[WIP] Add GLM-5.2-NVFP4 B300 AgentX recipes by csahithi · Pull Request #2654 · SemiAnalysisAI/InferenceX

csahithi · 2026-08-18T17:28:43Z

B300 Dynamo-SGLang agentic-coding sweep for GLM-5.2-NVFP4, 7 points across two config keys:

glm5.2-fp4-b300-dynamo-sglang-agentic-agg: aggregated single-node TP8 (conc 2/4)
glm5.2-fp4-b300-dynamo-sglang-agentic-disagg: low-latency 1P2D/1P3D/1P4D/1P6D TP4-decode (conc 63/61/59/58) and throughput 2P1D DEP8-decode (conc 139)

Note

Low Risk
Changes are benchmark recipes, master config entries, changelog, and conditional CI launch logic—no production serving or security-sensitive application code.

Overview
Adds GLM-5.2-NVFP4 on B300 agentic-coding benchmarks as two new nvidia-master config keys, wired to new srt-slurm base + zip_override_mtp_agentx_lowlat recipes under benchmarks/multi_node/srt-slurm-recipes/sglang/glm5.2/b300-fp4/agentic/.

glm5.2-fp4-b300-dynamo-sglang-agentic-agg covers aggregated single-node TP8 at concurrency 2 and 4 (glm5.2-agentx-agg.yaml). glm5.2-fp4-b300-dynamo-sglang-agentic-disagg sweeps five disaggregated points: 1P2D–1P6D TP4 prefill/decode (conc 63/61/59/58) plus 2P1D DEP8 decode (conc 139) via glm5.2-agentx.yaml. Both use Dynamo router, EAGLE/MTP, HiCache, nixl KV transfer where disagg, and AiPerf inferencex-agentx-mvp against SemiAnalysis trace data with session affinity hooks.

runners/launch_b300-dsxe.sh gains a narrow branch for agentic glm5.2 fp4 dynamo-sglang: clone NVIDIA/srt-slurm at open PR #313 (93fae85…) instead of the usual pin, overlay only the b300-fp4 recipe subtree (not the full in-repo recipes tree), and run srtctl apply with UCX_TLS unset for glm5.2 dynamo-sglang. perf-changelog.yaml documents the seven-point sweep and image lmsysorg/sglang:nightly-dev-cu13-20260824-95f5ecd3 on b300-dsxe.

^{Reviewed by Cursor Bugbot for commit 6588cb2. Bugbot is set up for automated code reviews on this repo. Configure here.}

B300 Dynamo-SGLang agentic-coding sweep for GLM-5.2-NVFP4, 7 points across two config keys: - glm5.2-fp4-b300-dynamo-sglang-agentic-agg: aggregated single-node TP8 (conc 2/4) - glm5.2-fp4-b300-dynamo-sglang-agentic-disagg: low-latency 1P2D/1P3D/1P4D/1P6D TP4-decode (conc 63/61/59/58) and throughput 2P1D DEP8-decode (conc 139)

github-actions · 2026-08-18T17:28:53Z

Thanks for the contribution! Please reach out to respective companies' CODEOWNER to fill in the latest PR_REVIEW_CHECKLIST.md before pinging core maintainer on Slack for review. In order for the signoff PR check bot to trigger, you must follow the PR_REVIEW_CHECKLIST.md template correctly, including the phrase As a PR reviewer and CODEOWNER, I have reviewed this and have.

For PR verification, add the full-sweep-fail-fast label (strongly recommended) to this PR — the benchmark sweep only runs on labeled PRs. Use full-sweep-enabled only if you need matrix jobs to keep running past a failure.

