2026-09-22T15:53:26.728101Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:compile_runtime: close time.busy=300µs time.idle=5.04µs 2026-09-22T15:53:45.296111Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:construct_mid:gemm_candidates: close time.busy=18.6s time.idle=4.35µs 2026-09-22T15:54:24.330913Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:construct_mid:gemm_candidates: close time.busy=39.0s time.idle=4.35µs 2026-09-22T15:54:26.278276Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:construct_mid: close time.busy=59.6s time.idle=2.49µs 2026-09-22T15:54:26.324910Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}: wrote planner memory profile path=artifacts/kernel-cost-mlp-final-20260922/fp8-b1-n1-default/memory/baseline-1179115-0000.json scope="baseline" peak_step=2 effective_peak_bytes=154624 2026-09-22T15:54:27.114862Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:group_exchanges: close time.busy=125µs time.idle=3.16µs 2026-09-22T15:54:27.116382Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:merge_copies: close time.busy=1.32ms time.idle=1.06µs 2026-09-22T15:54:27.144692Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:eliminate: eliminated padding clears using kernel access contracts removed=0 2026-09-22T15:54:27.146327Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:eliminate: close time.busy=29.9ms time.idle=1.17µs 2026-09-22T15:54:27.192871Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:place_provisional_storage: close time.busy=45.9ms time.idle=2.15µs 2026-09-22T15:54:27.494562Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: prepared large physical exchange phase phase=4 transfers=3096 maximum_identical_destinations=1 2026-09-22T15:54:27.494595Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: optimized physical exchange schedule phase=4 initial_horizon=7943 selected_horizon=7943 endpoint_lower_bound=7680 lower_bound_gap=263 selected_kind="full-duplex" neighborhood_improvements=0 2026-09-22T15:54:27.551612Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: retrying exchange schedule with incremental instruction-alignment validation transfers=50933 2026-09-22T15:54:28.028913Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: retrying exchange schedule with incremental instruction-alignment validation transfers=15488 2026-09-22T15:54:28.066968Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: finished incremental instruction-alignment validation transfers=50933 elapsed_ms=515 success=true 2026-09-22T15:54:28.352713Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: retrying exchange schedule with incremental instruction-alignment validation transfers=21111 2026-09-22T15:54:28.543661Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: retrying exchange schedule with incremental instruction-alignment validation transfers=131766 2026-09-22T15:54:28.571539Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: prepared large physical exchange phase phase=6 transfers=50933 maximum_identical_destinations=1 2026-09-22T15:54:28.571565Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: optimized physical exchange schedule phase=6 initial_horizon=474 selected_horizon=474 endpoint_lower_bound=288 lower_bound_gap=186 selected_kind="full-duplex" neighborhood_improvements=0 2026-09-22T15:54:28.965719Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: retrying exchange schedule with incremental instruction-alignment validation transfers=196058 2026-09-22T15:54:29.419649Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: finished incremental instruction-alignment validation transfers=15488 elapsed_ms=1390 success=true 2026-09-22T15:54:29.843420Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: finished incremental instruction-alignment validation transfers=21111 elapsed_ms=1490 success=true 2026-09-22T15:54:30.659348Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: finished incremental instruction-alignment validation transfers=131766 elapsed_ms=2115 success=true 2026-09-22T15:54:30.895558Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: retrying exchange schedule with incremental instruction-alignment validation transfers=15488 2026-09-22T15:54:31.993584Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: finished incremental instruction-alignment validation transfers=196058 elapsed_ms=3027 success=true 2026-09-22T15:54:32.334322Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: finished incremental instruction-alignment validation transfers=15488 elapsed_ms=1438 success=true 2026-09-22T15:54:32.568758Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: compared ordinary and paired exchange schedules phase=3 candidates=20310 ordinary_horizon=7868 paired_horizon=8729 use_paired=false 2026-09-22T15:54:32.626600Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: prepared large physical exchange phase phase=3 transfers=21111 maximum_identical_destinations=1 2026-09-22T15:54:32.626623Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: optimized physical exchange schedule phase=3 initial_horizon=7868 selected_horizon=7868 endpoint_lower_bound=6480 lower_bound_gap=1388 selected_kind="full-duplex" neighborhood_improvements=0 2026-09-22T15:54:32.922325Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: prepared large physical exchange phase phase=1 transfers=131766 maximum_identical_destinations=1 2026-09-22T15:54:32.922350Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: optimized physical exchange schedule phase=1 initial_horizon=8448 selected_horizon=8448 endpoint_lower_bound=6912 lower_bound_gap=1536 selected_kind="full-duplex" neighborhood_improvements=0 