Files
aituner/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v1/moe.csv

52 KiB

1time_stats.moe_gating_linear.mintime_stats.moe_gating_linear.maxtime_stats.moe_gating_linear.meantime_stats.moe_gating_linear.mediantime_stats.moe_gating_linear.stdtime_stats.moe_gating_routing_topk.mintime_stats.moe_gating_routing_topk.maxtime_stats.moe_gating_routing_topk.meantime_stats.moe_gating_routing_topk.mediantime_stats.moe_gating_routing_topk.stdtime_stats.moe_shuffling.mintime_stats.moe_shuffling.maxtime_stats.moe_shuffling.meantime_stats.moe_shuffling.mediantime_stats.moe_shuffling.stdtime_stats.moe_grouped_gemm.mintime_stats.moe_grouped_gemm.maxtime_stats.moe_grouped_gemm.meantime_stats.moe_grouped_gemm.mediantime_stats.moe_grouped_gemm.stdnum_tokensnum_expertsnum_experts_per_deviceexpert_parallel_sizerouting_runtime_pathrouting_assignment_policyrouting_weight_policyrouting_uses_router_logitsgating_runtime_contextgating_runtime_context_implrouter_topkhidden_dimexpert_hidden_dimuse_gatednum_tensor_parallel_workerstotal_routed_tokensmodel_expansion_ratiotokens_per_expert_avgtokens_to_experts_ratioexpert_utilizationmin_load_ratioload_imbalance_cvmax_load_ratioload_entropyload_gini_coefficientload_distributionseedmoe_grouped_gemm_backendmeasurement_typeprofiling_precisionmodel_archquant_signaturerouter_median_nonadditivity_ratioprojection_policy
20.025024000555276870.060927998274564740.0338399996981024740.0286720003932714460.0103152553437098180.0193600002676248550.03529600054025650.0234240002930164340.0220640003681182860.00388981022891945720.00.00.00.00.00.339264005422592160.4050239920616150.367804798483848560.365071997046470640.0169050764447477911281281standard_fused_topklogit_topksoftmax_renormTruestandalone_legacyvllm020_replicated_linear82048768True180.3750.06250.06250.06250.03.87298334620741716.03.00.9375uniform_random_logits20260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone1.0233364439829928measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
30.0252800006419420240.051327999681234360.0306415998376905930.0273439995944499970.0075624092327633040.0201600007712841030.0529600009322166440.0241456001996994030.0214240001514554020.0074501986455999850.00.00.00.00.01.19430398941040041.2867840528488161.2283840060234071.22732800245285030.0283254724238126381281281standard_fused_topklogit_topksoftmax_renormTruestandalone_legacyvllm020_replicated_linear82048768True1640.3750.50.50.39843750.01.3462912017836264.05.593750.661865234375uniform_random_logits20260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9806430689981738measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
40.0248319990932941440.0465600006282329560.0303712001070380220.027631999924778940.0058784872050286020.0203200001269578930.045600000768899920.026019200123846530.022704000584781170.007511299659067990.00.00.00.00.01.6797440052032471.7661440372467041.70958081483840951.70156800746917720.02438921262998535161281281standard_fused_topklogit_topksoftmax_renormTruestandalone_legacyvllm020_replicated_linear82048768True11280.3751.01.00.6250.01.0155048005794955.06.155164332129550.529052734375uniform_random_logits20260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9103623678483975measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
50.0242880005389451980.0493759997189044950.030860800016671420.02673599962145090.00708645313849972250.0204799994826316830.0309120006859302520.022747199889272450.0213599996641278270.00299668136506725050.00.00.00.00.02.15766406059265142.29216003417968752.20978240966796862.1889441013336180.045572321842012986321281281standard_fused_topklogit_topksoftmax_renormTruestandalone_legacyvllm020_replicated_linear82048768True12560.3752.02.00.8750.00.63430572281826372.56.643707484446390.35369873046875uniform_random_logits20260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9610778571819444measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
60.023935999721288680.043423999100923540.0293807999230921260.0266880001872777940.00573747820005265740.0200319997966289520.0366079993546009060.0224879999645054350.0209119999781250950.00381352030621032350.00.00.00.00.02.4223680496215822.51308798789978032.45166721343994132.4341599941253660.03278900287381846641281281standard_fused_topklogit_topksoftmax_renormTruestandalone_legacyvllm020_replicated_linear82048768True15120.3754.04.00.9843750.00.49212549212573822.256.8171900423447690.272369384765625uniform_random_logits20260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9952941013961014measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