PR authors are responsible for ensuring that after merging, all GitHub Action jobs fully pass. A lot of the time, failures are just flakes and simply re-running the failed jobs will fix it. See GitHub's docs on re-running failed jobs

感谢你的贡献！请联系相应公司的 CODEOWNER 填写最新的 PR_REVIEW_CHECKLIST.md，然后再在 Slack 上联系核心维护者进行审阅。为了触发 signoff PR 检查机器人，你必须正确遵循 PR_REVIEW_CHECKLIST.md 模板，包括保留英文语句 As a PR reviewer and CODEOWNER, I have reviewed this and have。

如需进行 PR 验证，请为此 PR 添加 full-sweep-fail-fast 标签（强烈推荐）— 基准测试 sweep 仅在带有标签的 PR 上运行。仅当需要矩阵任务在失败后继续运行时才使用 full-sweep-enabled。

PR 作者有责任确保合并后所有 GitHub Action 任务完全通过。 很多时候失败只是偶发抖动（flake），重新运行失败的任务即可解决。参见 GitHub 关于重新运行失败任务的文档

claude · 2026-08-18T17:43:14Z

    git checkout c180328b98c3793ca84a1e24a030f90545eb7d5d || exit 1
    mkdir -p recipes/sglang/deepseek-v4
    cp -rT "$GITHUB_WORKSPACE/benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4" recipes/sglang/deepseek-v4
+elif [[ $FRAMEWORK == "dynamo-sglang" && $MODEL_PREFIX == "glm5.2" && $PRECISION == "fp4" ]]; then
+    # GLM-5.2 B300 sglang AgentX: srt-slurm main carries the agentx-mvp scenario,
+    # session-affinity frontend, and custom benchmark schema these recipes need.
+    git clone https://github.com/NVIDIA/srt-slurm.git "$SRT_REPO_DIR"
+    cd "$SRT_REPO_DIR" || exit 1
+    git checkout main
+    mkdir -p recipes/sglang/glm5.2/b300-fp4
+    cp -rT "$GITHUB_WORKSPACE/benchmarks/multi_node/srt-slurm-recipes/sglang/glm5.2/b300-fp4" recipes/sglang/glm5.2/b300-fp4
 elif [[ $FRAMEWORK == "dynamo-vllm" && $MODEL_PREFIX == "minimaxm3" && $PRECISION == "fp4" && "$CONFIG_FILE" == recipes/vllm/minimax-m3/b300-fp4/8k1k/mtp/*.yaml ]]; then
    git clone --branch main --single-branch https://github.com/NVIDIA/srt-slurm.git "$SRT_REPO_DIR"
    cd "$SRT_REPO_DIR" || exit 1


🔴 The new glm5.2 elif branch in runners/launch_b300-nv.sh (lines 91-98) that clones NVIDIA/srt-slurm and stages the glm5.2 recipe files is unreachable dead code: the srt-slurm clone chain's very first branch is a bare if [[ "$IS_AGENTIC" == "1" ]]; then (line 76), and both new agentic-coding config keys set IS_AGENTIC=1, so bash always takes that catch-all instead. Fix by ANDing IS_AGENTIC==1 into the new glm5.2 condition (and moving the generic IS_AGENTIC catch-all after it), matching the ordering already used in launch_gb300-nv.sh and launch_gb200-nv.sh.

Extended reasoning...

The bug: runners/launch_b300-nv.sh builds its srt-slurm setup via an if/elif chain. The chain opens with:

if [[ "$IS_AGENTIC" == "1" ]]; then git clone --branch cam/sa-submission-q2-2026 --single-branch https://github.com/cquil11/srt-slurm-nv.git "$SRT_REPO_DIR" cd "$SRT_REPO_DIR" || exit 1 elif ... (several other elifs) ... elif [[ $FRAMEWORK == "dynamo-sglang" && $MODEL_PREFIX == "glm5.2" && $PRECISION == "fp4" ]]; then git clone https://github.com/NVIDIA/srt-slurm.git "$SRT_REPO_DIR" cd "$SRT_REPO_DIR" || exit 1 git checkout main mkdir -p recipes/sglang/glm5.2/b300-fp4 cp -rT "$GITHUB_WORKSPACE/benchmarks/multi_node/srt-slurm-recipes/sglang/glm5.2/b300-fp4" recipes/sglang/glm5.2/b300-fp4

The bare IS_AGENTIC check is first and does not exclude glm5.2 in any way, so in bash's if/elif chain it always wins whenever IS_AGENTIC == 1, regardless of $FRAMEWORK/$MODEL_PREFIX/$PRECISION. The new glm5.2 elif, added later in the same chain, can therefore never execute while IS_AGENTIC=1.

