[Klaud Cold] Update MiniMax-M3 FP4 GB300 Dynamo vLLM AgentX disagg image to v0.32.0 / [Klaud Cold] 将 MiniMax-M3 FP4 GB300 Dynamo vLLM AgentX 分离式镜像更新至 v0.32.0 by Klaud-Cold · Pull Request #4616 · SemiAnalysisAI/InferenceX

Klaud-Cold · 2026-10-28T16:19:22Z

Goal: Update vLLM image from vllm/vllm-openai:nightly-5e35a6f4f9bbc217c599692157ca985c894373f7 to vllm/vllm-openai:v0.32.0@sha256:befccf0e7e4089aa23926bb6f887a85c0d46c2a99e6de5c5ffdf5cb7957c5b34.
Baseline: 2026-08-21 · vllm/vllm-openai:nightly-5e35a6f4f9bbc217c599692157ca985c894373f7
Mean latency · Sources: API 1, API 2, API 3, API 4

Point	Total tok/s/GPU ↑	Output tok/s/GPU ↑	TTFT ms ↓	TPOT ms ↓
AgentX c1 PTP4x1/DTP4x1 b6fdfc	2,821.03	19.22	419.42	2.85
AgentX c20 PTP2x1/DTP4x1 553930	23,913.77	151.32	504.92	4.11
AgentX c24 PTP1x1/DTP4x3 8d4b12	14,070.32	83.58	779.31	3.27
AgentX c24 PTP2x1/DTP4x1 720f96	29,109.65	176.1	913.2	4.56
AgentX c48 PTP2x1/DTP2x3 b9ce69	38,508.87	254.63	1,770.09	5.67
AgentX c120 PTP2x2/DTP2x5 4463b4	43,977.29	346.08	2,529.72	7.55

Eval	Score ↑	Samples
gsm8k/em_strict · c24	96.97%	1,319
gsm8k/em_strict · c24	97.12%	1,319
gsm8k/em_strict · c48	97.27%	1,319
gsm8k/em_strict · c120	97.19%	1,319

Stop: Close this PR at any time to stop this candidate now; Klaud cancels its runs and deletes its branch.

中文

**目标：**将 vLLM 镜像从 vllm/vllm-openai:nightly-5e35a6f4f9bbc217c599692157ca985c894373f7 更新为 vllm/vllm-openai:v0.32.0@sha256:befccf0e7e4089aa23926bb6f887a85c0d46c2a99e6de5c5ffdf5cb7957c5b34。
**基线：**2026-08-21 · vllm/vllm-openai:nightly-5e35a6f4f9bbc217c599692157ca985c894373f7
平均延迟 · 来源： API 1, API 2, API 3, API 4；数值及异常说明见上表。
**停止：**随时关闭此 PR 即可立即停止该候选：Klaud 会取消其运行并删除其分支。

… v0.32.0 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

github-actions · 2026-10-28T16:19:33Z

Thanks for the contribution!

Review: If this PR changes files owned by someone other than a repository admin or @SemiAnalysisAI/core, ask one eligible CODEOWNER to complete the latest PR_REVIEW_CHECKLIST.md before contacting a core maintainer on Slack. Follow the template exactly, including As a PR reviewer and CODEOWNER, I have reviewed this and have, so sign-off verification triggers.
PR verification: Sweeps run only when the PR appends an inferencex-e2e/perf-changelog.yaml entry and carries exactly one primary label: full-sweep-fail-fast (strongly recommended; canary plus per-matrix fail-fast), full-sweep-enabled (canary; matrix jobs continue after a failure), or non-canary-full-sweep-enabled (no canary or fail-fast). The modifiers all-evals, evals-only, and agentx-fast require a primary label. On fork PRs, a maintainer applies the label. See sweep labels and reuse.
After merging: PR authors must ensure all GitHub Actions jobs pass. Transient failures often pass on rerun; see how to rerun failed jobs.