70.0254079997539520260.055103998631238940.0314303997904062240.027983999811112880.0073350502739827250.0204799994826316830.035615999251604080.022758399881422520.0215519992634654050.0035457183651658110.00.00.00.00.02.22176003456115722.2895998954772952.25713279247283932.2633759975433350.0216604160894494881281281281standard_fused_topklogit_topksoftmax_renormTruestandalone_legacyvllm020_replicated_linear82048768True110240.3758.08.01.00.1250.34868615006908431.8756.9081923101839970.197662353515625uniform_random_logits20260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9273256282883522measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
80.02441599965095520.053952001035213470.031889599841088060.0260319998487830160.009439959273874830.020511999726295470.0366079993546009060.0232479997910559170.021471999585628510.0039106238370696480.00.00.00.00.02.186687946319582.3180799484252932.22180161476135262.2110879421234130.0353808972131353162561281281standard_fused_topklogit_topksoftmax_renormTruestandalone_legacyvllm020_replicated_linear82048768True120480.37516.016.01.00.43750.25255046680069711.8756.9533477430530170.1410369873046875uniform_random_logits20260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone1.0380599882396606measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
90.0245120003819465640.046208001673221590.028846400231122970.0263200001791119580.0055165575833559250.020543999969959260.038463998585939410.0234016000293195240.0212639998644590380.00447429441982653740.00.00.00.00.02.22918391227722172.39296007156372072.29080963134765622.28046405315399170.044793489247862215121281281standard_fused_topklogit_topksoftmax_renormTruestandalone_legacyvllm020_replicated_linear82048768True140960.37532.032.01.00.656250.157659656801645041.56256.982298487282050.08779525756835938uniform_random_logits20260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9569603278386984measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
100.024127999320626260.069407999515533450.0308176004327833650.0268000001087784770.0100470851876529390.0201280005276203160.0440960004925727840.0246064000762999040.0223040003329515460.006228571668237140.00.00.00.00.02.06787204742431642.12976002693176272.08371198177337652.07791996002197270.01788004435798673510241281281standard_fused_topklogit_topksoftmax_renormTruestandalone_legacyvllm020_replicated_linear82048768True181920.37564.064.01.00.6250.121690816355040741.31256.98920296623563250.06879425048828125uniform_random_logits20260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9524274993623046measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
110.028319999575614930.0434879995882511140.0317440001294016850.029919999651610850.0039734158364289090.020703999325633050.0293439999222755430.0228864000178873530.0217439997941255570.00268730650450888730.00.00.00.00.02.9160320758819583.08195209503173832.98052480220794672.96563196182250980.0548219579901957220481281281standard_fused_topklogit_topksoftmax_renormTruestandalone_legacyvllm020_replicated_linear82048768True1163840.375128.0128.01.00.7968750.079354341476887511.17968756.99542979647503050.044734954833984375uniform_random_logits20260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.8971818172100244measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
120.038304001092910770.062688000500202180.0433375997468829140.0405119992792606350.0058230162479461160.0231359992176294330.037471998482942580.0252063999883830530.023935999721288680.0032713812562311610.00.00.00.00.04.4215998649597174.5353598594665534.4862943172454834.4970560073852540.03699024378754934440961281281standard_fused_topklogit_topksoftmax_renormTruestandalone_legacyvllm020_replicated_linear82048768True1327680.375256.0256.01.00.82031250.0608493264831030461.175781256.99731883758596850.033740997314453125uniform_random_logits20260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.8113207890716184measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
130.056607998907566070.079328000545501710.062499200180172920.060399999842047690.0056364616914808450.029567999765276910.037471998482942580.0311263998970389350.0302879996597766880.0021575710066315410.00.00.00.00.07.3025918006896977.4027519226074227.3547583103179937.34643197059631350.03214266240033556681921281281standard_fused_topklogit_topksoftmax_renormTruestandalone_legacyvllm020_replicated_linear82048768True1655360.375512.0512.01.00.8906250.04123230872663411.089843756.9987724331855780.02334284782409668uniform_random_logits20260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.883909666885606measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
140.025024000555276870.060927998274564740.0338399996981024740.0286720003932714460.0103152553437098180.0193600002676248550.03529600054025650.0234240002930164340.0220640003681182860.00388981022891945720.00.00.00.00.00.352800011634826660.396928012371063230.376627203822135940.371968001127243040.01340131805066530411281281standard_fused_topkfixed_hotset8softmax_renormFalsestandalone_legacyvllm020_replicated_linear82048768True180.3750.06250.06250.06250.03.87298334620741716.03.00.9375hotset820260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone1.0233364439829928measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
