  "output_throughput_tok_s": 18.11,
+  "input_throughput_tok_s": 4635.23,
+  "ttft_s": {
+   "mean": 22.3167,
+   "p50": 28.9744,
+   "max": 29.1154,
+   "min": 15.5008
+  },
+  "tpot_s": {
+   "mean": 0.067,
+   "p50": 0.08,
+   "max": 0.0808,
+   "min": 0.0535
+  },
+  "e2e_s": {
+   "mean": 56.5495,
+   "p50": 56.4636,
+   "max": 56.8831,
+   "min": 56.4112
+  },
+  "per_req_out_tok_s_e2e": {
+   "mean": 9.0541,
+   "p50": 9.0686,
+   "max": 9.0762,
+   "min": 9.0009
+  },
+  "per_req_decode_tok_s": {
+   "mean": 15.566,
+   "p50": 18.4753,
+   "max": 18.7252,
+   "min": 12.3965
+  },
+  "spec_accept_length_mean": null,
+  "retractions_total": 0,
+  "cache_hit_from_logs": {
+   "prefill_batches": 512,
+   "new_tokens": 4194304,
+   "cached_tokens": 0,
+   "hit_rate": 0.0
+  }
+ },
+ "arm3_TP2PP4_cc3": {
+  "concurrency": 3,
+  "num_requests": 8,
+  "run_id": 9503,
+  "corpus_window": {
+   "start": 12097152,
+   "end": 13145728
+  },
+  "ok": 8,
+  "failed": 0,
+  "wall_s": 209.0,
+  "input_len": 131072,
+  "shared_len": 0,
+  "unique_len": 131072,
+  "output_len": 512,
+  "output_tokens_total": 4096,
+  "output_throughput_tok_s": 19.6,
+  "input_throughput_tok_s": 5017.09,
+  "ttft_s": {
+   "mean": 27.8189,
+   "p50": 29.6062,
+   "max": 43.4436,
+   "min": 15.6499
+  },
+  "tpot_s": {
+   "mean": 0.085,
+   "p50": 0.09,
+   "max": 0.1191,
+   "min": 0.0533
+  },
+  "e2e_s": {
+   "mean": 71.2601,
+   "p50": 75.7564,
+   "max": 76.4978,
+   "min": 56.8297
+  },
+  "per_req_out_tok_s_e2e": {
+   "mean": 7.3001,
+   "p50": 6.7662,
+   "max": 9.0094,
+   "min": 6.693
+  },
+  "per_req_decode_tok_s": {
+   "mean": 12.6863,
+   "p50": 12.4942,
+   "max": 18.8075,
+   "min": 8.4145
+  },
+  "spec_accept_length_mean": null,
+  "retractions_total": 0,
+  "cache_hit_from_logs": {
+   "prefill_batches": 512,
+   "new_tokens": 4194304,
+   "cached_tokens": 0,
+   "hit_rate": 0.0
+  }
+ },
+ "arm3_TP2PP4_cc4": {
+  "concurrency": 4,
+  "num_requests": 8,
+  "run_id": 9504,
+  "corpus_window": {
+   "start": 13145728,
+   "end": 14194304
+  },
+  "ok": 8,
+  "failed": 0,
+  "wall_s": 177.53,
+  "input_len": 131072,
+  "shared_len": 0,
+  "unique_len": 131072,
+  "output_len": 512,
+  "output_tokens_total": 4096,
+  "output_throughput_tok_s": 23.07,
+  "input_throughput_tok_s": 5906.56,
+  "ttft_s": {
+   "mean": 36.6103,
+   "p50": 43.4223,
+   "max": 57.2876,
+   "min": 15.9191
+  },
+  "tpot_s": {
+   "mean": 0.102,
+   "p50": 0.1154,
+   "max": 0.1425,
+   "min": 0.0615
+  },
+  "e2e_s": {
+   "mean": 88.7369,
+   "p50": 88.7669,
+   "max": 88.8538,
+   "min": 88.5975
+  },
+  "per_req_out_tok_s_e2e": {
+   "mean": 5.7699,
+   "p50": 5.7685,
+   "max": 5.7789,