中文

感谢你的贡献！

**审阅：**如果 PR 修改的文件归属于仓库管理员及 @SemiAnalysisAI/core 之外的 CODEOWNER，请先联系一位有资格的 CODEOWNER 填写最新的 PR_REVIEW_CHECKLIST.md，再通过 Slack 联系核心维护者。必须严格遵循模板，并保留 As a PR reviewer and CODEOWNER, I have reviewed this and have，才能触发签核验证。
**PR 验证：**只有当 PR 在 inferencex-e2e/perf-changelog.yaml 末尾追加了条目，并且恰好带有一个主标签时，才会运行扫描：full-sweep-fail-fast（强烈推荐；canary 加逐矩阵 fail-fast）、full-sweep-enabled（有 canary；矩阵任务在失败后继续运行）或 non-canary-full-sweep-enabled（无 canary，也无 fail-fast）。修饰标签 all-evals、evals-only 和 agentx-fast 必须与主标签一起使用。fork PR 的标签由维护者添加。参见扫描标签与复用。
**合并后：**PR 作者必须确保所有 GitHub Actions 任务通过。临时性失败通常可以通过重新运行恢复；参见重新运行失败任务的说明。

Klaud-Cold · 2026-10-28T16:21:58Z

Initial attempt · Failed · Run 40724891253 / attempt 1 · 2026-10-28 16:49 UTC
vllm/vllm-openai:v0.32.0@sha256:befccf0e7e4089aa23926bb6f887a85c0d46c2a99e6de5c5ffdf5cb7957c5b34 · a19dd45d0386 · Mean latency
Change: Measured ref 8d49630 sets the master image to vLLM v0.32.0 by digest with recipe flags unchanged (vllm-project/vllm 5e35a6f4 versus v0.32.0 5df63a72 source keeps every used option); the c1 prefill worker failed during CUDA graph memory profiling when the EAGLE3 drafter's FLASH_ATTN backend selected the FA4 dense kernel and to_cute_tensor rejected the stride-0 FP8 KV descale tensors created by layer._k_scale.expand (RuntimeError: Expected strides[leading_dim] == 1, but got 0; source: vllm v0.32.0 vllm/v1/attention/backends/flash_attn.py L1290-1291 and L1399, vllm_flash_attn/cute/interface.py L1437), and fail-fast cancelled the other 4 benchmarks and 9 evals.

Point	Output tok/s/GPU ↑	TTFT ms ↓	TPOT ms ↓
AgentX c1 PTP4x1/DTP4x1 b6fdfc	N/A	N/A	N/A
AgentX c24 PTP1x1/DTP4x3 8d4b12	N/A	N/A	N/A
AgentX c20 PTP2x1/DTP4x1 553930	N/A	N/A	N/A
AgentX c48 PTP2x1/DTP2x3 b9ce69	N/A	N/A	N/A
AgentX c120 PTP2x2/DTP2x5 4463b4	N/A	N/A	N/A

Note: AgentX c1 PTP4x1/DTP4x1 b6fdfc: failed.
Note: All rows: request errors unavailable.
Note: All rows: Δ N/A: point unavailable.
Note: AgentX c24 PTP1x1/DTP4x3 8d4b12, AgentX c20 PTP2x1/DTP4x1 553930, AgentX c48 PTP2x1/DTP2x3 b9ce69, AgentX c120 PTP2x2/DTP2x5 4463b4: cancelled.

Next: Repair 1: switch only the EAGLE3 drafter speculative-config attention_backend from FLASH_ATTN to FLASHINFER (FP8 KV supported, as the sibling MiniMax-M3 vLLM B200 FP4 MTP recipe uses) and rerun the smoke.

中文

初次尝试 · 失败 · Run 40724891253 / attempt 1 · 2026-10-28 16:49 UTC
vllm/vllm-openai:v0.32.0@sha256:befccf0e7e4089aa23926bb6f887a85c0d46c2a99e6de5c5ffdf5cb7957c5b34 · a19dd45d0386 · 平均延迟
**变更：**测量 ref 8d49630 将主配置镜像设为按 digest 固定的 vLLM v0.32.0，配方参数不变（对比 vllm-project/vllm 5e35a6f4 与 v0.32.0 5df63a72 源码，所用选项均保留）；c1 的 prefill worker 在 CUDA graph 内存分析阶段失败：EAGLE3 草稿模型的 FLASH_ATTN 后端选择了 FA4 dense kernel，而 to_cute_tensor 拒绝了 layer._k_scale.expand 生成的 stride 为 0 的 FP8 KV descale 张量（RuntimeError: Expected strides[leading_dim] == 1, but got 0；来源：vllm v0.32.0 vllm/v1/attention/backends/flash_attn.py L1290-1291 与 L1399、vllm_flash_attn/cute/interface.py L1437），fail-fast 取消了其余 4 个 benchmark 和 9 个 eval。实测数值及异常说明见上表。
**下一步：**修复 1：仅将 EAGLE3 草稿模型 speculative-config 的 attention_backend 从 FLASH_ATTN 改为 FLASHINFER（支持 FP8 KV，与同系 MiniMax-M3 vLLM B200 FP4 MTP 配方一致），然后重新运行 smoke。

Klaud-Cold · 2026-10-28T16:22:28Z

parked · Session 1 is waiting on 40724891253. Recovery reruns preempted runs and starts the next session once they end.

