{ "schemaVersion": 1, "operationId": "20260830004055-24109c0f27", "checkedAt": "2026-08-30T00:52:34.274Z", "status": "pinned-zero-runtime-glm52-vllm-cpu-sparse-attention-audit", "sourceReceipts": [ { "id": "zai-glm52-release", "url": "https://z.ai/blog/glm-5.2", "finalUrl": "https://z.ai/blog/glm-5.2", "purpose": "mandatory first-party GLM-5.2 release check", "httpStatus": 200, "bytes": 598, "sha256": "a9e8c2b6f34717d69e3a0aa26bb117256a4d8c95bd299910c2a693000ee88fe8" }, { "id": "zhipu-research-index", "url": "http://zhipuai.cn/zh/research", "finalUrl": "https://www.zhipuai.cn/zh/research", "purpose": "mandatory Zhipu AI research-discovery check", "httpStatus": 200, "bytes": 1235119, "sha256": "7d50f4c290fbc240f50fabc8d18f7a499688f968f53951655572909ddf547d4e" }, { "id": "hf-glm52-config", "url": "https://huggingface.co/zai-org/GLM-5.2-FP8/resolve/ba978f7d347eaf65d22f1a86833408afdb953541/config.json", "finalUrl": "https://huggingface.co/api/resolve-cache/models/zai-org/GLM-5.2-FP8/ba978f7d347eaf65d22f1a86833408afdb953541/config.json?%2Fzai-org%2FGLM-5.2-FP8%2Fresolve%2Fba978f7d347eaf65d22f1a86833408afdb953541%2Fconfig.json=&etag=%224e1f0168afd127189fb1c4ddb1d4476a4fca96ac%22", "purpose": "pin the GLM-5.2 DSA architecture and positive index_topk value", "httpStatus": 200, "bytes": 29464, "sha256": "22e49334abf8562fecf70ca3292ba3f5b33f5602fb2bf10b52dd64a66cfe65ff" }, { "id": "vllm-glm52-recipe", "url": "https://raw.githubusercontent.com/vllm-project/recipes/5943215a27acb4a243e9d27bd69daf491034cfea/models/zai-org/GLM-5.2.yaml", "finalUrl": "https://raw.githubusercontent.com/vllm-project/recipes/5943215a27acb4a243e9d27bd69daf491034cfea/models/zai-org/GLM-5.2.yaml", "purpose": "pin the official GLM-5.2 vLLM serving recipe", "httpStatus": 200, "bytes": 15247, "sha256": "517662028453e8a0b62f53a8b507d6b8ac502aaecbd5452b396c37ea2fbae4a7" }, { "id": "vllm-issue-54018", "url": "https://api.github.com/repos/vllm-project/vllm/issues/54018", "finalUrl": "https://api.github.com/repos/vllm-project/vllm/issues/54018", "purpose": "capture the CPU sparse-attention support report", "httpStatus": 200, "bytes": 5923, "sha256": "ce77f8784be45a59e559b4903fb0dccdf6e73493d3d02706cd978e42d8a97156" }, { "id": "vllm-issue-54018-comments", "url": "https://api.github.com/repos/vllm-project/vllm/issues/54018/comments", "finalUrl": "https://api.github.com/repos/vllm-project/vllm/issues/54018/comments", "purpose": "capture public follow-up on the CPU report", "httpStatus": 200, "bytes": 2, "sha256": "4f53cda18c2baa0c0354bb5f9a3ecbe5ed12ab4d8e11ba873c2f11161202b945" }, { "id": "vllm-pr-54029", "url": "https://api.github.com/repos/vllm-project/vllm/pulls/54029", "finalUrl": "https://api.github.com/repos/vllm-project/vllm/pulls/54029", "purpose": "capture the proposed dense-MLA opt-out state and validation boundary", "httpStatus": 200, "bytes": 19227, "sha256": "a469e45ac512726b410640c38dfea533b40ef54b2517351b905856cf638ca7cd" }, { "id": "vllm-pr-54029-files", "url": "https://api.github.com/repos/vllm-project/vllm/pulls/54029/files", "finalUrl": "https://api.github.com/repos/vllm-project/vllm/pulls/54029/files", "purpose": "capture the exact proposed source and test scope", "httpStatus": 200, "bytes": 6413, "sha256": "33d044ab50e0b27445ec287a0076362b7f8ac104557eaa4bf856b06304c904de" }, { "id": "vllm-pr-54029-reviews", "url": "https://api.github.com/repos/vllm-project/vllm/pulls/54029/reviews", "finalUrl": "https://api.github.com/repos/vllm-project/vllm/pulls/54029/reviews", "purpose": "check whether the proposed fix has a maintainer approval", "httpStatus": 200, "bytes": 