2026-09-22T15:54:33.090066Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: compared ordinary and paired exchange schedules phase=0 candidates=23 ordinary_horizon=15608 paired_horizon=15657 use_paired=false 2026-09-22T15:54:33.164787Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: prepared large physical exchange phase phase=0 transfers=15488 maximum_identical_destinations=1 2026-09-22T15:54:33.164806Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: optimized physical exchange schedule phase=0 initial_horizon=15608 selected_horizon=15608 endpoint_lower_bound=14832 lower_bound_gap=776 selected_kind="full-duplex" neighborhood_improvements=0 2026-09-22T15:54:35.358933Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: prepared large physical exchange phase phase=2 transfers=196058 maximum_identical_destinations=1 2026-09-22T15:54:35.358961Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: optimized physical exchange schedule phase=2 initial_horizon=1705 selected_horizon=1705 endpoint_lower_bound=1152 lower_bound_gap=553 selected_kind="full-duplex" neighborhood_improvements=0 2026-09-22T15:54:37.257555Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: retrying exchange schedule with incremental instruction-alignment validation transfers=454896 2026-09-22T15:54:50.271165Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: finished incremental instruction-alignment validation transfers=454896 elapsed_ms=13013 success=true 2026-09-22T15:55:01.974397Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: prepared large physical exchange phase phase=5 transfers=454896 maximum_identical_destinations=1 2026-09-22T15:55:01.974452Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: optimized physical exchange schedule phase=5 initial_horizon=9273 selected_horizon=9273 endpoint_lower_bound=7488 lower_bound_gap=1785 selected_kind="full-duplex" neighborhood_improvements=0 2026-09-22T15:55:02.832424Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:schedule_provisional_exchanges: close time.busy=35.6s time.idle=3.42µs 2026-09-22T15:55:02.885142Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:plan_kernels: close time.busy=52.6ms time.idle=3.59µs 2026-09-22T15:55:02.886683Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:compile_kernels: close time.busy=1.52ms time.idle=1.76µs 2026-09-22T15:55:02.886938Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:link_runtime: close time.busy=168µs time.idle=882ns 2026-09-22T15:55:03.075660Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}: exchange table and repeat patch storage exchange_table_bytes=12900 maximum_uncompressed_repeat_patch_bytes=0 maximum_elided_arithmetic_patch_bytes=0 2026-09-22T15:55:03.394688Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:size_tile_code: close time.busy=25.0ms time.idle=5.07µs 2026-09-22T15:55:03.394746Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}: placing generated tile programs linked_end=318984 host_code_base=376832 host_code_bytes=620 generated_code_bytes=1168 exchange_table_bytes=12900 executable_ranges=[(377520, 524288)] 2026-09-22T15:55:03.394797Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}: allocated package support memory linked_end=318984 profile_bytes=96 exchange_table_bytes=12900 host_code_bytes=620 host_data_bytes=404 generated_code_bytes=1168 code_address=377520 available_ranges=[(327680, 360448), (362612, 376832), (409600, 949168)] 2026-09-22T15:55:03.468756Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:place_storage: close time.busy=62.2ms time.idle=2.70µs 2026-09-22T15:55:03.583806Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: reused exchange optimization after validating relocated rows phase=4 2026-09-22T15:55:03.587432Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: prepared large physical exchange phase phase=4 transfers=3096 maximum_identical_destinations=1 2026-09-22T15:55:03.587442Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: optimized physical exchange schedule phase=4 initial_horizon=7943 selected_horizon=7943 endpoint_lower_bound=7680 lower_bound_gap=263 selected_kind="reused" neighborhood_improvements=0 2026-09-22T15:55:04.406524Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: retrying exchange schedule with incremental instruction-alignment validation transfers=50933 2026-09-22T15:55:04.890163Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: finished incremental instruction-alignment validation transfers=50933 elapsed_ms=483 success=true 2026-09-22T15:55:05.072921Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: reused exchange optimization after validating relocated rows phase=3 2026-09-22T15:55:05.096336Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: prepared large physical exchange phase phase=3 transfers=21111 maximum_identical_destinations=1 2026-09-22T15:55:05.096358Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: optimized physical exchange schedule phase=3 initial_horizon=7868 selected_horizon=7868 endpoint_lower_bound=6480 lower_bound_gap=1388 selected_kind="reused" neighborhood_improvements=0 2026-09-22T15:55:05.418646Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: prepared large physical exchange phase phase=6 transfers=50933 maximum_identical_destinations=1 2026-09-22T15:55:05.418677Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: optimized physical exchange schedule phase=6 initial_horizon=474 selected_horizon=474 endpoint_lower_bound=288 lower_bound_gap=186 selected_kind="full-duplex" neighborhood_improvements=0 2026-09-22T15:55:05.439947Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: