Repository navigation
Start model-agnostic EXL3 SM121 kernel lab - #75
Conversation
plotarmordev
added
triage/skip
Review follow-up —
|
plotarmordev
removed
the
awaiting-test-evidence
Keeps both sides of the README path table: main's TP3/TP4/NFS rows and the kernel-lab rows added by this branch. No other integration change.
Maintainer decision on MiaAI-Lab#75: retain kernel_lab/, its CPU oracles/tests and docs/exl3-sm121-kernel-lab.md in this PR as development-only work, and remove every production Dockerfile addition from it. The ARG/ENV runtime stamp and the kernel_lab COPY are gone, so the Dockerfile is byte-identical to main's (pinned EXLLAMAV3_COMMIT build argument, compile/overlay layers and late recipe-stamp conventions untouched). README.md, kernel_lab/README.md and the engineering log now describe the lab as a source checkout that the production Dockerfile neither copies nor installs, drop the candidate-image build recipe and the in-image stamp claim, and keep measured provenance fail-closed and declarative: --runtime-commit or the clean checkout's Git identity, the environment override only when it is exactly the loaded source, and a developer-provided backend must set its real EXLLAMAV3_COMMIT before the metadata module is imported, since the pinned fallback describes only the production pin. Historical author results stay attributed and no speedup is claimed for the untimed SM121 sweep. No code changes: kernel_lab/exl3/**, tests/test_exl3_kernel_lab.py are unchanged from the author's head.
…rdev/pr75-cleanup-7c8b
plotarmordev
left a comment
plotarmordev
left a comment
There was a problem hiding this comment.
SIX exact-head approval: 289e365. Maintainer scope decision implemented in this same PR: kernel_lab, its CPU tests and docs are development-only; zero PR-attributable Dockerfile changes and no production launch hook. Independently reviewed all carried code and corrected developer provenance instructions. On this exact cumulative main-integrated head, /usr/bin/python3 ran 21 focused tests successfully, K1-K8 CPU oracle passed 8/8, and provenance smoke resolved the exact source SHA while refusing nonrepository and nonfull identity. PyTorch was absent; no CUDA/GPU inspection or measurement occurred. Complete publishing tree and pushed range scanned; author source branch/history preserved. No workflows/checks exist, not CI green. Runtime/backend identity remains a caller declaration, not binary attestation; historical CUDA claims remain author-attributed, and no SM121 speedup is claimed. Production-image scope hold is resolved; no hardware sweep was commissioned or required for development-only inclusion.
plotarmordev
removed
the
merge-hold
Summary
Validation
python3 -m pytest tests/test_exl3_kernel_lab.py -q: 22 passedHardware gate
This development host has no NVIDIA GPU. The PR intentionally makes no SM121 speed claim; the next step is to run the included isolated tactic sweep on a GB10/SM121 host and choose reconstruction, launch, or representation work from the measured bottleneck. The production ExLlamaV3 pin is unchanged.