Why IS_AGENTIC is always 1 for these new recipes: configs/nvidia-master.yaml adds two new config keys, glm5.2-fp4-b300-dynamo-sglang-agentic-agg and glm5.2-fp4-b300-dynamo-sglang-agentic-disagg, both declaring scenarios: agentic-coding:. .github/workflows/benchmark-multinode-tmpl.yml sets IS_AGENTIC: ${{ inputs.scenario-type == 'agentic-coding' && '1' || '0' }}. So every job spawned from these two keys runs with IS_AGENTIC=1.

Step-by-step proof of failure:

A sweep point from glm5.2-fp4-b300-dynamo-sglang-agentic-agg launches with FRAMEWORK=dynamo-sglang, MODEL_PREFIX=glm5.2, PRECISION=fp4, IS_AGENTIC=1, and CONFIG_FILE=recipes/sglang/glm5.2/b300-fp4/agentic/glm5.2-agentx-agg.yaml.

launch_b300-nv.sh reaches the srt-slurm clone chain. The first condition [[ "$IS_AGENTIC" == "1" ]] is true, so bash takes that branch and never evaluates the later glm5.2 elif.

The cquil11/srt-slurm-nv fork (branch cam/sa-submission-q2-2026) is cloned instead of NVIDIA/srt-slurm main. No mkdir/cp -rT of any recipe happens in this branch.

Later in the script: CONFIG_PATH="${CONFIG_FILE%%:*}"; then if [[ ! -f "$CONFIG_PATH" ]]; then echo "Error: CONFIG_FILE does not exist after srt-slurm setup: $CONFIG_PATH"; exit 1; fi. Since recipes/sglang/glm5.2/b300-fp4/agentic/glm5.2-agentx-agg.yaml was never staged into the cloned repo, this check fails and the job exits 1 for every one of the 7 sweep points added by this PR.

Why nothing else catches it: there's no lint or CI step validating condition ordering in this bash chain; the failure only surfaces at runtime when the missing CONFIG_PATH check fires, deep into the job after the (slow) srt-slurm clone, srtctl install, and container import have already run.

The fix, proven by sibling launchers: runners/launch_gb300-nv.sh (and launch_gb200-nv.sh) solve this exact problem correctly — every model-specific agentic branch ANDs "$IS_AGENTIC" == "1" into its own condition (e.g. line 239: elif [[ "$IS_AGENTIC" == "1" && $FRAMEWORK == "dynamo-sglang" && $MODEL_PREFIX == "glm5.2" ]]), and the generic bare elif [[ "$IS_AGENTIC" == "1" ]] catch-all is placed LAST (line 249), after all specific cases. launch_b300-nv.sh should follow the same pattern: either AND $IS_AGENTIC == "1" into the new glm5.2 condition and move it before the generic catch-all, or move the generic if [[ "$IS_AGENTIC" == "1" ]] block to the end of the elif chain.

github-actions · 2026-08-18T17:44:37Z

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=32165814158
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=32165814158

The srt-slurm clone chain opens with a bare `if [[ "$IS_AGENTIC" == "1" ]]` catch-all, so the glm5.2 dynamo-sglang elif never ran for agentic jobs (all 7 GLM-5.2 B300 points hit the catch-all, cloned the wrong fork, and failed the CONFIG_PATH check). Promote the glm5.2 branch to the opening `if` with IS_AGENTIC==1 ANDed in and demote the generic catch-all to elif, matching the ordering in launch_gb300-nv.sh / launch_gb200-nv.sh.