+   "min": 5.7623
+  },
+  "per_req_decode_tok_s": {
+   "mean": 10.8261,
+   "p50": 11.3182,
+   "max": 16.279,
+   "min": 7.0293
+  },
+  "spec_accept_length_mean": null,
+  "retractions_total": 0,
+  "cache_hit_from_logs": {
+   "prefill_batches": 512,
+   "new_tokens": 4194304,
+   "cached_tokens": 0,
+   "hit_rate": 0.0
+  }
+ },
+ "arm3_TP2PP4_cc5": {
+  "concurrency": 5,
+  "num_requests": 8,
+  "run_id": 9505,
+  "corpus_window": {
+   "start": 10000000,
+   "end": 11048576
+  },
+  "ok": 8,
+  "failed": 0,
+  "wall_s": 182.46,
+  "input_len": 131072,
+  "shared_len": 0,
+  "unique_len": 131072,
+  "output_len": 512,
+  "output_tokens_total": 4096,
+  "output_throughput_tok_s": 22.45,
+  "input_throughput_tok_s": 5746.99,
+  "ttft_s": {
+   "mean": 37.5353,
+   "p50": 42.5635,
+   "max": 69.4847,
+   "min": 15.615
+  },
+  "tpot_s": {
+   "mean": 0.1127,
+   "p50": 0.117,
+   "max": 0.1789,
+   "min": 0.0641
+  },
+  "e2e_s": {
+   "mean": 95.1469,
+   "p50": 106.9463,
+   "max": 107.052,
+   "min": 75.3993
+  },
+  "per_req_out_tok_s_e2e": {
+   "mean": 5.5367,
+   "p50": 4.7905,
+   "max": 6.7905,
+   "min": 4.7827
+  },
+  "per_req_decode_tok_s": {
+   "mean": 9.8884,
+   "p50": 10.0708,
+   "max": 15.6197,
+   "min": 5.6013
+  },
+  "spec_accept_length_mean": null,
+  "retractions_total": 0,
+  "cache_hit_from_logs": {
+   "prefill_batches": 512,
+   "new_tokens": 4194304,
+   "cached_tokens": 0,
+   "hit_rate": 0.0
+  }
+ },
+ "arm3_TP2PP4_cc6": {
+  "concurrency": 6,
+  "num_requests": 8,
+  "run_id": 9506,
+  "corpus_window": {
+   "start": 11048576,
+   "end": 12097152
+  },
+  "ok": 8,
+  "failed": 0,
+  "wall_s": 179.26,
+  "input_len": 131072,
+  "shared_len": 0,
+  "unique_len": 131072,
+  "output_len": 512,
+  "output_tokens_total": 4096,
+  "output_throughput_tok_s": 22.85,
+  "input_throughput_tok_s": 5849.32,
+  "ttft_s": {
+   "mean": 42.7136,
+   "p50": 42.8146,
+   "max": 83.2404,
+   "min": 15.5862
+  },
+  "tpot_s": {
+   "mean": 0.1242,
+   "p50": 0.13,
+   "max": 0.2096,
+   "min": 0.0536
+  },
+  "e2e_s": {
+   "mean": 106.1613,
+   "p50": 122.6846,
+   "max": 122.8274,
+   "min": 56.4675
+  },
+  "per_req_out_tok_s_e2e": {
+   "mean": 5.3952,
+   "p50": 4.1733,
+   "max": 9.0672,
+   "min": 4.1684
+  },
+  "per_req_decode_tok_s": {
+   "mean": 9.7806,
+   "p50": 9.6772,
+   "max": 18.6901,
+   "min": 4.7795
+  },
+  "spec_accept_length_mean": null,
+  "retractions_total": 0,
+  "cache_hit_from_logs": {
+   "prefill_batches": 512,
+   "new_tokens": 4194304,
+   "cached_tokens": 0,
+   "hit_rate": 0.0
+  }
+ },
+ "arm3_TP2PP4_cc7": {
+  "concurrency": 7,
+  "num_requests": 8,
+  "run_id": 9507,
+  "corpus_window": {
