[aien-spark]

README L410 said the Omega Bridge framework attachments (PyTorch, llama.cpp/GGML) were "planned". They now exist in draft PRs and are exercised op by op on one GB10; this changes the line to exactly that scope.

• llama.cpp/GGML: OmegaGB10 backend: model checks, symbol isolation and empty-graph handling ggml-omega#2, opt-in OmegaGB10 device (MUL_MAT via omega_gpu_gemm_ex, RMS_NORM, swiglu; rest on ggml CPU). test-backend-ops on the chip: MUL_MAT 871 / RMS_NORM 7 / SWIGLU 12 OK, 0 FAIL; gbmm-k mutant 385 FAIL. Receipt in that PR (evidence/gb10-20261027T0726Z).
• PyTorch: native.om Python mode + PyTorch attachment on CUDA++ (pending) omega-bridge#30, torch.library ops cudapp::mm, cudapp::swiglu; checked against torch CPU, opcheck, mutant caught. Receipt evidence/attach-gb10-20261027T0726Z.
• Both depend on draft gemm_ex: general GEMM domain + fp8 fallback for refused shapes (draft) #139 (omega_gpu_gemm_ex); this should merge after it.

NOT_RUN and stated in the line: any model through either attachment on the GB10 (exclusive window). Docs only; crumbs verify OK (21 current). Pending review; no DCO.

Independent review (Main desk, 2026-10-31)

• This branch does not contain gemm_ex: general GEMM domain + fp8 fallback for refused shapes (draft) #139/gemm_ex: general GEMM domain + fp8 fallback for refused shapes (replaces #139) #214 commits; its diff against main is README.md plus evidence/GGML-OMEGA-OPS-117d8d7/ only. git merge-tree of gemm_ex: general GEMM domain + fp8 fallback for refused shapes (draft) #139 head and this head is conflict-free. Merge after gemm_ex: general GEMM domain + fp8 fallback for refused shapes (replaces #139) #214.
• Rerun on the gemm_ex: general GEMM domain + fp8 fallback for refused shapes (draft) #139 kernels (117d8d7, tree-identical to gemm_ex: general GEMM domain + fp8 fallback for refused shapes (replaces #139) #214 head f95dca3), spark-window 20261031T115524Z-7a14f4 shared: MUL_MAT 871/871, RMS_NORM 7/7, SWIGLU 12/12; gbmm-k mutant CAUGHT (301 failures). Evidence: evidence/GGML-OMEGA-OPS-117d8d7/.
• Upstream gemm_ex: general GEMM domain + fp8 fallback for refused shapes (draft) #139 measurements: gemm_ex chip PASS (902 checks); semver NC rc=1; runtime_api_check FAILs on the abi mutant.
• crumb verify OK (33). Trailers: 0.

NOT_RUN: full model generation through ggml-omega on this head; PyTorch attachment rerun; hardware qualification.

[drakestapleton]

Classification: blocked (stacked). The body says this must merge after draft #139 (omega_gpu_gemm_ex). The README L410 GB10 claims (MUL_MAT 871 / RMS_NORM 7 / SWIGLU 12 OK) rest on receipts in other repos' draft PRs (aien-dev/ggml-omega#2, aien-dev/omega-bridge#30), not in this repo.

To become merge-ready: land #139 first, then link the receipts at a fixed commit (or copy them under evidence/) and re-check the wording against what is merged.

[drakestapleton]

Classification: blocked (stacked on draft #139).

• The body says this must merge after #139 (omega_gpu_gemm_ex), which is still draft with a failing build.
• The README line's GB10 numbers (MUL_MAT 871 / RMS_NORM 7 / SWIGLU 12 OK) rest on receipts in draft PRs in other repos (ggml-omega#2, omega-bridge#30), not in this repo. Link those receipts at a fixed commit, or copy them under evidence/.
• To become merge-ready: #139 lands, the README wording is re-checked against the merged API, and checks are re-run.

— Grokbot