1751, "sha256": "52d6a6c87541f715d13860bf0ae65d56ee8981d339e05f4df747991bb7cb3f72" }, { "id": "vllm-pr-51471", "url": "https://api.github.com/repos/vllm-project/vllm/pulls/51471", "finalUrl": "https://api.github.com/repos/vllm-project/vllm/pulls/51471", "purpose": "separate merged dense MLA CPU plumbing from GLM sparse-attention support", "httpStatus": 200, "bytes": 44005, "sha256": "98fc93ae802d17fe17b2577f4db2194d045b4b2f71ab71e9de5e5b01eb52359d" }, { "id": "vllm-stable-commit", "url": "https://api.github.com/repos/vllm-project/vllm/commits/v0.28.0", "finalUrl": "https://api.github.com/repos/vllm-project/vllm/commits/v0.28.0", "purpose": "resolve the v0.28.0 release tag", "httpStatus": 200, "bytes": 4868, "sha256": "07376e59bec891c8edfeb322579383c523f352d524a5a043eb56a5c67bb1bac9" }, { "id": "vllm-main-commit", "url": "https://api.github.com/repos/vllm-project/vllm/commits/680e2177e473ed8dfaa9773f7ead185b369cab46", "finalUrl": "https://api.github.com/repos/vllm-project/vllm/commits/680e2177e473ed8dfaa9773f7ead185b369cab46", "purpose": "pin the checked main-branch snapshot", "httpStatus": 200, "bytes": 7752, "sha256": "32532bd3c41db3111f9f1c914f43c338744153ab2e14e38258900a4a48385990" }, { "id": "transformers-commit", "url": "https://api.github.com/repos/huggingface/transformers/commits/93c8b7b485963a10800c91f55304db6be211c2bd", "finalUrl": "https://api.github.com/repos/huggingface/transformers/commits/93c8b7b485963a10800c91f55304db6be211c2bd", "purpose": "pin the Transformers 5.16.1 source snapshot", "httpStatus": 200, "bytes": 5415, "sha256": "089064864d16b9a5a2d7dce33e8099c44f83feb013f3b99b0cfdcbad6772e60b" }, { "id": "stable-cpu-platform", "url": "https://raw.githubusercontent.com/vllm-project/vllm/2cf0a6915ce544dc493a0990f2ea38d81601128a/vllm/platforms/cpu.py", "finalUrl": "https://raw.githubusercontent.com/vllm-project/vllm/2cf0a6915ce544dc493a0990f2ea38d81601128a/vllm/platforms/cpu.py", "purpose": "check the stable CPU sparse-attention backend gate", "httpStatus": 200, "bytes": 21705, "sha256": "6bac9048589b7fbe3051f5099225c9d54808e3949461f206a0dc6164a711de94" }, { "id": "stable-deepseek-v2", "url": "https://raw.githubusercontent.com/vllm-project/vllm/2cf0a6915ce544dc493a0990f2ea38d81601128a/vllm/model_executor/models/deepseek_v2.py", "finalUrl": "https://raw.githubusercontent.com/vllm-project/vllm/2cf0a6915ce544dc493a0990f2ea38d81601128a/vllm/model_executor/models/deepseek_v2.py", "purpose": "check the stable index_topk presence test and sparse routing", "httpStatus": 200, "bytes": 75645, "sha256": "58d8916458de7c6f73b40bfef9d2f57bdd6fa0fb79be9e269331af6e66149fe2" }, { "id": "stable-model-config", "url": "https://raw.githubusercontent.com/vllm-project/vllm/2cf0a6915ce544dc493a0990f2ea38d81601128a/vllm/model_executor/models/config.py", "finalUrl": "https://raw.githubusercontent.com/vllm-project/vllm/2cf0a6915ce544dc493a0990f2ea38d81601128a/vllm/model_executor/models/config.py", "purpose": "check stable GLM config verification for a CPU opt-out", "httpStatus": 200, "bytes": 41289, "sha256": "4111381726c3e5dc55870f4a1412c8c29c786455d024b74db34775eefb70847c" }, { "id": "main-cpu-platform", "url": "https://raw.githubusercontent.com/vllm-project/vllm/680e2177e473ed8dfaa9773f7ead185b369cab46/vllm/platforms/cpu.py", "finalUrl": "https://raw.githubusercontent.com/vllm-project/vllm/680e2177e473ed8dfaa9773f7ead185b369cab46/vllm/platforms/cpu.py", "purpose": "check the pinned main CPU sparse-attention backend gate", "httpStatus": 200, "bytes": 