llama.cpp
Extended SYCL oneDNN SDPA to non-FP16 KV caches (Q4_0–Q8_0 and FP32)
#25874
Merged

Loading