UPSTREAM PR #16829: cpu: introduce chunking for flash attention #6
LOCI Review Dev / Performance Review #6
succeeded
Oct 28, 2025 in 1h 8m 57s
Performance varied across binaries, overall acceptable
0 binaries improved · 4 binaries unchanged · 0 binaries stable ~ within threshold · 0 binaries degraded ~ beyond threshold
| Binary | Δ % Response | Δ % Throughput | Performance (based on response time) |
|---|---|---|---|
build.bin.libllama.so |
-0.00 | -0.00 | unchanged |
build.bin.libggml.so |
-0.00 | -0.00 | unchanged |
build.bin.libggml-cpu.so |
-0.00 | -0.00 | unchanged |
build.bin.libggml-base.so |
-0.00 | -0.00 | unchanged |
Performance threshold: 30%
Default configuration used.
Note: Performance status is evaluated only from Δ% Response. Throughput is displayed for reference.
Access the complete analysis in the LOCI Dashboard.
Open the Pull Request linked to this check-run.
Loading