Select hybrid expert placement from benchmark

scripts/serve-frank-small.sh

@@ -7,8 +7,9 @@
 # 3B-active MoE-dense hybrid with the MTP speculative head bundled in the GGUF.
 #
 # Expert offload: the expert (FFN exps) tensors selected by the layer-bounded
-# --override-tensor pattern are pinned to CPU. The final output-adjacent
-# blocks stay on CUDA with all non-expert tensors, including the MTP head.
+# --override-tensor pattern are pinned to CPU. Selected full-attention and
+# final output-adjacent blocks stay on CUDA with all non-expert tensors,
+# including the MTP head.
 #
 # Spec: MoE model with a bundled MTP head -> native draft-mtp (the head is
 # auto-detected from the target model; b10121 creates the MTP draft context
@@ -49,7 +50,7 @@ fi
 # Expert offload: --override-tensor regex pinned to the CPU buft. Same pattern
 # the internal --cpu-moe / --n-cpu-moe use; explicit here so we only move the
 # MoE expert weights and leave every other tensor on the GPU.
-: "${OVERRIDE_EXPERT:=blk\.(0|1|2|3|4|5|6|7|8|9|10|11|12|13|14|15|16|17|18|19|20|21|22|23|24|25|26|27|28|29|30|31|32)\.ffn_(up|down|gate|gate_up)_(ch|)exps}"
+: "${OVERRIDE_EXPERT:=blk\.(0|1|2|3|4|5|6|8|9|10|11|12|13|14|15|16|17|18|19|20|21|22|23|24|25|26|28|29|30|32)\.ffn_(up|down|gate|gate_up)_(ch|)exps}"
 : "${OVERRIDE_BUFT:=CPU}"
 : "${N_GPU_LAYERS:=all}"
 : "${BATCH_SIZE:=4096}"