UPSTREAM PR #18039: [Speculative decoding] feat: add EAGLE3 speculative decoding support #568
LOCI Review / Performance Review #568
succeeded
Dec 14, 2025 in 37m 2s
Performance varied across binaries, overall acceptable
2 binaries improved · 11 binaries unchanged · 3 binaries stable ~ within threshold · 0 binaries degraded ~ beyond threshold
| Binary | Δ % Response | Δ % Throughput | Performance (based on response time) |
|---|---|---|---|
build.bin.libggml-base.so |
-0.64 | 0.05 | improved |
build.bin.libggml-cpu.so |
0 | 0 | unchanged |
build.bin.libggml.so |
0 | 0 | unchanged |
build.bin.libllama.so |
0.2 | -0.02 | stable |
build.bin.libmtmd.so |
0 | 0 | unchanged |
build.bin.llama-bench |
0 | 0.01 | unchanged |
build.bin.llama-cvector-generator |
-0.41 | 0.14 | improved |
build.bin.llama-gemma3-cli |
0 | 0 | unchanged |
build.bin.llama-gguf-split |
0 | 0.01 | unchanged |
build.bin.llama-llava-cli |
0 | 0 | unchanged |
build.bin.llama-minicpmv-cli |
0 | 0 | unchanged |
build.bin.llama-quantize |
0 | 0.01 | unchanged |
build.bin.llama-qwen2vl-cli |
0 | 0 | unchanged |
build.bin.llama-run |
0.98 | 0.09 | stable |
build.bin.llama-tokenize |
0 | 0.02 | unchanged |
build.bin.llama-tts |
0.91 | 0.07 | 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.
Loading