150.0252800006419420240.051327999681234360.0306415998376905930.0273439995944499970.0075624092327633040.0201600007712841030.0529600009322166440.0241456001996994030.0214240001514554020.0074501986455999850.00.00.00.00.00.46924799680709840.55238401889801030.51347520351409910.51006400585174560.0229143346413578481281281standard_fused_topkfixed_hotset8softmax_renormFalsestandalone_legacyvllm020_replicated_linear82048768True1640.3750.50.50.06250.03.87298334620741716.03.00.9375hotset820260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9806430689981738measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
160.0248319990932941440.0465600006282329560.0303712001070380220.027631999924778940.0058784872050286020.0203200001269578930.045600000768899920.026019200123846530.022704000584781170.007511299659067990.00.00.00.00.00.346527993679046630.41190400719642640.37894719839096070.385504007339477540.02073945105803335161281281standard_fused_topkfixed_hotset8softmax_renormFalsestandalone_legacyvllm020_replicated_linear82048768True11280.3751.01.00.06250.03.87298334620741716.03.00.9375hotset820260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9103623678483975measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
170.0242880005389451980.0493759997189044950.030860800016671420.02673599962145090.00708645313849972250.0204799994826316830.0309120006859302520.022747199889272450.0213599996641278270.00299668136506725050.00.00.00.00.00.314624011516571040.74569600820541380.386172804236412040.345456004142761230.12230201266253077321281281standard_fused_topkfixed_hotset8softmax_renormFalsestandalone_legacyvllm020_replicated_linear82048768True12560.3752.02.00.06250.03.87298334620741716.03.00.9375hotset820260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9610778571819444measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
180.023935999721288680.043423999100923540.0293807999230921260.0266880001872777940.00573747820005265740.0200319997966289520.0366079993546009060.0224879999645054350.0209119999781250950.00381352030621032350.00.00.00.00.00.325215995311737060.4198719859123230.363251197338104240.349680006504058840.03161798848223672641281281standard_fused_topkfixed_hotset8softmax_renormFalsestandalone_legacyvllm020_replicated_linear82048768True15120.3754.04.00.06250.03.87298334620741716.03.00.9375hotset820260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9952941013961014measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
190.0254079997539520260.055103998631238940.0314303997904062240.027983999811112880.0073350502739827250.0204799994826316830.035615999251604080.022758399881422520.0215519992634654050.0035457183651658110.00.00.00.00.00.2897920012474060.46633601188659670.40918399989604950.416559994220733640.04460019865066151281281281standard_fused_topkfixed_hotset8softmax_renormFalsestandalone_legacyvllm020_replicated_linear82048768True110240.3758.08.00.06250.03.87298334620741716.03.00.9375hotset820260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9273256282883522measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
200.02441599965095520.053952001035213470.031889599841088060.0260319998487830160.009439959273874830.020511999726295470.0366079993546009060.0232479997910559170.021471999585628510.0039106238370696480.00.00.00.00.00.37612798810005190.44163200259208680.406863999366760270.405407994985580440.022577787698996452561281281standard_fused_topkfixed_hotset8softmax_renormFalsestandalone_legacyvllm020_replicated_linear82048768True120480.37516.016.00.06250.03.87298334620741716.03.00.9375hotset820260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone1.0380599882396606measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
210.0245120003819465640.046208001673221590.028846400231122970.0263200001791119580.0055165575833559250.020543999969959260.038463998585939410.0234016000293195240.0212639998644590380.00447429441982653740.00.00.00.00.00.71724802255630490.86630398035049440.77238079905509950.75918400287628170.041647723794512425121281281standard_fused_topkfixed_hotset8softmax_renormFalsestandalone_legacyvllm020_replicated_linear82048768True140960.37532.032.00.06250.03.87298334620741716.03.00.9375hotset820260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9569603278386984measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
220.024127999320626260.069407999515533450.0308176004327833650.0268000001087784770.0100470851876529390.0201280005276203160.0440960004925727840.0246064000762999040.0223040003329515460.006228571668237140.00.00.00.00.01.01955199241638181.22166395187377931.12538881301879881.14536005258560180.0654800532224359410241281281standard_fused_topkfixed_hotset8softmax_renormFalsestandalone_legacyvllm020_replicated_linear82048768True181920.37564.064.00.06250.03.87298334620741716.03.00.9375hotset820260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9524274993623046measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
