ollama - 💡(How to fix) Fix Performance decrease on Beelink GTR9 Pro AMD Ryzen™ AI Max+ 395

Environment="OLLAMA_NUM_GPU=99"
Environment="OLLAMA_MAX_LOADED_MODELS=3"
Environment="OLLAMA_FLASH_ATTENTION=1"
Environment="OLLAMA_KEEP_ALIVE=60m"
Environment="HSA_OVERRIDE_GFX_VERSION=11.5.1"
Environment="ROCR_VISIBLE_DEVICE=0"
Environment="HIP_VISIBLE_DEVICE=0"
Environment="OLLAMA_LLM_LIBRARY=rocm"
Environment="TORCH_BLAS_PREFER_HIPBLASLT=1"

---

total duration:       23.206342987s
load duration:        10.715541252s
prompt eval count:    13 token(s)
prompt eval duration: 77.061094ms
prompt eval rate:     168.70 tokens/s
eval count:           542 token(s)
eval duration:        12.26131007s
eval rate:            44.20 tokens/s

---

total duration:       20.519295811s
load duration:        6.97770851s
prompt eval count:    19 token(s)
prompt eval duration: 81.412507ms
prompt eval rate:     233.38 tokens/s
eval count:           693 token(s)
eval duration:        13.174333915s
eval rate:            52.60 tokens/s

---

total duration:       12.825772198s
load duration:        213.465616ms
prompt eval count:    4 token(s)
prompt eval duration: 52.509ms
prompt eval rate:     76.18 tokens/s
eval count:           372 token(s)
eval duration:        12.531793s
eval rate:            29.68 tokens/s

---

total duration:       41.32533364s
load duration:        16.666850597s
prompt eval count:    19 token(s)
prompt eval duration: 158.064ms
prompt eval rate:     120.20 tokens/s
eval count:           753 token(s)
eval duration:        24.498851s
eval rate:            30.74 tokens/s

---