community pool / model answer

KAT-Coder-V2.5-Dev

code · Kwaipilot/KAT-Coder-V2.5-Dev

75 tok/s · model median · 4 runs in the model record.

Reference rigQuantizationMedian resultBasisTTFT (ms)Peak VRAMMax contextEngines
Radeon AI Pro R9700 32GB ×38-bit75.8 tok/sreported34442.8 GB787llama.cpp
Ryzen AI Max 395 128GB4-bit70.3 tok/sreported1,359- GB787llama.cpp

Context tested

The table above is the tested context for each exact rig and quantization cell. Empty fields remain empty rather than being filled with an estimate.