Misc. bug: M1 Pro 8gb: the CLIP graph uses unsupported operators by the backend #21941

warmup: flash attention is enabled
warmup: *****************************************************************
warmup: WARNING: the CLIP graph uses unsupported operators by the backend
warmup:          the performance will be suboptimal                      
warmup:          list of unsupported ops (backend=MTL0):
warmup:             ROLL: type = f32, ne = [128 8 1512 1]
warmup:             ROLL: type = f32, ne = [128 8 1512 1]
warmup:             ROLL: type = f32, ne = [754 1024 1 1]
warmup:             ROLL: type = f32, ne = [128 8 1512 1]
warmup:             ROLL: type = f32, ne = [128 8 1512 1]
warmup:             ROLL: type = f32, ne = [754 1024 1 1]
warmup:             ROLL: type = f32, ne = [128 8 1512 1]
warmup:             ROLL: type = f32, ne = [128 8 1512 1]
warmup:             ROLL: type = f32, ne = [754 1024 1 1]
warmup:             ROLL: type = f32, ne = [128 8 1512 1]
warmup:             ROLL: type = f32, ne = [128 8 1512 1]
warmup:             ROLL: type = f32, ne = [754 1024 1 1]
warmup:             ROLL: type = f32, ne = [128 8 1512 1]
warmup:             ROLL: type = f32, ne = [128 8 1512 1]
warmup:             ROLL: type = f32, ne = [754 1024 1 1]
warmup:             ROLL: type = f32, ne = [128 8 1512 1]
warmup:             ROLL: type = f32, ne = [128 8 1512 1]
warmup:             ROLL: type = f32, ne = [754 1024 1 1]
warmup:             ROLL: type = f32, ne = [128 8 1512 1]
warmup:             ROLL: type = f32, ne = [128 8 1512 1]
warmup:             ROLL: type = f32, ne = [754 1024 1 1]
warmup:             ROLL: type = f32, ne = [128 8 1512 1]
warmup:             ROLL: type = f32, ne = [128 8 1512 1]
warmup:             ROLL: type = f32, ne = [754 1024 1 1]
warmup:             ROLL: type = f32, ne = [128 8 1512 1]
warmup:             ROLL: type = f32, ne = [128 8 1512 1]
warmup:             ROLL: type = f32, ne = [754 1024 1 1]
warmup:             ROLL: type = f32, ne = [128 8 1512 1]
warmup:             ROLL: type = f32, ne = [128 8 1512 1]
warmup:             ROLL: type = f32, ne = [754 1024 1 1]
warmup:             ROLL: type = f32, ne = [128 8 1512 1]
warmup:             ROLL: type = f32, ne = [128 8 1512 1]
warmup:             ROLL: type = f32, ne = [754 1024 1 1]
warmup:             ROLL: type = f32, ne = [128 8 1512 1]
warmup:             ROLL: type = f32, ne = [128 8 1512 1]
warmup:             ROLL: type = f32, ne = [754 1024 1 1]
warmup: flash attention is enabled
warmup: please report this on github as an issue
warmup: ref: https://github.com/ggml-org/llama.cpp/pull/16837#issuecomment-3461676118
warmup: *****************************************************************

llama-server -m ./models/gemma-4-E2B-it-Q4_K_M.gguf \
                --ctx-size 4096 \
                --fit-target 512 \
                --flash-attn on \
                --cache-type-k q4_0 \
                --cache-type-v q4_0 \
                --parallel 1 \
                --temp 0.3 \
                --top-k 64 \
                --top-p 0.95 \
                --frequency-penalty 1.0 \
                --repeat-penalty 1.1 \
                --mmproj ./models/mmproj-BF16.gguf \
                --image-min-tokens 300 \
                --image-max-tokens 512 \
                --host 0.0.0.0 \
                --port 8080 --webui-mcp-proxy --webui-config-file ./uiConfig.json