github-actions · 2026-08-18T17:49:35Z

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=32167101624
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=32167101624

github-actions · 2026-08-18T18:00:57Z

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=32167548413
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=32167548413

GLM-5.2-NVFP4 is staged under compute-node-local /scratch/models, which the GHA login host running srtctl preflight cannot stat, so preflight fails with "path is unavailable" even though the workers can load it. Extend the existing dsv4 dynamo-sglang --no-preflight bypass to glm5.2 (runtime model loading still validates the path).

…entx # Conflicts: # perf-changelog.yaml

The disagg decode worker's NIXL KV manager crashed at buffer registration ("VRAM memory is detected as host by UCX ... VRAM registration cannot proceed" -> NIXL_ERR_BACKEND) because the recipe shipped no UCX config, so UCX had no CUDA memory transport. Add UCX_TLS=cuda_copy,rc + MC_FORCE_MNNVL + NCCL_MNNVL_ENABLE to prefill/decode env, matching the proven dsv4 B300 nixl disagg recipes.

github-actions · 2026-08-18T19:54:01Z

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=32174659553
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=32174659553

github-actions · 2026-08-18T23:09:25Z

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=32179040742
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=32179040742

Use the requested UCX_TLS override for both prefill and decode environments.

…entx # Conflicts: # perf-changelog.yaml

…entx # Conflicts: # configs/nvidia-master.yaml # perf-changelog.yaml

github-actions · 2026-08-19T22:12:31Z

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=32281705398
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=32281705398

github-actions · 2026-08-20T08:01:11Z

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=32307702676
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=32307702676

将 B300 AgentX 分离式 SGLang 侧的 prefill 与 decode 环境变量中固定的 UCX_TLS 显式设置替换为 UCX_LOG_LEVEL: debug，恢复由 UCX 自行选择传输通道，同时打开 UCX 侧调试日志。

github-actions · 2026-08-20T20:54:24Z

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=32410588562
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=32410588562

github-actions · 2026-08-20T22:40:37Z

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=32410588562
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=32410588562

github-actions · 2026-08-21T01:30:20Z

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=32435060137
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=32435060137

将 GLM-5.2 B300 AgentX 的两份 recipe 替换为上游 srt-slurm 版本（base + zip_override_mtp_agentx_lowlat），将 sglang 镜像升级到 nightly-dev-cu13-20260824-95f5ecd3，并把 nvidia-master.yaml 的 -agg 条目拆分为对应两个 zip_override 索引（c2 / c4）。

cursor · 2026-08-25T10:09:37Z

+    agg_nodes: 1
+    agg_workers: 1
+    gpus_per_decode: 8
+    gpus_per_prefill: 8


Aggregated recipe missing GPU field

High Severity

The aggregated zip override sets agg_nodes/agg_workers but uses gpus_per_decode and gpus_per_prefill instead of gpus_per_agg. Every other aggregated recipe in-repo (GB200/GB300 GLM, DeepSeek H200/GB300) uses gpus_per_agg. Aggregated worker GPU allocation is likely wrong or rejected.

^{Reviewed by Cursor Bugbot for commit e8c0a18. Configure here.}

cursor · 2026-08-25T10:09:37Z

+        enable-dp-attention: [false, false, false, false, true]
+        enable-dp-lm-head: [false, false, false, false, true]
+        max-running-requests: [16, 16, 16, 16, 139]
+        tensor-parallel-size: [4, 4, 4, 4, 8]


DEP8 decode missing expert parallel

High Severity

The throughput point enables decode data-parallel-size: 8 with enable-dp-attention: true but never overrides expert-parallel-size, which stays at base 1. Master config also labels that decode arm as ep: 1 while calling it DEP8. Sibling GLM DEP recipes keep ep equal to dp under dp-attention.