23334, "sha256": "0204d451540f1b9cd26cac1061d2a14e50a9abf2eac0e377367ec79f6baa5530" }, { "id": "main-deepseek-v2", "url": "https://raw.githubusercontent.com/vllm-project/vllm/680e2177e473ed8dfaa9773f7ead185b369cab46/vllm/model_executor/models/deepseek_v2.py", "finalUrl": "https://raw.githubusercontent.com/vllm-project/vllm/680e2177e473ed8dfaa9773f7ead185b369cab46/vllm/model_executor/models/deepseek_v2.py", "purpose": "check pinned main index_topk presence routing", "httpStatus": 200, "bytes": 77732, "sha256": "8f34352a6a86da98a727c6c6c734dfb4d439fdb5f5f30e2a1472b54ca81f7a23" }, { "id": "main-model-config", "url": "https://raw.githubusercontent.com/vllm-project/vllm/680e2177e473ed8dfaa9773f7ead185b369cab46/vllm/model_executor/models/config.py", "finalUrl": "https://raw.githubusercontent.com/vllm-project/vllm/680e2177e473ed8dfaa9773f7ead185b369cab46/vllm/model_executor/models/config.py", "purpose": "check pinned main GLM config verification for a CPU opt-out", "httpStatus": 200, "bytes": 42161, "sha256": "f1e9741512ba55871af5ef2a08e3d3ad16e2f01d6fd5c76485ccfcfcd5831095" }, { "id": "proposed-model-config", "url": "https://raw.githubusercontent.com/vllm-project/vllm/3d6db6b56381dc26d595509683fb556c24591e5a/vllm/model_executor/models/config.py", "finalUrl": "https://raw.githubusercontent.com/vllm-project/vllm/3d6db6b56381dc26d595509683fb556c24591e5a/vllm/model_executor/models/config.py", "purpose": "pin the open PR dense-opt-out helper and early CPU error", "httpStatus": 200, "bytes": 42569, "sha256": "9c2332f639120dfa59a663fb165a9fff2c043beb0baba9adae681ab5a17cbf40" }, { "id": "proposed-deepseek-v2", "url": "https://raw.githubusercontent.com/vllm-project/vllm/3d6db6b56381dc26d595509683fb556c24591e5a/vllm/model_executor/models/deepseek_v2.py", "finalUrl": "https://raw.githubusercontent.com/vllm-project/vllm/3d6db6b56381dc26d595509683fb556c24591e5a/vllm/model_executor/models/deepseek_v2.py", "purpose": "pin the open PR index_topk value test", "httpStatus": 200, "bytes": 77329, "sha256": "0597278fbe5dac845226689044ddd760b0d1d585ade793af94198bcf05a5bd50" }, { "id": "proposed-tests", "url": "https://raw.githubusercontent.com/vllm-project/vllm/3d6db6b56381dc26d595509683fb556c24591e5a/tests/test_config.py", "finalUrl": "https://raw.githubusercontent.com/vllm-project/vllm/3d6db6b56381dc26d595509683fb556c24591e5a/tests/test_config.py", "purpose": "pin the open PR config-only regression tests", "httpStatus": 200, "bytes": 85922, "sha256": "1ce8a82c393ceb3cbe954d4e657fb35c8a23e9e82327234fd3eca1ec2be88831" }, { "id": "transformers-glm-config", "url": "https://raw.githubusercontent.com/huggingface/transformers/93c8b7b485963a10800c91f55304db6be211c2bd/src/transformers/models/glm_moe_dsa/configuration_glm_moe_dsa.py", "finalUrl": "https://raw.githubusercontent.com/huggingface/transformers/93c8b7b485963a10800c91f55304db6be211c2bd/src/transformers/models/glm_moe_dsa/configuration_glm_moe_dsa.py", "purpose": "pin the typed index_topk default that survives JSON omission", "httpStatus": 200, "bytes": 7926, "sha256": "8c8b95bb0ecfcb502f768905964926ea55a83234e4b2f494b6f3b68333891a37" }, { "id": "vllm-cpu-docs", "url": "https://raw.githubusercontent.com/vllm-project/vllm/2cf0a6915ce544dc493a0990f2ea38d81601128a/docs/getting_started/installation/cpu.md", "finalUrl": "https://raw.githubusercontent.com/vllm-project/vllm/2cf0a6915ce544dc493a0990f2ea38d81601128a/docs/getting_started/installation/cpu.md", "purpose": "pin general CPU backend installation and serving guidance", "httpStatus": 200, "bytes": 