中文

parked · 第 1 个会话正在等待 40724891253。恢复流程会重跑被抢占的运行，并在运行结束后启动下一个会话。

Klaud-Cold · 2026-10-28T16:45:48Z

resumed · The session ended before finishing; session 2 continues it in a new run of its planner.

中文

resumed · 会话未完成即已结束；第 2 个会话将在其规划运行的新运行中继续。

…LLM v0.32.0 vLLM v0.32.0 routes FLASH_ATTN to the FA4 dense kernel on SM100, which rejects the stride-0 FP8 KV descale tensors. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Klaud-Cold · 2026-10-28T16:51:08Z

Repair 1/5 · Failed · Run 40729765879 / attempt 1 · 2026-10-28 17:51 UTC
vllm/vllm-openai:v0.32.0@sha256:befccf0e7e4089aa23926bb6f887a85c0d46c2a99e6de5c5ffdf5cb7957c5b34 · a19dd45d0386 · Mean latency
Change: Measured ref bac80d5 (EAGLE3 drafter attention_backend FLASHINFER) cleared the FA4 failure: engines loaded and captured graphs, but the c48 prefill worker (and the c48 eval and c120 prefill) then exited when Dynamo 1.4.0.dev20260730 parsed the node-inherited DYN_TCP_RPC_HOST=enP22s22f0np0 interface name as a socket address (Exception: Invalid TCP bind address: invalid socket address syntax; source: ai-dynamo/dynamo a3725995 lib/runtime/src/pipeline/network/manager.rs L270-272), and fail-fast cancelled the rest.

Point	Output tok/s/GPU ↑	TTFT ms ↓	TPOT ms ↓
AgentX c1 PTP4x1/DTP4x1 b6fdfc	N/A	N/A	N/A
AgentX c24 PTP1x1/DTP4x3 8d4b12	N/A	N/A	N/A
AgentX c20 PTP2x1/DTP4x1 553930	N/A	N/A	N/A
AgentX c48 PTP2x1/DTP2x3 b9ce69	N/A	N/A	N/A
AgentX c120 PTP2x2/DTP2x5 4463b4	N/A	N/A	N/A

Note: AgentX c1 PTP4x1/DTP4x1 b6fdfc, AgentX c24 PTP1x1/DTP4x3 8d4b12, AgentX c20 PTP2x1/DTP4x1 553930, AgentX c120 PTP2x2/DTP2x5 4463b4: cancelled.
Note: All rows: request errors unavailable.
Note: All rows: Δ N/A: point unavailable.
Note: AgentX c48 PTP2x1/DTP2x3 b9ce69: failed.

Next: Repair 2: bump only the family recipe's Dynamo pin to 1.6.0.dev20261010, which resolves interface names since ai-dynamo/dynamo#11870 and runs green in gb300-nv dsv41flash Dynamo vLLM recipes, then rerun the smoke.

中文

修复 1/5 · 失败 · Run 40729765879 / attempt 1 · 2026-10-28 17:51 UTC
vllm/vllm-openai:v0.32.0@sha256:befccf0e7e4089aa23926bb6f887a85c0d46c2a99e6de5c5ffdf5cb7957c5b34 · a19dd45d0386 · 平均延迟
**变更：**测量 ref bac80d5（EAGLE3 草稿模型 attention_backend 改为 FLASHINFER）消除了 FA4 失败：引擎已加载并完成 graph 捕获，但随后 c48 prefill worker（以及 c48 eval 和 c120 prefill）退出，原因是 Dynamo 1.4.0.dev20260730 将节点继承的接口名 DYN_TCP_RPC_HOST=enP22s22f0np0 当作 socket 地址解析（Exception: Invalid TCP bind address: invalid socket address syntax；来源：ai-dynamo/dynamo a3725995 lib/runtime/src/pipeline/network/manager.rs L270-272），fail-fast 取消了其余任务。实测数值及异常说明见上表。
**下一步：**修复 2：仅将该系列配方的 Dynamo 版本固定为 1.6.0.dev20261010（自 ai-dynamo/dynamo#11870 起支持解析接口名，且在 gb300-nv dsv41flash Dynamo vLLM 配方中运行通过），然后重新运行 smoke。

Klaud-Cold · 2026-10-28T16:51:22Z

parked · Session 2 is waiting on 40729765879. Recovery reruns preempted runs and starts the next session once they end.