Additional Locations (1)

configs/nvidia-master.yaml#L8857-L8862

^{Reviewed by Cursor Bugbot for commit e8c0a18. Configure here.}

合并 origin/main，将 perf-changelog.yaml 解析为 main 的完整条目列表并把本分支的 PR #2654 条目附加在末尾。

cursor

Stale Bugbot comment from a previous run.

github-actions · 2026-08-25T10:51:00Z

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=32836132886
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=32836132886

修复 GLM-5.2 B300 AgentX 配方镜像解析。

github-actions · 2026-08-26T00:35:29Z

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=32912628075
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=32912628075

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

…entx-runner-update # Conflicts: # configs/nvidia-master.yaml # perf-changelog.yaml # runners/launch_b300-nv.sh

launch_b300-nv.sh was retired on main (#2826) in favor of the B300 DSXE cluster, so PR #2654's runner move to cluster:b300-dsxe needs the GLM-5.2 AgentX special-casing carried over: the pinned NVIDIA/srt-slurm#313 fork checkout (scoped to just the glm5.2/b300-fp4 recipes), and the UCX_TLS-unset wrapper around srtctl apply for dynamo-sglang. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

Master config now rejects disagg=false entries that use prefill/decode instead of a single worker: block (validated by infx.matrix.validation.MultiNodeMasterConfigEntry, which landed on main after this PR's branch point). Convert glm5.2-fp4-b300-dynamo-sglang-agentic-agg's two search-space points to worker:/num-nodes: to match, mirroring the sibling gb200-mtp-agg entry. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

cursor

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

There are 3 total unresolved issues (including 2 from previous reviews).

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

Want higher recall? High effort reviews run extra passes and find more bugs. A team admin can switch effort levels in the Cursor dashboard.

^{Reviewed by Cursor Bugbot for commit 6588cb2. Configure here.}

cursor · 2026-09-10T21:47:37Z

+              --ui simple \
+              --server-metrics-formats json \
+              --output-artifact-dir /logs/aiperf
+      rm -f /logs/aiperf/inputs.json


Recipes skip InferenceX agentic harness

High Severity

Both recipes invoke /aiperf-venv/bin/aiperf inline and write raw artifacts to /logs/aiperf instead of running agentic_srt.sh. The multinode workflow requires ${RESULT_FILENAME}_conc*.json files with num_requests_successful, which only the InferenceX harness writes. Every point in this sweep will fail result collection even if aiperf itself succeeds.

Additional Locations (1)

benchmarks/multi_node/srt-slurm-recipes/sglang/glm5.2/b300-fp4/agentic/glm5.2-agentx-agg.yaml#L95-L130

^{Reviewed by Cursor Bugbot for commit 6588cb2. Configure here.}

github-actions · 2026-09-11T01:25:35Z

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=34533625594
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=34533625594

csahithi requested a review from a team August 18, 2026 17:28

csahithi added the NVIDIA label Aug 18, 2026

csahithi requested review from Ankur-singh and kedarpotdar-nv as code owners August 18, 2026 17:28

csahithi added the full-sweep-enabled label Aug 18, 2026

github-project-automation Bot added this to InferenceMAX Board Aug 18, 2026

changelog: point GLM-5.2 B300 AgentX entry at PR #2654

fb7e871

claude Bot reviewed Aug 18, 2026

View reviewed changes

csahithi added 3 commits August 18, 2026 11:58

Merge remote-tracking branch 'origin/main' into nv-glm5.2-fp4-b300-ag…

5403dd9

…entx # Conflicts: # perf-changelog.yaml

csahithi added 3 commits August 19, 2026 10:16

fix: update B300 GLM-5.2 AgentX UCX transports

55209a8

Use the requested UCX_TLS override for both prefill and decode environments.