+   "start": 12097152,
+   "end": 13145728
+  },
+  "ok": 8,
+  "failed": 0,
+  "wall_s": 179.89,
+  "input_len": 131072,
+  "shared_len": 0,
+  "unique_len": 131072,
+  "output_len": 512,
+  "output_tokens_total": 4096,
+  "output_throughput_tok_s": 22.77,
+  "input_throughput_tok_s": 5829.09,
+  "ttft_s": {
+   "mean": 58.4235,
+   "p50": 56.691,
+   "max": 138.7352,
+   "min": 15.683
+  },
+  "tpot_s": {
+   "mean": 0.1241,
+   "p50": 0.1298,
+   "max": 0.2103,
+   "min": 0.0532
+  },
+  "e2e_s": {
+   "mean": 121.849,
+   "p50": 123.0753,
+   "max": 179.69,
+   "min": 56.7662
+  },
+  "per_req_out_tok_s_e2e": {
+   "mean": 4.6041,
+   "p50": 4.1611,
+   "max": 9.0195,
+   "min": 2.8494
+  },
+  "per_req_decode_tok_s": {
+   "mean": 9.8211,
+   "p50": 9.7386,
+   "max": 18.8168,
+   "min": 4.7644
+  },
+  "spec_accept_length_mean": null,
+  "retractions_total": 0,
+  "cache_hit_from_logs": {
+   "prefill_batches": 512,
+   "new_tokens": 4194304,
+   "cached_tokens": 0,
+   "hit_rate": 0.0
+  }
+ },
+ "arm3_TP2PP4_hit90_cc4": {
+  "concurrency": 4,
+  "num_requests": 8,
+  "run_id": 9604,
+  "corpus_window": {
+   "start": 15000000,
+   "end": 15104832
+  },
+  "ok": 8,
+  "failed": 0,
+  "wall_s": 59.49,
+  "input_len": 131072,
+  "shared_len": 117968,
+  "unique_len": 13104,
+  "output_len": 512,
+  "output_tokens_total": 4096,
+  "output_throughput_tok_s": 68.85,
+  "input_throughput_tok_s": 17624.84,
+  "ttft_s": {
+   "mean": 6.9268,
+   "p50": 7.2908,
+   "max": 8.986,
+   "min": 4.5732
+  },
+  "tpot_s": {
+   "mean": 0.0446,
+   "p50": 0.0451,
+   "max": 0.0489,
+   "min": 0.041
+  },
+  "e2e_s": {
+   "mean": 29.7055,
+   "p50": 29.9081,
+   "max": 30.0152,
+   "min": 29.4371
+  },
+  "per_req_out_tok_s_e2e": {
+   "mean": 17.237,
+   "p50": 17.3599,
+   "max": 17.393,
+   "min": 17.058
+  },
+  "per_req_decode_tok_s": {
+   "mean": 22.5696,
+   "p50": 23.1193,
+   "max": 24.449,
+   "min": 20.4936
+  },
+  "spec_accept_length_mean": null,
+  "retractions_total": 0,
+  "cache_hit_from_logs": {
+   "prefill_batches": 56,
+   "new_tokens": 419840,
+   "cached_tokens": 3774464,
+   "hit_rate": 0.8999
+  }
+ },
+ "arm3_TP2PP4_single512k": {
+  "concurrency": 1,
+  "num_requests": 1,
+  "run_id": 9601,
+  "corpus_window": {
+   "start": 10000000,
+   "end": 10523776
+  },
+  "ok": 1,
+  "failed": 0,
+  "wall_s": 114.38,
+  "input_len": 523776,
+  "shared_len": 0,
+  "unique_len": 523776,
+  "output_len": 512,
+  "output_tokens_total": 512,
+  "output_throughput_tok_s": 4.48,
+  "input_throughput_tok_s": 4579.23,
+  "ttft_s": {
+   "mean": 88.3831,
+   "p50": 88.3831,
+   "max": 88.3831,
+   "min": 88.3831
+  },
+  "tpot_s": {
+   "mean": 0.0509,
