UPSTREAM PR #18471: Add self‑speculative decoding (no draft model required) #750
LOCI Review / Performance Per Binary #750
succeeded
Jan 27, 2026
Performance varied across binaries, overall acceptable
2 binaries improved · 9 binaries unchanged · 4 binaries stable ~ within threshold · 0 binaries degraded ~ beyond threshold
| Binary | Δ % Response | Δ % Throughput | Performance (based on response time) |
|---|---|---|---|
build.bin.libggml-base.so |
0 | 0 | unchanged |
build.bin.libggml-cpu.so |
0 | 0 | unchanged |
build.bin.libggml.so |
0 | 0 | unchanged |
build.bin.libllama.so |
0 | 0 | unchanged |
build.bin.libmtmd.so |
0 | 0 | unchanged |
build.bin.llama-bench |
0.02 | 0.04 | stable |
build.bin.llama-cvector-generator |
0.93 | 3 | stable |
build.bin.llama-gemma3-cli |
0 | 0 | unchanged |
build.bin.llama-gguf-split |
-0.03 | -0.11 | improved |
build.bin.llama-llava-cli |
0 | 0 | unchanged |
build.bin.llama-minicpmv-cli |
0 | 0 | unchanged |
build.bin.llama-quantize |
-0.01 | -0.07 | improved |
build.bin.llama-qwen2vl-cli |
0 | 0 | unchanged |
build.bin.llama-tokenize |
0.01 | -0.07 | stable |
build.bin.llama-tts |
0.49 | 2.85 | stable |
Performance threshold: 30%
Default configuration used.
Note: Performance status is evaluated only from Δ% Response. Throughput is displayed for reference.
Explore the complete analysis inside the Version Insights.
Open the Pull Request linked to this check-run.
Loading