230.028319999575614930.0434879995882511140.0317440001294016850.029919999651610850.0039734158364289090.020703999325633050.0293439999222755430.0228864000178873530.0217439997941255570.00268730650450888730.00.00.00.00.01.74905598163604741.96448004245758061.85290240049362191.8143039941787720.0804256561732728820481281281standard_fused_topkfixed_hotset8softmax_renormFalsestandalone_legacyvllm020_replicated_linear82048768True1163840.375128.0128.00.06250.03.87298334620741716.03.00.9375hotset820260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.8971818172100244measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
240.038304001092910770.062688000500202180.0433375997468829140.0405119992792606350.0058230162479461160.0231359992176294330.037471998482942580.0252063999883830530.023935999721288680.0032713812562311610.00.00.00.00.03.24793601036071783.3852798938751223.2960704088211063.28009605407714840.0452552902644204640961281281standard_fused_topkfixed_hotset8softmax_renormFalsestandalone_legacyvllm020_replicated_linear82048768True1327680.375256.0256.00.06250.03.87298334620741716.03.00.9375hotset820260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.8113207890716184measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
250.056607998907566070.079328000545501710.062499200180172920.060399999842047690.0056364616914808450.029567999765276910.037471998482942580.0311263998970389350.0302879996597766880.0021575710066315410.00.00.00.00.06.3447999954223636.5178561210632326.4644384384155276.4786238670349120.0511667444314509881921281281standard_fused_topkfixed_hotset8softmax_renormFalsestandalone_legacyvllm020_replicated_linear82048768True1655360.375512.0512.00.06250.03.87298334620741716.03.00.9375hotset820260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.883909666885606measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
260.025024000555276870.060927998274564740.0338399996981024740.0286720003932714460.0103152553437098180.0193600002676248550.03529600054025650.0234240002930164340.0220640003681182860.00388981022891945720.00.00.00.00.00.26480001211166380.3254399895668030.288528001308441140.283903986215591430.0193377837763507711281281standard_fused_topklogit_topksoftmax_renormTruestandalone_legacyvllm020_replicated_linear82048768True280.3750.06250.06250.06250.03.87298334620741716.03.00.9375uniform_random_logits20260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone1.0233364439829928measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
270.0252800006419420240.051327999681234360.0306415998376905930.0273439995944499970.0075624092327633040.0201600007712841030.0529600009322166440.0241456001996994030.0214240001514554020.0074501986455999850.00.00.00.00.00.73478400707244870.8624960184097290.77693440318107610.7692160010337830.0329048532879628581281281standard_fused_topklogit_topksoftmax_renormTruestandalone_legacyvllm020_replicated_linear82048768True2640.3750.50.50.4218750.01.3462912017836266.05.6521146483360870.636962890625uniform_random_logits20260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9806430689981738measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
280.0248319990932941440.0465600006282329560.0303712001070380220.027631999924778940.0058784872050286020.0203200001269578930.045600000768899920.026019200123846530.022704000584781170.007511299659067990.00.00.00.00.00.91983997821807860.96460801362991330.94120639562606810.94118398427963260.014939365085478117161281281standard_fused_topklogit_topksoftmax_renormTruestandalone_legacyvllm020_replicated_linear82048768True21280.3751.01.00.57031250.01.1180339887498955.06.0086417735188980.580810546875uniform_random_logits20260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9103623678483975measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
290.0242880005389451980.0493759997189044950.030860800016671420.02673599962145090.00708645313849972250.0204799994826316830.0309120006859302520.022747199889272450.0213599996641278270.00299668136506725050.00.00.00.00.01.27968001365661621.34841597080230711.30069760084152231.29291200637817380.020807998177176254321281281standard_fused_topklogit_topksoftmax_renormTruestandalone_legacyvllm020_replicated_linear82048768True22560.3752.02.00.88281250.00.69597054535375273.06.608728506155830.38055419921875uniform_random_logits20260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9610778571819444measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