+   "p50": 0.0509,
+   "max": 0.0509,
+   "min": 0.0509
+  },
+  "e2e_s": {
+   "mean": 114.3797,
+   "p50": 114.3797,
+   "max": 114.3797,
+   "min": 114.3797
+  },
+  "per_req_out_tok_s_e2e": {
+   "mean": 4.4763,
+   "p50": 4.4763,
+   "max": 4.4763,
+   "min": 4.4763
+  },
+  "per_req_decode_tok_s": {
+   "mean": 19.6952,
+   "p50": 19.6952,
+   "max": 19.6952,
+   "min": 19.6952
+  },
+  "spec_accept_length_mean": null,
+  "retractions_total": 0,
+  "cache_hit_from_logs": {
+   "prefill_batches": 256,
+   "new_tokens": 2095104,
+   "cached_tokens": 0,
+   "hit_rate": 0.0
+  }
+ },
+ "arm3_TP2PP4_single900k": {
+  "concurrency": 1,
+  "num_requests": 1,
+  "run_id": 9602,
+  "corpus_window": {
+   "start": 10000000,
+   "end": 10900000
+  },
+  "ok": 1,
+  "failed": 0,
+  "wall_s": 237.77,
+  "input_len": 900000,
+  "shared_len": 0,
+  "unique_len": 900000,
+  "output_len": 512,
+  "output_tokens_total": 512,
+  "output_throughput_tok_s": 2.15,
+  "input_throughput_tok_s": 3785.22,
+  "ttft_s": {
+   "mean": 210.587,
+   "p50": 210.587,
+   "max": 210.587,
+   "min": 210.587
+  },
+  "tpot_s": {
+   "mean": 0.0532,
+   "p50": 0.0532,
+   "max": 0.0532,
+   "min": 0.0532
+  },
+  "e2e_s": {
+   "mean": 237.7662,
+   "p50": 237.7662,
+   "max": 237.7662,
+   "min": 237.7662
+  },
+  "per_req_out_tok_s_e2e": {
+   "mean": 2.1534,
+   "p50": 2.1534,
+   "max": 2.1534,
+   "min": 2.1534
+  },
+  "per_req_decode_tok_s": {
+   "mean": 18.8381,
+   "p50": 18.8381,
+   "max": 18.8381,
+   "min": 18.8381
+  },
+  "spec_accept_length_mean": null,
+  "retractions_total": 0,
+  "cache_hit_from_logs": {
+   "prefill_batches": 440,
+   "new_tokens": 3600128,
+   "cached_tokens": 0,
+   "hit_rate": 0.0
+  }
+ },
+ "arm3_TP2PP4_warmup": {
+  "concurrency": 1,
+  "num_requests": 1,
+  "run_id": 9991,
+  "corpus_window": {
+   "start": 17000000,
+   "end": 17131072
+  },
+  "ok": 1,
+  "failed": 0,
+  "wall_s": 31.74,
+  "input_len": 131072,
+  "shared_len": 0,
+  "unique_len": 131072,
+  "output_len": 64,
+  "output_tokens_total": 64,
+  "output_throughput_tok_s": 2.02,
+  "input_throughput_tok_s": 4129.93,
+  "ttft_s": {
+   "mean": 28.361,
+   "p50": 28.361,
+   "max": 28.361,
+   "min": 28.361
+  },
+  "tpot_s": {
+   "mean": 0.0536,
+   "p50": 0.0536,
+   "max": 0.0536,
+   "min": 0.0536
+  },
+  "e2e_s": {
+   "mean": 31.736,
+   "p50": 31.736,
+   "max": 31.736,
+   "min": 31.736
+  },
+  "per_req_out_tok_s_e2e": {
+   "mean": 2.0166,
+   "p50": 2.0166,
+   "max": 2.0166,
+   "min": 2.0166
+  },
+  "per_req_decode_tok_s": {
+   "mean": 18.9644,
+   "p50": 18.9644,
+   "max": 18.9644,
+   "min": 18.9644
+  },
+  "spec_accept_length_mean": null,