17744, "sha256": "1fee5e78f149a8ffcff714c222a4cb7130b9ece0274a91dd3a19d674d8c4d342" }, { "id": "vllm-apple-cpu-docs", "url": "https://raw.githubusercontent.com/vllm-project/vllm/2cf0a6915ce544dc493a0990f2ea38d81601128a/docs/getting_started/installation/cpu.apple.inc.md", "finalUrl": "https://raw.githubusercontent.com/vllm-project/vllm/2cf0a6915ce544dc493a0990f2ea38d81601128a/docs/getting_started/installation/cpu.apple.inc.md", "purpose": "pin Apple Silicon CPU support status", "httpStatus": 200, "bytes": 3184, "sha256": "00e3ec14d5de01fb6c70857699630fdcf4c03d12df51a6f2e7d50bce72769a97" }, { "id": "vllm-cpu-model-matrix", "url": "https://raw.githubusercontent.com/vllm-project/vllm/2cf0a6915ce544dc493a0990f2ea38d81601128a/docs/models/hardware_supported_models/cpu.md", "finalUrl": "https://raw.githubusercontent.com/vllm-project/vllm/2cf0a6915ce544dc493a0990f2ea38d81601128a/docs/models/hardware_supported_models/cpu.md", "purpose": "check whether GLM-5.2 appears in the validated CPU model matrix", "httpStatus": 200, "bytes": 5092, "sha256": "ab37c09f3216e499f56f0c2419979a6661e97050db5862852abd448cdd7fa0b8" } ], "pins": { "modelRevision": "ba978f7d347eaf65d22f1a86833408afdb953541", "stableVllmTag": "v0.28.0", "stableVllmRevision": "2cf0a6915ce544dc493a0990f2ea38d81601128a", "checkedMainRevision": "680e2177e473ed8dfaa9773f7ead185b369cab46", "proposedFixRevision": "3d6db6b56381dc26d595509683fb556c24591e5a", "transformersTag": "v5.16.1", "transformersRevision": "93c8b7b485963a10800c91f55304db6be211c2bd", "cpuMlaMergeRevision": "6f7df92a8e6cdc74a725b8f10b4d0b48ba2b37ef", "recipeRevision": "5943215a27acb4a243e9d27bd69daf491034cfea" }, "modelContract": { "architectures": [ "GlmMoeDsaForCausalLM" ], "modelType": "glm_moe_dsa", "hiddenLayers": 78, "routedExperts": 256, "activeExpertsPerToken": 8, "indexTopK": 2048, "indexHeads": 32, "checkpointTransformersVersion": "5.12.0", "quantizationMethod": "fp8" }, "admissionAudit": { "stableV0280": { "tag": "v0.28.0", "resolvedSha": "2cf0a6915ce544dc493a0990f2ea38d81601128a", "committedAt": "2026-08-24T23:42:42Z", "sparseCpuErrorPresent": true, "indexTopkPresenceChecks": 2, "passesSparseFlag": true, "denseOptOutHelperPresent": false, "actionableCpuOverridePresent": false }, "checkedMain": { "resolvedSha": "680e2177e473ed8dfaa9773f7ead185b369cab46", "committedAt": "2026-08-29T23:48:22Z", "sparseCpuErrorPresent": true, "indexTopkPresenceChecks": 2, "passesSparseFlag": true, "denseOptOutHelperPresent": false, "actionableCpuOverridePresent": false }, "transformers": { "tag": "v5.16.1", "resolvedSha": "93c8b7b485963a10800c91f55304db6be211c2bd", "committedAt": "2026-08-26T14:30:13Z", "typedDefault2048": true, "documentationDefault2048": true, "assignsIndexerScheduleFromTopK": true }, "stableRejectsSparseCpu": true, "checkedMainStillRejectsSparseCpu": true, "deletingJsonKeyDoesNotDisableSparsePath": true, "stableAndMainLackDenseOptOut": true }, "issueStatus": { "number": 54018, "title": "[Bug][CPU] GLM-5.x (glm_moe_dsa) cannot run on CPU: sparse attention forced by config, no way to disable", "url": "https://github.com/vllm-project/vllm/issues/54018", "state": "open", "createdAt": "2026-08-27T08:07:57Z", "updatedAt": "2026-08-27T08:08:05Z", "commentCount": 0, "environmentVllm0280": true, "environmentAppleM3": true, "configOnlyDummyReproReported": true, "deletingJsonFieldStillAddsAttributeReported": true, "localDensePatchReported": true, "localDensePatchFullModelValidated": false }, "proposedFix": { "number": 54029, "title": "[Bugfix][CPU] Allow dense attention fallback for GLM DSA", "url": "https://github.com/vllm-project/vllm/pull/54029", "state": "open", "draft": false, "mergedAt": null, "headSha": "3d6db6b56381dc26d595509683fb556c24591e5a", "baseSha": "fd57c4b7afebc0b43d25ed7f5848fc35786463d0", "maintainerApprovals": 0, "changedFiles": [ "tests/test_config.py", "vllm/model_executor/models/config.py", "vllm/model_executor/models/deepseek_v2.py" ], "expectedThreeFilesOnly": true, "explicitZeroMeansDense": true, "earlyCpuError": true, "actionableOverride": true, "modelUsesHelperTwice": true, "configFixturesCoverZeroAnd2048": true, "configTestsReported": true, "fullLinuxGlmStartupReported": false, "outputParityReported": false }, "priorCpuMlaWork": { "number": 51471, "title": "[CPU][MLA] Fix prefill backend selection so MLA runs end-to-end on CPU", "url": "https://github.com/vllm-project/vllm/pull/51471", "merged": true, "mergedAt": "2026-08-28T04:36:56Z", "mergeCommitSha": "6f7df92a8e6cdc74a725b8f10b4d0b48ba2b37ef", "purposeNamesDeepSeekV2Lite": true, "providesGlmSparseOptOut": false }, "documentationAudit": { "cpuVariantsIncludeAppleSilicon": true, "appleSupportExperimental": true, "appleRequiresSourceBuild": true, "cpuMatrixListsGlm52": false, "cpuMatrixListsGlm49b": true, "officialRecipeNamesGlm52": true }, "decision": { "currentVerdict": "The pinned vLLM v0.28.0 and checked-main snapshots cannot admit the default GLM-5.2 configuration on CPU because positive index_topk selects sparse attention and the CPU backend rejects sparse attention.", "safeAction": "Keep CPU attempts out of production, use a supported GPU or hosted route, and wait for a merged pinned dense-opt-out followed by full-model startup and output-parity receipts.", "unsafeAction": "Do not delete index_topk from config.json or patch self.is_v32 locally and treat a process start as correctness evidence.", "evidenceBoundary": "Static source can prove the current admission path and proposal state, but it cannot prove full-model CPU correctness, output parity, acceptable speed, memory fit, or a future merge." }, "canaryGate": { "algorithm": [ "Pin the GLM-5.2 checkpoint and exact vLLM commit; inspect the effective typed index_topk after config loading.", "Reject positive index_topk on CPU while the backend explicitly rejects sparse attention.", "Treat index_topk zero as experimental until the opt-out is merged into the pinned runtime and its exact tests pass.", "Require full-model startup plus deterministic prompt, stop-token, reasoning, tool-call, and long-context parity before a bounded functional canary.", "Measure memory and speed only after correctness; a CPU functional path does not imply the 744B checkpoint is practical." ], "fixtures": [ { "name": "v0.28.0-default-config", "effectiveIndexTopK": 2048, "runtimeUnderstandsZeroAsDense": false, "changeMerged": false, "fullModelStartupPassed": false, "outputParityPassed": false, "decision": "reject", "reason": "positive-index-topk-enters-unsupported-cpu-sparse-attention" }, { "name": "index-topk-deleted-from-json", "effectiveIndexTopK": 2048, "runtimeUnderstandsZeroAsDense": false, "changeMerged": false, "fullModelStartupPassed": false, "outputParityPassed": false, "decision": "reject", "reason": "positive-index-topk-enters-unsupported-cpu-sparse-attention" }, { "name": "open-pr-index-topk-zero", "effectiveIndexTopK": 0, "runtimeUnderstandsZeroAsDense": true, "changeMerged": false, "fullModelStartupPassed": false, "outputParityPassed": false, "decision": "hold", "reason": "dense-opt-out-is-not-in-a-merged-pinned-runtime" }, { "name": "future-merged-and-validated-zero-override", "effectiveIndexTopK": 0, "runtimeUnderstandsZeroAsDense": true, "changeMerged": true, "fullModelStartupPassed": true, "outputParityPassed": true, "decision": "eligible-for-bounded-functional-canary", "reason": "merged-pin-startup-and-parity-receipts-present" } ] }, "overlapAudit": { "prepublicationSitemapUrls": 