中文

parked · 第 2 个会话正在等待 40729765879。恢复流程会重跑被抢占的运行，并在运行结束后启动下一个会话。

Klaud-Cold · 2026-10-28T17:39:36Z

resumed · The session ended before finishing; session 3 continues it in a new run of its planner.

中文

resumed · 会话未完成即已结束；第 3 个会话将在其规划运行的新运行中继续。

…disagg Dynamo 1.4.0.dev20260730 parses DYN_TCP_RPC_HOST as a raw socket address, so the gb300-nv node value enP22s22f0np0 (an interface name) fails with 'Invalid TCP bind address'. Dynamo resolves interface names since ai-dynamo/dynamo#11870 (2026-09-05); 1.6.0.dev20261010 matches the green gb300-nv dsv41flash Dynamo vLLM recipes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Klaud-Cold · 2026-10-28T17:51:36Z

Repair 2/5 · Failed · Run 40739495555 / attempt 1 · 2026-10-28 18:12 UTC
vllm/vllm-openai:v0.32.0@sha256:befccf0e7e4089aa23926bb6f887a85c0d46c2a99e6de5c5ffdf5cb7957c5b34 · a19dd45d0386 · Mean latency
Change: Measured ref b5f1ca3 (FLASHINFER drafter, recipe Dynamo pin 1.6.0.dev20261010) failed before engine start: the c24 prefill worker and c20 eval exited at import because Dynamo 1.6.0.dev20261010 dynamo/vllm/embedding_worker_processes.py L32 imports get_engine_zmq_addresses, which vLLM v0.32.0 vllm/v1/engine/utils.py lacks (ImportError; v0.32.0 offers get_engine_zmq_bind_addresses), and fail-fast cancelled the rest.

Point	Output tok/s/GPU ↑	TTFT ms ↓	TPOT ms ↓
AgentX c1 PTP4x1/DTP4x1 b6fdfc	N/A	N/A	N/A
AgentX c24 PTP1x1/DTP4x3 8d4b12	N/A	N/A	N/A
AgentX c20 PTP2x1/DTP4x1 553930	N/A	N/A	N/A
AgentX c48 PTP2x1/DTP2x3 b9ce69	N/A	N/A	N/A
AgentX c120 PTP2x2/DTP2x5 4463b4	N/A	N/A	N/A

Note: AgentX c1 PTP4x1/DTP4x1 b6fdfc, AgentX c20 PTP2x1/DTP4x1 553930, AgentX c48 PTP2x1/DTP2x3 b9ce69, AgentX c120 PTP2x2/DTP2x5 4463b4: cancelled.
Note: All rows: request errors unavailable.
Note: All rows: Δ N/A: point unavailable.
Note: AgentX c24 PTP1x1/DTP4x3 8d4b12: failed.

Next: Repair 3: pin the recipe Dynamo to 1.7.0.dev20261025, the first nightly after ai-dynamo/dynamo#16757 bumped core to vLLM v0.32.0 (wheel requires vllm==0.32.0 and resolves the helper lazily per #16378), then rerun the smoke.

中文

修复 2/5 · 失败 · Run 40739495555 / attempt 1 · 2026-10-28 18:12 UTC
vllm/vllm-openai:v0.32.0@sha256:befccf0e7e4089aa23926bb6f887a85c0d46c2a99e6de5c5ffdf5cb7957c5b34 · a19dd45d0386 · 平均延迟
**变更：**测量 ref b5f1ca3（FLASHINFER 草稿模型，配方 Dynamo 固定为 1.6.0.dev20261010）在引擎启动前失败：c24 prefill worker 与 c20 eval 在导入时退出，原因是 Dynamo 1.6.0.dev20261010 的 dynamo/vllm/embedding_worker_processes.py L32 导入 get_engine_zmq_addresses，而 vLLM v0.32.0 的 vllm/v1/engine/utils.py 中没有该函数（ImportError；v0.32.0 提供的是 get_engine_zmq_bind_addresses），fail-fast 取消了其余任务。实测数值及异常说明见上表。
**下一步：**修复 3：将配方 Dynamo 固定为 1.7.0.dev20261025，这是 ai-dynamo/dynamo#16757 将核心升级到 vLLM v0.32.0 之后的首个 nightly（wheel 依赖 vllm==0.32.0，并按 #16378 延迟解析该辅助函数），然后重新运行 smoke。

Klaud-Cold · 2026-10-28T17:51:57Z

parked · Session 3 is waiting on 40739495555. Recovery reruns preempted runs and starts the next session once they end.