+  "retractions_total": 0,
+  "cache_hit_from_logs": {
+   "prefill_batches": 64,
+   "new_tokens": 524288,
+   "cached_tokens": 0,
+   "hit_rate": 0.0
+  }
+ }
+}
diff --git a/experiments/pro6000/glm53_nvfp4_128k_capacity_topology/results/20260909/md5s.txt b/experiments/pro6000/glm53_nvfp4_128k_capacity_topology/results/20260909/md5s.txt
new file mode 100644
index 0000000..51192f7
--- /dev/null
+++ b/experiments/pro6000/glm53_nvfp4_128k_capacity_topology/results/20260909/md5s.txt
@@ -0,0 +1,3 @@
+e782a8767d4478711757669f83adb64b *scripts/arm_runner.sh
+f6f5a49937ac6e8a0157ba1eb11b8ee7 *scripts/deploy_glm53_605_v2.sh
+8eb75693f01f6055b15ecc3d62feff08 *scripts/val_runner.sh
diff --git a/experiments/pro6000/glm53_nvfp4_128k_capacity_topology/scripts/arm_runner.sh b/experiments/pro6000/glm53_nvfp4_128k_capacity_topology/scripts/arm_runner.sh
new file mode 100644
index 0000000..08040ed
--- /dev/null
+++ b/experiments/pro6000/glm53_nvfp4_128k_capacity_topology/scripts/arm_runner.sh
@@ -0,0 +1,36 @@
+#!/bin/bash
+# 128k capacity experiment - per-arm bench runner (i131072/o512, cc 1-4, nreq 8)
+# usage: nohup bash arm_runner.sh <arm-name> >/dev/null 2>&1 &
+# per-cc-point corpus windows are distinct but the mapping is identical for every arm:
+#   cc1 base 10000000, cc2 11048576, cc3 12097152, cc4 13145728 (all in consumed space, flushed)
+ARM=${1:?arm name required}
+URL=http://127.0.0.1:30000
+LOGD=/root/bench_logs/128kexp
+mkdir -p $LOGD
+echo "$(date +%T) [$ARM] start" >> $LOGD/${ARM}_progress.log
+
+# wait until server idle (no running/queued requests in recent logs)
+for i in $(seq 1 60); do
+  R=$(docker logs --since 20s glm53-nvfp4 2>&1 | grep -cE 'running-req: [1-9]|queue-req: [1-9]')
+  if [ "$R" -eq 0 ]; then break; fi
+  sleep 10
+done
+
+flush() { curl -s -X POST $URL/flush_cache >/dev/null; sleep 2; }
+
+# shape warmup (not scored): one 128k prefill to absorb kernel first-touch cost
+flush
+python3 /root/bench_corpus.py --input-len 131072 --output-len 64 --shared-frac 0 \
+  --concurrency 1 --num-requests 1 --run-id 9991 --pool-override 17000000 \
+  > $LOGD/${ARM}_warmup.log 2>&1
+echo "$(date +%T) [$ARM] warmup done" >> $LOGD/${ARM}_progress.log
+
+declare -A BASE=( [1]=10000000 [2]=11048576 [3]=12097152 [4]=13145728 )
+for CC in 1 2 3 4; do
+  flush
+  python3 /root/bench_corpus.py --input-len 131072 --output-len 512 --shared-frac 0 \
+    --concurrency $CC --num-requests 8 --run-id 950${CC} --pool-override ${BASE[$CC]} \
+    > $LOGD/${ARM}_cc${CC}.log 2>&1
+  echo "$(date +%T) [$ARM] cc$CC done rc=$?" >> $LOGD/${ARM}_progress.log
+done
+echo "ALL-DONE" >> $LOGD/${ARM}_progress.log