97, "registryPages": 90, "distinctIntent": true, "primaryIntent": "audit whether GLM-5.2 can enter the vLLM CPU backend and distinguish the current sparse-attention rejection from an unmerged dense-MLA proposal", "readerJob": "pin the checkpoint and runtime source, detect the index_topk sparse gate, avoid unsupported local source edits, and admit a CPU functional canary only after a merged fix and output parity", "nearby": [ [ "https://glm52.ai/guides/run-glm-5-2-locally/", "Owns RAM, VRAM, storage, context, and cost fit across local hardware; it does not audit the vLLM CPU sparse-attention gate." ], [ "https://glm52.ai/guides/glm-5-2-vllm-torch-compile/", "Owns torch.compile activation and eager fallback on accelerator-oriented model runners; it does not cover CPU attention-backend admission." ], [ "https://glm52.ai/guides/glm-5-2-amd-rocm-sglang/", "Owns SGLang deployment on AMD accelerators, not vLLM CPU inference." ], [ "https://glm52.ai/guides/glm-5-2-fp8-download-verify/", "Owns immutable checkpoint download and shard verification, not CPU runtime support." ] ] }, "searchSupply": { "query": "GLM-5.2 vLLM CPU", "serpApiRequests": 1, "requestId": "serpapi-763fcd7646244aefb1e091d13d656a09", "knownMonthlyUsageAfter": 841, "result": "http-error", "evaluation": "failed", "retryCount": 0, "decisionImpact": "No Google evidence was returned. Topic selection and overlap remain based on the committed registry, live sitemap, and pinned upstream sources." }, "aiHot": [ { "itemId": "cmtf0ibgi091wrovjvv5ee7qv", "permalink": "https://aihot.virxact.com/items/cmtf0ibgi091wrovjvv5ee7qv", "classification": "duplicate-intent", "note": "The existing Hugging Face breach forensics guide already owns incident verification and disputed GLM-5.2 attribution; a civilization retelling adds no distinct reader task." }, { "itemId": "cmtdssm0205s8ro2mzzv9s9kq", "permalink": "https://aihot.virxact.com/items/cmtdssm0205s8ro2mzzv9s9kq", "classification": "duplicate-event", "note": "This is another Cursor and OpenAI access statement from the same event cluster and still creates no direct GLM-5.2 runtime task." } ], "discoveryChecks": { "zaiGlm52Release": "fetched and hashed with HTTP 200", "zhipuResearchIndex": "fetched and hashed after redirect with non-empty bytes" }, "protectedBaseline": { "homepageSha256": "5d7f8327af1af910c0484cdd0b7fde8695a107339c207cc585452210f7fd05a8", "backlinkPageSha256": "bce62493373dcf3e21e18a2b5ba7dcebf7d83a34a1faed21d3395cb477998de0", "homepageTitle": "GLM-5.2 Developer Guide: API, Pricing & Tools | GLM52.ai", "backlinkTitle": "GLM-5.2 Local Hardware Guide: RAM, VRAM & Cost | GLM52.ai" }, "runtimeBoundary": { "modelCalls": 0, "modelWeightBytesDownloaded": 0, "vllmImports": 0, "transformersImports": 0, "cpuInferenceRuns": 0, "gpuRuns": 0, "servingProcesses": 0, "containers": 0, "note": "Static source can prove the current admission path and proposal state, but it cannot prove full-model CPU correctness, output parity, acceptable speed, memory fit, or a future merge." }, "invariants": { "twentySixSourceReceipts": true, "allSourcesReturned200": true, "allSourceHashesValid": true, "immutablePinsMatch": true, "glmArchitecturePinned": true, "stableAndMainGatePinned": true, "transformerDefaultReinsertsTopK": true, "issueOpenWithConfigOnlyReport": true, "proposedFixIsOpenAndUnvalidated": true, "priorMlaWorkDoesNotCloseGap": true, "docsDoNotClaimGlm52CpuValidation": true, "fourGateFixturesCoverCurrentAndFuture": true, "distinctIntent": true, "oneFailedSerpRequestNoRetry": true, "twoAiHotItemsClassified": true, "zeroRuntimeExecution": true } }