300.023935999721288680.043423999100923540.0293807999230921260.0266880001872777940.00573747820005265740.0200319997966289520.0366079993546009060.0224879999645054350.0209119999781250950.00381352030621032350.00.00.00.00.01.36303997039794921.44304001331329351.39092159271240231.38923197984695430.022335744492366926641281281standard_fused_topklogit_topksoftmax_renormTruestandalone_legacyvllm020_replicated_linear82048768True25120.3754.04.00.9843750.00.52010365553416373.06.7988265091588510.28302001953125uniform_random_logits20260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9952941013961014measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
310.0254079997539520260.055103998631238940.0314303997904062240.027983999811112880.0073350502739827250.0204799994826316830.035615999251604080.022758399881422520.0215519992634654050.0035457183651658110.00.00.00.00.01.279487967491151.39040005207061771.31760638952255251.3094400167465210.0380608878273127751281281281standard_fused_topklogit_topksoftmax_renormTruestandalone_legacyvllm020_replicated_linear82048768True210240.3758.08.01.00.250.35112820397256611.8756.910022663052380.1970977783203125uniform_random_logits20260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9273256282883522measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
320.02441599965095520.053952001035213470.031889599841088060.0260319998487830160.009439959273874830.020511999726295470.0366079993546009060.0232479997910559170.021471999585628510.0039106238370696480.00.00.00.00.01.2644159793853761.31459200382232671.27919999361038221.27534401416778560.0141306052495683322561281281standard_fused_topklogit_topksoftmax_renormTruestandalone_legacyvllm020_replicated_linear82048768True220480.37516.016.01.00.3750.246929384832486051.68756.9554811307752850.13909912109375uniform_random_logits20260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone1.0380599882396606measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
330.0245120003819465640.046208001673221590.028846400231122970.0263200001791119580.0055165575833559250.020543999969959260.038463998585939410.0234016000293195240.0212639998644590380.00447429441982653740.00.00.00.00.01.30819201469421391.3476480245590211.32922559976577761.3299679756164550.0145588636790169335121281281standard_fused_topklogit_topksoftmax_renormTruestandalone_legacyvllm020_replicated_linear82048768True240960.37532.032.01.00.6250.171430533261653831.56256.97866752757540350.09520339965820312uniform_random_logits20260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9569603278386984measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
340.024127999320626260.069407999515533450.0308176004327833650.0268000001087784770.0100470851876529390.0201280005276203160.0440960004925727840.0246064000762999040.0223040003329515460.006228571668237140.00.00.00.00.01.2427519559860231.31120002269744871.27479358911514271.2662079930305480.02109307351769505710241281281standard_fused_topklogit_topksoftmax_renormTruestandalone_legacyvllm020_replicated_linear82048768True281920.37564.064.01.00.781250.110000998752568151.2968756.9913088712136790.062183380126953125uniform_random_logits20260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9524274993623046measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
350.028319999575614930.0434879995882511140.0317440001294016850.029919999651610850.0039734158364289090.020703999325633050.0293439999222755430.0228864000178873530.0217439997941255570.00268730650450888730.00.00.00.00.01.73881602287292481.80777597427368161.7727648019790651.7727040052413940.02105664407728428320481281281standard_fused_topklogit_topksoftmax_renormTruestandalone_legacyvllm020_replicated_linear82048768True2163840.375128.0128.01.00.781250.08646301501976781.17968756.9945525263941390.048796653747558594uniform_random_logits20260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.8971818172100244measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
360.038304001092910770.062688000500202180.0433375997468829140.0405119992792606350.0058230162479461160.0231359992176294330.037471998482942580.0252063999883830530.023935999721288680.0032713812562311610.00.00.00.00.02.65635204315185552.70636796951293952.67850558757781962.67916798591613770.0163996346305294440961281281standard_fused_topklogit_topksoftmax_renormTruestandalone_legacyvllm020_replicated_linear82048768True2327680.375256.0256.01.00.86718750.061276865147219371.160156256.9972915830271460.03497934341430664uniform_random_logits20260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.8113207890716184measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
370.056607998907566070.079328000545501710.062499200180172920.060399999842047690.0056364616914808450.029567999765276910.037471998482942580.0311263998970389350.0302879996597766880.0021575710066315410.00.00.00.00.04.3868799209594734.4522562026977544.41084804534912144.4063038825988770.01976816179193763681921281281standard_fused_topklogit_topksoftmax_renormTruestandalone_legacyvllm020_replicated_linear82048768True2655360.375512.0512.01.00.8847656250.0417237685253241951.11718756.9987464343189340.02298593521118164uniform_random_logits20260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.883909666885606measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