diff --git a/experiments/pro6000/glm53_nvfp4_128k_capacity_topology/scripts/deploy_glm53_605_v2.sh b/experiments/pro6000/glm53_nvfp4_128k_capacity_topology/scripts/deploy_glm53_605_v2.sh
new file mode 100644
index 0000000..e428fd5
--- /dev/null
+++ b/experiments/pro6000/glm53_nvfp4_128k_capacity_topology/scripts/deploy_glm53_605_v2.sh
@@ -0,0 +1,129 @@
+#!/bin/bash
+# deploy_glm53_605_v2.sh — GLM-5.3-NVFP4 60.5 生产容量扩容配置（2026-09-09 交付）
+#
+# 目的：解决 60.5 现役 TP8+EAGLE 配置的 KV 池容量问题（池 276,864 token）：
+#   - 单条请求上限 264k（--context-length 270336 所致，模型原生支持 1M）
+#   - 并发>1 且总 token>~26 万时排队（i128k 下 c3 起必然排队）
+# 本配置 = 2026-09-09 在 60.8 同型机（8×RTX 6000D）四臂对拍优胜者：
+#   TP2 PP4 nomtp + radix/hicache 开 + ctx 1M + cu13 栈 9 挂载
+#   实测池 909,632 token（3.29×）：i128k 并发上限 c2→c6，单条上限 264k→909k，
+#   c4 输入/输出吞吐 5907/23.1 tok/s（现役口径 3228/12.6 的 1.83×/1.83×），
+#   质量门 7/7。完整数据：experiments/pro6000/glm53_nvfp4_128k_capacity_topology/
+#
+# ⚠️ 运维提示：
+#   1) 在维护窗口执行：会移除现役 glm53-nvfp4 容器，停机约 10-15 分钟（权重加载+健康）。
+#   2) 回滚 = 原脚本原样留盘：bash /root/deploy_glm53_605.sh（TP8+EAGLE 原配置）。
+#   3) 与 60.8 现役（本配置已留役）一致，脚本可互相对拍 md5。
+#
+# 用法: bash deploy_glm53_605_v2.sh          # 部署到 30000 端口
+#       ROLLBACK_ONLY=1 bash deploy_glm53_605_v2.sh   # 仅回滚到原配置
+
+set -u
+IMAGE=lmsysorg/sglang:nightly-dev-cu13-20260901-07c8f729
+PORT=30000
+
+# ---- 前置检查（缺什么给什么清单，不猜） ----
+MISSING=0
+if [ -z "$(docker images -q $IMAGE 2>/dev/null)" ]; then
+  echo "[MISS] 镜像不存在: $IMAGE"
+  echo "       拉取: docker pull $IMAGE"
+  echo "       或从 60.8 导: ssh 60.8 'docker save $IMAGE | gzip' | gunzip | docker load"
+  MISSING=1
+fi
+MOUNTS=(
+  "/root/sglang_patch2/layer_setup.py"
+  "/root/sglang_patch2/validation_hook.py"
+  "/root/eagle_worker_v2_mask.py"
+  "/root/sglang_patch2/eagle_worker_common.py"
+  "/root/sglang_patch2/deepseek_nextn.py"
+  "/root/scheduler_pp_mixin_r35.py"
+  "/root/request_receiver_degloo.py"
+  "/root/sglang_patch2/deepseek_v2.py"
+  "/root/decode_cuda_graph_runner_fix.py"
+)
+for f in "${MOUNTS[@]}"; do
+  if [ ! -f "$f" ]; then
+    echo "[MISS] 补丁文件缺失: $f"
+    MISSING=1
+  fi
+done
+if [ "$MISSING" = "1" ]; then
+  echo "[STOP] 上述文件在 60.8 /root 均有（含 sglang_patch2/ 子目录），scp 全部补齐后重跑本脚本。"
+  exit 1
+fi
+
+if [ "${ROLLBACK_ONLY:-0}" = "1" ]; then
+  echo "[rollback] 恢复 60.5 原生产配置（TP8+EAGLE）..."
+  exec bash /root/deploy_glm53_605.sh
+fi
+
+# ---- 移除旧容器（等待显存归零，纪律：重部署前必等显存<1500MiB） ----
+docker update --restart=no glm53-nvfp4 >/dev/null 2>&1
+for i in 1 2 3 4 5 6 7 8; do
+  docker rm -f glm53-nvfp4 >/dev/null 2>&1
+  docker ps -a --format '{{.Names}}' | grep -q '^glm53-nvfp4$' || break
+  sleep 5
+done
+if docker ps -a --format '{{.Names}}' | grep -q '^glm53-nvfp4$'; then
+  echo "[ERROR] 旧容器删不掉（zombie 时重试即可）"; exit 1
+fi
+echo "[wait] 等待显存释放 (<1500MiB/卡)..."