reused exchange optimization after validating relocated rows phase=1 2026-09-22T15:55:05.469668Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: prepared large physical exchange phase phase=1 transfers=131766 maximum_identical_destinations=1 2026-09-22T15:55:05.469683Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: optimized physical exchange schedule phase=1 initial_horizon=8448 selected_horizon=8448 endpoint_lower_bound=6912 lower_bound_gap=1536 selected_kind="reused" neighborhood_improvements=0 2026-09-22T15:55:05.910146Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: reused exchange optimization after validating relocated rows phase=0 2026-09-22T15:55:05.960469Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: prepared large physical exchange phase phase=0 transfers=15488 maximum_identical_destinations=1 2026-09-22T15:55:05.960492Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: optimized physical exchange schedule phase=0 initial_horizon=15608 selected_horizon=15608 endpoint_lower_bound=14832 lower_bound_gap=776 selected_kind="reused" neighborhood_improvements=0 2026-09-22T15:55:06.981411Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: reused exchange optimization after validating relocated rows phase=2 2026-09-22T15:55:07.025981Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: prepared large physical exchange phase phase=2 transfers=196058 maximum_identical_destinations=1 2026-09-22T15:55:07.026001Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: optimized physical exchange schedule phase=2 initial_horizon=1705 selected_horizon=1705 endpoint_lower_bound=1152 lower_bound_gap=553 selected_kind="reused" neighborhood_improvements=0 2026-09-22T15:55:12.338325Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: reused exchange optimization after validating relocated rows phase=5 2026-09-22T15:55:12.455263Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: prepared large physical exchange phase phase=5 transfers=454896 maximum_identical_destinations=1 2026-09-22T15:55:12.455281Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: optimized physical exchange schedule phase=5 initial_horizon=9273 selected_horizon=9273 endpoint_lower_bound=7488 lower_bound_gap=1785 selected_kind="reused" neighborhood_improvements=0 2026-09-22T15:55:13.030144Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:lower_exchanges: close time.busy=9.56s time.idle=5.78µs 2026-09-22T15:55:13.038062Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}: costed final placed program final_cycles=205271 final_exchange=55519 2026-09-22T15:55:13.332868Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:prepare_tile_code: close time.busy=48.6ms time.idle=5.37µs 2026-09-22T15:55:13.352356Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:emit_tile_code: close time.busy=19.4ms time.idle=2.44µs 2026-09-22T15:55:13.501996Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}:build_tile_images: close time.busy=144ms time.idle=1.40µs 2026-09-22T15:55:13.687324Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}: wrote exact tile placement profile path=artifacts/kernel-cost-mlp-final-20260922/fp8-b1-n1-default/memory/placement-1179115.html 2026-09-22T15:55:13.779961Z INFO ipu_codegen.package.build{tile_count=1472 operations=3}: close time.busy=107s time.idle=5.24µs 2026-09-22T15:55:13.784772Z INFO writing application package tiles=1472 2026-09-22T15:55:13.936363Z INFO application package written 2026-09-22T15:55:13.938754Z INFO reading application package 2026-09-22T15:55:14.107438Z INFO application package read tiles=1472 compiler=0.1.0 2026-09-22T15:55:14.111490Z INFO waiting for hardware ownership path=/tmp/ipu-stack-device.lock 2026-09-22T15:55:14.111502Z INFO acquired hardware ownership path=/tmp/ipu-stack-device.lock 2026-09-22T15:55:14.111507Z INFO opening IPU device path="/dev/ipu0" 2026-09-22T15:55:14.111536Z INFO IPU device attached path=/dev/ipu0 2026-09-22T15:55:14.111539Z INFO initializing IPU device 2026-09-22T15:55:14.119567Z INFO IPU device initialized 2026-09-22T15:55:14.119611Z INFO replaying device configuration register_writes=2003 2026-09-22T15:55:14.125259Z INFO loading application tile_count=1472 final_mark=23 2026-09-22T15:55:14.312525Z INFO application loaded and initial synchronization completed tile_count=1472 2026-09-22T15:55:15.180110Z INFO prepared reference inputs elapsed_ms=864 2026-09-22T15:55:15.512238Z INFO configured host reference BLAS threads=4 referencePrecision=device 2026-09-22T15:55:15.940857Z INFO evaluated host reference elapsed_ms=760 2026-09-22T15:55:15.945686Z INFO starting host exchange session startup_mark=23 2026-09-22T15:55:15.945712Z INFO attaching host exchange pages pages=2 2026-09-22T15:55:15.946482Z INFO host exchange session started 2026-09-22T15:55:15.946519Z INFO invoking host exchange call call="initialize" command=0 phases=4 2026-09-22T15:55:15.948574Z INFO invoking host exchange call call="run" command=0 phases=13 referenceInference=1 numericalTest=PASS profile=artifacts/kernel-cost-mlp-final-20260922/fp8-b1-n1-default/profile.capnp tiles=1472 samples=26676 workload=siglip-mlp-f16-b1-t729-d1152-h4304-n1 benchmark=siglip-mlp-f16 batch=1 tokens=729 rows=729 dimension=1152 hiddenDimension=4304 blocks=1 biases=false activeTiles=1469 executionTiles=1472 inputBytes=839808 weightBytes=9934848 cycles=204006 minimumTileCycles=199980 deviceMicroseconds=136.004 effectiveGemmTflops=106.307 peakTflops=282.624 efficiencyPercent=37.61 maximumAbsoluteError=0.005554 package=artifacts/kernel-cost-mlp-final-20260922/fp8-b1-n1-default/model.ipuexe tiles=1472 hardwareTest=PASS