380.025024000555276870.060927998274564740.0338399996981024740.0286720003932714460.0103152553437098180.0193600002676248550.03529600054025650.0234240002930164340.0220640003681182860.00388981022891945720.00.00.00.00.00.242080003023147580.40284800529479980.30415360033512120.2771039903163910.0566072148488158411281281standard_fused_topkfixed_hotset8softmax_renormFalsestandalone_legacyvllm020_replicated_linear82048768True280.3750.06250.06250.06250.03.87298334620741716.03.00.9375hotset820260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone1.0233364439829928measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
390.0252800006419420240.051327999681234360.0306415998376905930.0273439995944499970.0075624092327633040.0201600007712841030.0529600009322166440.0241456001996994030.0214240001514554020.0074501986455999850.00.00.00.00.00.24470399320125580.305023998022079470.264467203617095970.262656003236770630.01674454836443537281281281standard_fused_topkfixed_hotset8softmax_renormFalsestandalone_legacyvllm020_replicated_linear82048768True2640.3750.50.50.06250.03.87298334620741716.03.00.9375hotset820260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9806430689981738measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
400.0248319990932941440.0465600006282329560.0303712001070380220.027631999924778940.0058784872050286020.0203200001269578930.045600000768899920.026019200123846530.022704000584781170.007511299659067990.00.00.00.00.00.23379200696945190.28815999627113340.260748803615570070.2643840014934540.016469850143940968161281281standard_fused_topkfixed_hotset8softmax_renormFalsestandalone_legacyvllm020_replicated_linear82048768True21280.3751.01.00.06250.03.87298334620741716.03.00.9375hotset820260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9103623678483975measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
410.0242880005389451980.0493759997189044950.030860800016671420.02673599962145090.00708645313849972250.0204799994826316830.0309120006859302520.022747199889272450.0213599996641278270.00299668136506725050.00.00.00.00.00.233695998787879940.285919994115829470.254659201204776750.253856003284454350.01593204652619194321281281standard_fused_topkfixed_hotset8softmax_renormFalsestandalone_legacyvllm020_replicated_linear82048768True22560.3752.02.00.06250.03.87298334620741716.03.00.9375hotset820260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9610778571819444measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
420.023935999721288680.043423999100923540.0293807999230921260.0266880001872777940.00573747820005265740.0200319997966289520.0366079993546009060.0224879999645054350.0209119999781250950.00381352030621032350.00.00.00.00.00.22959999740123750.265567988157272340.246742400527000420.24976000189781190.010345732066199003641281281standard_fused_topkfixed_hotset8softmax_renormFalsestandalone_legacyvllm020_replicated_linear82048768True25120.3754.04.00.06250.03.87298334620741716.03.00.9375hotset820260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9952941013961014measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
430.0254079997539520260.055103998631238940.0314303997904062240.027983999811112880.0073350502739827250.0204799994826316830.035615999251604080.022758399881422520.0215519992634654050.0035457183651658110.00.00.00.00.00.217215999960899350.290208011865615840.23942080140113830.23464000225067140.0187473303577685851281281281standard_fused_topkfixed_hotset8softmax_renormFalsestandalone_legacyvllm020_replicated_linear82048768True210240.3758.08.00.06250.03.87298334620741716.03.00.9375hotset820260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9273256282883522measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
440.02441599965095520.053952001035213470.031889599841088060.0260319998487830160.009439959273874830.020511999726295470.0366079993546009060.0232479997910559170.021471999585628510.0039106238370696480.00.00.00.00.00.27177599072456360.3051840066909790.288137599825859070.288095995783805850.011834226005458542561281281standard_fused_topkfixed_hotset8softmax_renormFalsestandalone_legacyvllm020_replicated_linear82048768True220480.37516.016.00.06250.03.87298334620741716.03.00.9375hotset820260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone1.0380599882396606measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