+for i in $(seq 1 60); do
+  MAXMI=$(nvidia-smi --query-gpu=memory.used --format=csv,noheader,nounits | sort -rn | head -1)
+  [ "$MAXMI" -lt 1500 ] && break
+  sleep 5
+done
+echo "[wait] 显存就绪 (${MAXMI}MiB max)"
+
+# ---- 部署优胜配置（与 60.8 实验臂 arm3_TP2PP4 完全一致） ----
+docker run -d --name glm53-nvfp4 --gpus all --shm-size 64g --ipc=host --cap-add SYS_PTRACE \
+  -v /root/sglang_patch2/layer_setup.py:/sgl-workspace/sglang/python/sglang/srt/model_executor/model_runner_components/layer_setup.py:ro \
+  -v /root/sglang_patch2/validation_hook.py:/sgl-workspace/sglang/python/sglang/srt/arg_groups/validation_hook.py:ro \
+  -v /root/eagle_worker_v2_mask.py:/sgl-workspace/sglang/python/sglang/srt/speculative/eagle_worker_v2.py:ro \
+  -v /root/sglang_patch2/eagle_worker_common.py:/sgl-workspace/sglang/python/sglang/srt/speculative/eagle_worker_common.py:ro \
+  -v /root/sglang_patch2/deepseek_nextn.py:/sgl-workspace/sglang/python/sglang/srt/models/deepseek_nextn.py:ro \
+  -v /root/scheduler_pp_mixin_r35.py:/sgl-workspace/sglang/python/sglang/srt/managers/scheduler_pp_mixin.py:ro \
+  -v /root/request_receiver_degloo.py:/sgl-workspace/sglang/python/sglang/srt/managers/scheduler_components/request_receiver.py:ro \
+  -v /root/sglang_patch2/deepseek_v2.py:/sgl-workspace/sglang/python/sglang/srt/models/deepseek_v2.py:ro \
+  -v /root/decode_cuda_graph_runner_fix.py:/sgl-workspace/sglang/python/sglang/srt/model_executor/runner/decode_cuda_graph_runner.py:ro \
+  -e SGLANG_PP_DEGLOO=1 -e SGLANG_PP_SPEC_FORCE_EAGER_DRAFT=1 -e SGLANG_PP_FORCE_EAGER_VERIFY=0 -e SGLANG_PP_SPEC_DEBUG=0 \
+  --restart no -p ${PORT}:${PORT} \
+  -v /data/hf_models:/data/hf_models \
+  $IMAGE \
+  python3 -m sglang.launch_server \
+    --model-path /data/hf_models/GLM-5.3-NVFP4 \
+    --tp 8 \
+    --mem-fraction-static 0.85 \
+    --max-running-requests 16 \
+    --chunked-prefill-size 8192 \
+    --disable-shared-experts-fusion \
+    --moe-runner-backend flashinfer_cutlass \
+    --disable-flashinfer-autotune \
+    --reasoning-parser glm45 --tool-call-parser glm47 \
+    --enable-hierarchical-cache --hicache-ratio 3 \
+    --tp 2 --pp-size 4 --disable-overlap-schedule --max-prefill-tokens 16384 \
+    --disable-custom-all-reduce --context-length 1048576 \
+    --host 0.0.0.0 --port ${PORT}
+
+echo "[deploy] TP2PP4-nomtp 容量扩容配置启动，等待健康（最长 1800s）..."