Merge remote-tracking branch 'origin/main' into nv-glm5.2-fp4-b300-ag…

d0aa47d

…entx # Conflicts: # perf-changelog.yaml

Merge remote-tracking branch 'origin/main' into nv-glm5.2-fp4-b300-ag…

1d4c73d

…entx # Conflicts: # configs/nvidia-master.yaml # perf-changelog.yaml

hshrivastava-droid added 2 commits August 20, 2026 12:34

chore(glm5.2): enable UCX debug logging on B300 AgentX disagg recipe

e0d3952

将 B300 AgentX 分离式 SGLang 侧的 prefill 与 decode 环境变量中固定的 UCX_TLS 显式设置替换为 UCX_LOG_LEVEL: debug，恢复由 UCX 自行选择传输通道，同时打开 UCX 侧调试日志。

Merge branch 'main' into nv-glm5.2-fp4-b300-agentx

6dc1fbd

fix(glm5.2): unset UCX_TLS for dynamo-sglang srtctl apply

6bd2424

cursor Bot reviewed Aug 25, 2026

View reviewed changes

Merge origin/main and resolve perf changelog

eb01708

合并 origin/main，将 perf-changelog.yaml 解析为 main 的完整条目列表并把本分支的 PR #2654 条目附加在末尾。

cursor Bot reviewed Aug 25, 2026

View reviewed changes

Comment thread benchmarks/multi_node/srt-slurm-recipes/sglang/glm5.2/b300-fp4/agentic/glm5.2-agentx.yaml Outdated

hshrivastava-droid added 2 commits August 25, 2026 16:45

fix(glm5.2): resolve B300 AgentX recipe image

a11f922

修复 GLM-5.2 B300 AgentX 配方镜像解析。

Merge branch 'main' into nv-glm5.2-fp4-b300-agentx

39da2c9

hshrivastava-droid and others added 4 commits September 10, 2026 14:33

feat(glm5.2): route B300 AgentX agg/disagg to cluster:b300-dsxe

6f5d148

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

Merge remote-tracking branch 'origin/main' into nv-glm5.2-fp4-b300-ag…

a220b2d

…entx-runner-update # Conflicts: # configs/nvidia-master.yaml # perf-changelog.yaml # runners/launch_b300-nv.sh

cursor Bot reviewed Sep 10, 2026

View reviewed changes

Uh oh!

Conversation

csahithi commented Aug 18, 2026 • edited by cursor Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions Bot commented Aug 18, 2026

Uh oh!

claude Bot Aug 18, 2026

Choose a reason for hiding this comment

Uh oh!

github-actions Bot commented Aug 18, 2026

Uh oh!

github-actions Bot commented Aug 18, 2026

Uh oh!

github-actions Bot commented Aug 18, 2026

Uh oh!

github-actions Bot commented Aug 18, 2026

Uh oh!

github-actions Bot commented Aug 18, 2026

Uh oh!

github-actions Bot commented Aug 19, 2026

Uh oh!

github-actions Bot commented Aug 20, 2026

Uh oh!

github-actions Bot commented Aug 20, 2026

Uh oh!

github-actions Bot commented Aug 20, 2026

Uh oh!

github-actions Bot commented Aug 21, 2026

Uh oh!

Uh oh!

cursor Bot Aug 25, 2026

Choose a reason for hiding this comment

Aggregated recipe missing GPU field

Uh oh!

cursor Bot Aug 25, 2026

Choose a reason for hiding this comment

DEP8 decode missing expert parallel

Uh oh!

cursor Bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

github-actions Bot commented Aug 25, 2026

Uh oh!

github-actions Bot commented Aug 26, 2026

Uh oh!

cursor Bot left a comment

Choose a reason for hiding this comment

Uh oh!

cursor Bot Sep 10, 2026

Choose a reason for hiding this comment

Recipes skip InferenceX agentic harness

Uh oh!

github-actions Bot commented Sep 11, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

csahithi commented Aug 18, 2026 •

edited by cursor Bot

Loading

cursor Bot left a comment •

edited

Loading