Skip to content

VALIDATION ONLY: current-main PR132 correction composition - #28

Closed
nekomario28 wants to merge 1 commit into
validation/pr132-current-main-compositionfrom
validation/pr132-current-main-composition-carrier
Closed

VALIDATION ONLY: current-main PR132 correction composition#28
nekomario28 wants to merge 1 commit into
validation/pr132-current-main-compositionfrom
validation/pr132-current-main-composition-carrier

Conversation

@nekomario28

@nekomario28 nekomario28 commented Aug 25, 2026

Copy link
Copy Markdown
Owner

One-shot hosted composition carrier. Closed unmerged after successful evidence capture. Base is an integration-only revision that overlays the seven independently validated correction blobs onto GitHub's current synthetic merge of upstream main and PR FlashML-org#132.

Pinned identities:

  • upstream current main: 2757bb5f91156fc8a44d88ec4b302a81f10c9e81
  • upstream PR feat(rocm): add RDNA3 and RDNA4 runtime foundation FlashML-org/FreeToken#132 head: 45417560beeba896a9b0866e50154d2a7a320b8b
  • GitHub synthetic merge: 66156d38240f1e17640c8e90fa7b7fdcc1c6c409
  • corrected current-merge composition: 9a72fcf4c2490f72e4fa05880d6315ba0279961c
  • carrier head: 97377b81207612eabca57d365be56e70d521f069

Hosted result

Run 32816161007, job 97704765383: SUCCESS on Ubuntu 24.04.

Observed markers:

ROCM_CURRENT_MERGE_MATERIAL=EXACT_VALIDATED_BLOBS
ROCM_HOST_POINTER_SOURCE_CONTRACT=PASS
ROCM_VERSIONED_RUNTIME_SELECTION=PASS_NUMERIC_7_14_OVER_7_9
ROCM_JIT_RUNTIME_RESOLUTION=PASS_NUMERIC_SELECTION_AND_CACHE_LIFETIME
ROCM_CURRENT_MERGE_GPU_SELECT=PASS_NVML_OPTIONAL_VISIBLE_INDEX
ROCM_CURRENT_MERGE_COMPOSITION=PASS_SOURCE_CONTRACTS

The job verified exact synthetic-merge/candidate/carrier ancestry, exact seven-file material scope, all seven expected Git blob identities, git diff --check, syntax, all three correction contracts, and the current-main FlashML-org#117 no-NVML visible-index source contract: after forcing NVML discovery unavailable, resolve_gpu_uuids(('0',)) returns None, and set_assigned_gpu('0') preserves visible ordinal 0 without importing Torch.

Interpretation: the current FlashML-org#117 composition does not establish a source-level blocker for the single-GPU/no-NVML visible-index path. This is not a physical AMD serve test and does not validate AMD UUID UX, TP/multi-GPU communication, performance, or NVIDIA parity.

Carrier is intentionally not merged.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant