Loading evidence and metric snapshots
Loading evidence and metric snapshots
rocm · sgl-project/sglang
Live candidate for Scaled dot-product attention in sgl-project/sglang has 4 of 10 observed or reviewed stages and 1 of 9 verified edges. Missing relationships remain typed gaps.
Each item names a stage whose evidence or required incoming relationship is unproven. These gaps are not failures and do not show that the capability lacks support.
Implementation
This stage or its required incoming relationship is not proven.
Gap kind: implementation
Test Mapping
This stage or its required incoming relationship is not proven.
Gap kind: test_mapping
Pull Request
This stage or its required incoming relationship is not proven.
Gap kind: pull_request
Test Execution
This stage or its required incoming relationship is not proven.
Gap kind: test_execution
Release
This stage or its required incoming relationship is not proven.
Gap kind: release
Benchmark
This stage or its required incoming relationship is not proven.
Gap kind: benchmark
Analyst Decision
This stage or its required incoming relationship is not proven.
Gap kind: analyst_decision
Report Version
This stage or its required incoming relationship is not proven.
Gap kind: report_version
01 / Scaled dot-product attention
Scaled dot-product attention
02 / ROCM implementation at python/sglang/srt/layers/attention/aiter_backend.py
Implements · Inferred, not completion eligible
ROCM implementation at python/sglang/srt/layers/attention/aiter_backend.py
The implementation is registered, but an allowlisted parity review has not approved it.
03 / Human-approved capability-to-test mapping required
Maps To Test · Relationship missing
Human-approved capability-to-test mapping required
No effective human-approved capability-to-test mapping has public evidence.
04 / #17327 Disable mla persistent kernel when not using fp8 kv_cache
Changed By · Relationship missing
#17327 Disable mla persistent kernel when not using fp8 kv_cache
Both an approved mapping and an explicit implementation pull-request link are required.
05 / PR Test (AMD) attempt 2
Validated By · Provider verified
PR Test (AMD) attempt 2
The pull request and CI attempt resolve to exact commit 35443039efe9c82aa73af6e8692fe7b2c5bd308d.
06 / Exact mapped-test execution required
Executed In · Relationship missing
Exact mapped-test execution required
A CI success does not prove that the mapped test executed.
07 / Verified containing release required
Included In · Relationship missing
Verified containing release required
No explicit release containment link joins the tested commit to a release.
08 / Exact capability and release benchmark mapping required
Measured By · Relationship missing
Exact capability and release benchmark mapping required
Contextual hardware results do not complete an exact capability and release mapping.
09 / Human analyst decision required
Reviewed As · Relationship missing
Human analyst decision required
A new source-derived chain version requires a new allowlisted human decision.
10 / Cited weekly report required
Cited In · Relationship missing
Cited weekly report required
No frozen weekly report cites a human decision for this chain version.
Scroll horizontally to inspect every column.
| From | Relation | To | Basis |
|---|---|---|---|
| ROCM implementation at python/sglang/srt/layers/attention/aiter_backend.py | Maps To Test | Human-approved capability-to-test mapping required | Missing |
| PR Test (AMD) attempt 2 | Executed In | Exact mapped-test execution required | Missing |
| Verified containing release required | Measured By | Exact capability and release benchmark mapping required | Missing |
| Human analyst decision required | Cited In | Cited weekly report required | Missing |
| Scaled dot-product attention | Implements | ROCM implementation at python/sglang/srt/layers/attention/aiter_backend.py | Inferred |
| Human-approved capability-to-test mapping required | Changed By | #17327 Disable mla persistent kernel when not using fp8 kv_cache | Missing |
| Exact capability and release benchmark mapping required | Reviewed As | Human analyst decision required | Missing |
| #17327 Disable mla persistent kernel when not using fp8 kv_cache | Validated By | PR Test (AMD) attempt 2 | Provider Verified |
| Exact mapped-test execution required | Included In | Verified containing release required | Missing |
SGLang PR 17327 disables the persistent MLA kernel without FP8 KV cache
Merged SGLang change 6988a0f5706c05ebc40ba9ef5c09243bcbf9886c updates the ROCm AITER attention backend guard.
sha256:95155a85c57c66b0935e33dd53407de965e70a4cbe822c0000dde3f3d75b9f01
SGLang PR Test AMD run 21151393674 attempt 2
GitHub Actions reports attempt 2 completed successfully for PR head 35443039efe9c82aa73af6e8692fe7b2c5bd308d.
sha256:0d700cb8a0f3462740d0970475cc3ded8b82d685bfd6e6bb6ed234e685eaef14