中文

parked · 第 3 个会话正在等待 40739495555。恢复流程会重跑被抢占的运行，并在运行结束后启动下一个会话。

Klaud-Cold · 2026-10-28T18:08:33Z

resumed · The session ended before finishing; session 4 continues it in a new run of its planner.

中文

resumed · 会话未完成即已结束；第 4 个会话将在其规划运行的新运行中继续。

…v0.32.0 Dynamo 1.6.0.dev20261010 imports get_engine_zmq_addresses, which vLLM v0.32.0 lacks; 1.7.0.dev20261025 follows ai-dynamo/dynamo#16757 (vLLM v0.32.0 core) and resolves the helper lazily (#16378). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Klaud-Cold · 2026-10-28T18:13:15Z

Repair 3/5 · Failed · Run 40742966902 / attempt 1 · 2026-10-28 18:50 UTC
vllm/vllm-openai:v0.32.0@sha256:befccf0e7e4089aa23926bb6f887a85c0d46c2a99e6de5c5ffdf5cb7957c5b34 · a19dd45d0386 · Mean latency
Change: Measured ref 1bb2867 (FLASHINFER drafter, recipe Dynamo 1.7.0.dev20261025) cleared the Dynamo import error and all engines loaded and captured graphs, but the 4-rank prefill workers of the c24 DPA benchmark (im-gb300-r03-c001) and c1 eval (im-gb300-r03-c002) were host OOM-killed while Mooncake 0.3.13.post1 registered the unchanged 4x200 GiB embedded segments (node AnonPages reached ~800 of 957 GiB); PR #4557's passing c24 DPA job ran the same Mooncake build and segment layout on im-gb300-r01-c010, so this looks node-dependent.

Point	Output tok/s/GPU ↑	TTFT ms ↓	TPOT ms ↓
AgentX c1 PTP4x1/DTP4x1 b6fdfc	N/A	N/A	N/A
AgentX c24 PTP1x1/DTP4x3 8d4b12	N/A	N/A	N/A
AgentX c20 PTP2x1/DTP4x1 553930	N/A	N/A	N/A
AgentX c48 PTP2x1/DTP2x3 b9ce69	N/A	N/A	N/A
AgentX c120 PTP2x2/DTP2x5 4463b4	N/A	N/A	N/A

Note: AgentX c1 PTP4x1/DTP4x1 b6fdfc, AgentX c20 PTP2x1/DTP4x1 553930, AgentX c48 PTP2x1/DTP2x3 b9ce69, AgentX c120 PTP2x2/DTP2x5 4463b4: cancelled.
Note: All rows: request errors unavailable.
Note: All rows: Δ N/A: point unavailable.
Note: AgentX c24 PTP1x1/DTP4x3 8d4b12: failed.

Next: Infrastructure retry 1: after a 45-minute back-off, rerun the failed and fail-fast-cancelled jobs of run 40742966902 on the same head with no recipe change; if the 4-rank prefill host OOM repeats on r03 nodes, stop, because avoiding it would require shrinking the frozen Mooncake DRAM offload capacity.

中文

修复 3/5 · 失败 · Run 40742966902 / attempt 1 · 2026-10-28 18:50 UTC
vllm/vllm-openai:v0.32.0@sha256:befccf0e7e4089aa23926bb6f887a85c0d46c2a99e6de5c5ffdf5cb7957c5b34 · a19dd45d0386 · 平均延迟
**变更：**测量 ref 1bb2867（FLASHINFER 草稿模型，配方 Dynamo 1.7.0.dev20261025）消除了 Dynamo 导入错误，所有引擎均已加载并完成 graph 捕获；但 c24 DPA 基准（im-gb300-r03-c001）与 c1 eval（im-gb300-r03-c002）的 4-rank prefill worker 在 Mooncake 0.3.13.post1 注册未改动的 4x200 GiB embedded 段时被主机 OOM 终止（节点 AnonPages 达到约 800/957 GiB）；PR #4557 中通过的 c24 DPA 任务在 im-gb300-r01-c010 上使用相同的 Mooncake 版本与段布局，因此该问题看起来与节点相关。实测数值及异常说明见上表。
**下一步：**基础设施重试 1：等待 45 分钟后，在同一 head 上重跑 run 40742966902 中失败及被 fail-fast 取消的任务，不修改配方；若 4-rank prefill 主机 OOM 在 r03 节点上再次出现则停止，因为规避它需要缩小已冻结的 Mooncake DRAM offload 容量。

Klaud-Cold · 2026-10-28T18:14:26Z

parked · Session 4 is waiting on 40742966902. Recovery reruns preempted runs and starts the next session once they end.

中文

parked · 第 4 个会话正在等待 40742966902。恢复流程会重跑被抢占的运行，并在运行结束后启动下一个会话。

Klaud-Cold · 2026-10-28T18:42:46Z

resumed · The session ended before finishing; session 5 continues it in a new run of its planner.

中文

resumed · 会话未完成即已结束；第 5 个会话将在其规划运行的新运行中继续。

Klaud-Cold · 2026-10-28T18:50:38Z

parked · Session 5 is waiting for recovery to rerun the failed jobs of 40742966902 after 2026-10-28 19:35 UTC. Recovery starts the next session once they end.

中文

parked · 第 5 个会话正在等待恢复流程在 2026-10-28 19:35 UTC 后重跑 40742966902 的失败作业。运行结束后，恢复流程将启动下一个会话。

chore(configs): bump MiniMax-M3 FP4 GB300 Dynamo vLLM disagg image to…

8d49630

… v0.32.0 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

github-project-automation Bot added this to InferenceX Board Oct 28, 2026

github-actions Bot added the klaud-parked Klaud parked until its queued or running benchmark runs end label Oct 28, 2026

github-actions Bot added klaud-in-progress Klaud is working on this candidate or will resume it and removed klaud-parked Klaud parked until its queued or running benchmark runs end labels Oct 28, 2026

fix(recipes): use FLASHINFER for MiniMax-M3 GB300 EAGLE3 drafter on v…

bac80d5

…LLM v0.32.0 vLLM v0.32.0 routes FLASH_ATTN to the FA4 dense kernel on SM100, which rejects the stride-0 FP8 KV descale tensors. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

github-actions Bot added klaud-parked Klaud parked until its queued or running benchmark runs end and removed klaud-in-progress Klaud is working on this candidate or will resume it labels Oct 28, 2026

github-actions Bot added klaud-in-progress Klaud is working on this candidate or will resume it and removed klaud-parked Klaud parked until its queued or running benchmark runs end labels Oct 28, 2026

github-actions Bot added klaud-parked Klaud parked until its queued or running benchmark runs end and removed klaud-in-progress Klaud is working on this candidate or will resume it labels Oct 28, 2026

github-actions Bot added klaud-in-progress Klaud is working on this candidate or will resume it and removed klaud-parked Klaud parked until its queued or running benchmark runs end labels Oct 28, 2026

github-actions Bot added klaud-parked Klaud parked until its queued or running benchmark runs end and removed klaud-in-progress Klaud is working on this candidate or will resume it labels Oct 28, 2026

github-actions Bot added klaud-in-progress Klaud is working on this candidate or will resume it and removed klaud-parked Klaud parked until its queued or running benchmark runs end labels Oct 28, 2026

github-actions Bot added klaud-parked Klaud parked until its queued or running benchmark runs end and removed klaud-in-progress Klaud is working on this candidate or will resume it labels Oct 28, 2026

Conversation

Klaud-Cold commented Oct 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions Bot commented Oct 28, 2026

Uh oh!

Klaud-Cold commented Oct 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Klaud-Cold commented Oct 28, 2026

Uh oh!

Klaud-Cold commented Oct 28, 2026

Uh oh!

Klaud-Cold commented Oct 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Klaud-Cold commented Oct 28, 2026

Uh oh!

Klaud-Cold commented Oct 28, 2026

Uh oh!

Klaud-Cold commented Oct 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Klaud-Cold commented Oct 28, 2026

Uh oh!

Klaud-Cold commented Oct 28, 2026

Uh oh!

Klaud-Cold commented Oct 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Klaud-Cold commented Oct 28, 2026

Uh oh!

Klaud-Cold commented Oct 28, 2026

Uh oh!

Klaud-Cold commented Oct 28, 2026

Uh oh!

This branch has not been deployed

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Klaud-Cold commented Oct 28, 2026 •

edited

Loading

Klaud-Cold commented Oct 28, 2026 •

edited

Loading

Klaud-Cold commented Oct 28, 2026 •

edited

Loading

Klaud-Cold commented Oct 28, 2026 •

edited

Loading

Klaud-Cold commented Oct 28, 2026 •

edited

Loading