+for i in $(seq 10 10 1800); do
+  code=$(curl -s -o /dev/null -m3 -w '%{http_code}' http://127.0.0.1:${PORT}/health 2>/dev/null)
+  if [ "$code" = "200" ]; then
+    echo "[OK] healthy after ${i}s"
+    docker update --restart=unless-stopped glm53-nvfp4 >/dev/null && echo "[OK] restart=unless-stopped 已设置"
+    POOL=$(docker logs glm53-nvfp4 2>&1 | grep -m1 -oE 'max_total_num_tokens=[0-9]+')
+    echo "[INFO] ${POOL}  (60.8 同型机实测 909632；若显著低于此值请停下核查)"
+    echo "[NEXT] 质量门: bash /root/quality_gate_605.sh   期望 PASS=7 FAIL=0"
+    echo "[NEXT] 容量冒烟: 300k 单条请求应能正常完成（旧配置会直接拒绝）"
+    exit 0
+  fi
+  if ! docker ps --format '{{.Names}}' | grep -q '^glm53-nvfp4$'; then
+    echo "[DIED] 容器启动失败，最近错误："
+    docker logs --tail 40 glm53-nvfp4 2>&1 | grep -iE 'error|assert|not support' | tail -8
+    echo "[回滚] bash /root/deploy_glm53_605.sh"
+    exit 1
+  fi
+  sleep 10
+done
+echo "[TIMEOUT] 健康等待超时；回滚: bash /root/deploy_glm53_605.sh"
+exit 1
diff --git a/experiments/pro6000/glm53_nvfp4_128k_capacity_topology/scripts/val_runner.sh b/experiments/pro6000/glm53_nvfp4_128k_capacity_topology/scripts/val_runner.sh
new file mode 100644
index 0000000..67ef397
--- /dev/null
+++ b/experiments/pro6000/glm53_nvfp4_128k_capacity_topology/scripts/val_runner.sh
@@ -0,0 +1,48 @@
+#!/bin/bash
+# winner validation runner (TP2PP4): big singles, cc boundary, 90%-hit point
+# usage: nohup bash val_runner.sh <arm-name> >/dev/null 2>&1 &
+ARM=${1:?arm name required}
+URL=http://127.0.0.1:30000
+LOGD=/root/bench_logs/128kexp
+echo "$(date +%T) [$ARM] val start" >> $LOGD/${ARM}_progress.log
+
+for i in $(seq 1 60); do
+  R=$(docker logs --since 20s glm53-nvfp4 2>&1 | grep -cE 'running-req: [1-9]|queue-req: [1-9]')
+  if [ "$R" -eq 0 ]; then break; fi
+  sleep 10
+done
+
+flush() { curl -s -X POST $URL/flush_cache >/dev/null; sleep 2; }
+
+# 1) 512k single (real corpus): 523776+512 = 524288
+flush
+python3 /root/bench_corpus.py --input-len 523776 --output-len 512 --shared-frac 0 \
+  --concurrency 1 --num-requests 1 --run-id 9601 --pool-override 10000000 \
+  > $LOGD/${ARM}_single512k.log 2>&1
+echo "$(date +%T) [$ARM] single512k rc=$?" >> $LOGD/${ARM}_progress.log
+
+# 2) 900k single (pool ceiling probe): 900000+512 = 900512 vs pool 909632
+flush
+python3 /root/bench_corpus.py --input-len 900000 --output-len 512 --shared-frac 0 \
+  --concurrency 1 --num-requests 1 --run-id 9602 --pool-override 10000000 \
+  > $LOGD/${ARM}_single900k.log 2>&1
+echo "$(date +%T) [$ARM] single900k rc=$?" >> $LOGD/${ARM}_progress.log
+
+# 3-5) cc boundary: 5 / 6 fit, 7 exceeds pool (921088 > 909632) -> expect queuing
+declare -A BASEB=( [5]=10000000 [6]=11048576 [7]=12097152 )
+for CC in 5 6 7; do
+  flush
+  python3 /root/bench_corpus.py --input-len 131072 --output-len 512 --shared-frac 0 \
+    --concurrency $CC --num-requests 8 --run-id 950${CC} --pool-override ${BASEB[$CC]} \
+    > $LOGD/${ARM}_cc${CC}.log 2>&1
+  echo "$(date +%T) [$ARM] cc$CC rc=$?" >> $LOGD/${ARM}_progress.log
+done
+
+# 6) 90%-hit cc4 point (mirrors 60.5 real traffic: shared 117968 + unique 13104x8)
+flush
+python3 /root/bench_corpus.py --input-len 131072 --output-len 512 --shared-frac 0.9 \
+  --concurrency 4 --num-requests 8 --run-id 9604 --pool-override 15000000 \
+  > $LOGD/${ARM}_hit90_cc4.log 2>&1
+echo "$(date +%T) [$ARM] hit90_cc4 rc=$?" >> $LOGD/${ARM}_progress.log
+
+echo "VAL-ALL-DONE" >> $LOGD/${ARM}_progress.log