450.0245120003819465640.046208001673221590.028846400231122970.0263200001791119580.0055165575833559250.020543999969959260.038463998585939410.0234016000293195240.0212639998644590380.00447429441982653740.00.00.00.00.00.39177599549293520.437727987766265870.411308795213699340.41315199434757230.0129924736408052275121281281standard_fused_topkfixed_hotset8softmax_renormFalsestandalone_legacyvllm020_replicated_linear82048768True240960.37532.032.00.06250.03.87298334620741716.03.00.9375hotset820260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9569603278386984measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
460.024127999320626260.069407999515533450.0308176004327833650.0268000001087784770.0100470851876529390.0201280005276203160.0440960004925727840.0246064000762999040.0223040003329515460.006228571668237140.00.00.00.00.00.61766397953033450.70099198818206790.6427679955959320.63307198882102970.02407408491993875610241281281standard_fused_topkfixed_hotset8softmax_renormFalsestandalone_legacyvllm020_replicated_linear82048768True281920.37564.064.00.06250.03.87298334620741716.03.00.9375hotset820260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9524274993623046measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
470.028319999575614930.0434879995882511140.0317440001294016850.029919999651610850.0039734158364289090.020703999325633050.0293439999222755430.0228864000178873530.0217439997941255570.00268730650450888730.00.00.00.00.01.08208000659942631.16748797893524171.10343040227891.09774398803710940.023306729298156620481281281standard_fused_topkfixed_hotset8softmax_renormFalsestandalone_legacyvllm020_replicated_linear82048768True2163840.375128.0128.00.06250.03.87298334620741716.03.00.9375hotset820260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.8971818172100244measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
480.038304001092910770.062688000500202180.0433375997468829140.0405119992792606350.0058230162479461160.0231359992176294330.037471998482942580.0252063999883830530.023935999721288680.0032713812562311610.00.00.00.00.01.98099195957183842.04153609275817872.0037151813507081.9927519559860230.02206607643464573740961281281standard_fused_topkfixed_hotset8softmax_renormFalsestandalone_legacyvllm020_replicated_linear82048768True2327680.375256.0256.00.06250.03.87298334620741716.03.00.9375hotset820260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.8113207890716184measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
490.056607998907566070.079328000545501710.062499200180172920.060399999842047690.0056364616914808450.029567999765276910.037471998482942580.0311263998970389350.0302879996597766880.0021575710066315410.00.00.00.00.03.7901120185852053.86511993408203123.8291391611099243.82308793067932130.02517746416011056481921281281standard_fused_topkfixed_hotset8softmax_renormFalsestandalone_legacyvllm020_replicated_linear82048768True2655360.375512.0512.00.06250.03.87298334620741716.03.00.9375hotset820260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.883909666885606measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
500.025024000555276870.060927998274564740.0338399996981024740.0286720003932714460.0103152553437098180.0193600002676248550.03529600054025650.0234240002930164340.0220640003681182860.00388981022891945720.00.00.00.00.00.2123519927263260.243839994072914120.227600000798702240.227232001721858980.0105056857583759411281281standard_fused_topklogit_topksoftmax_renormTruestandalone_legacyvllm020_replicated_linear82048768True480.3750.06250.06250.06250.03.87298334620741716.03.00.9375uniform_random_logits20260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone1.0233364439829928measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
510.0252800006419420240.051327999681234360.0306415998376905930.0273439995944499970.0075624092327633040.0201600007712841030.0529600009322166440.0241456001996994030.0214240001514554020.0074501986455999850.00.00.00.00.00.474943995475769040.51840001344680790.49207040071487430.491696000099182130.01199106470147185581281281standard_fused_topklogit_topksoftmax_renormTruestandalone_legacyvllm020_replicated_linear82048768True4640.3750.50.50.39843750.01.39194109070750546.05.5701597655573920.667236328125uniform_random_logits20260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9806430689981738measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
520.0248319990932941440.0465600006282329560.0303712001070380220.027631999924778940.0058784872050286020.0203200001269578930.045600000768899920.026019200123846530.022704000584781170.007511299659067990.00.00.00.00.00.63609600067138670.70044797658920290.66083840131759640.65723198652267460.020416877242438597161281281standard_fused_topklogit_topksoftmax_renormTruestandalone_legacyvllm020_replicated_linear82048768True41280.3751.01.00.6250.01.03077640640441514.06.1382518552828270.5382080078125uniform_random_logits20260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9103623678483975measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
530.0242880005389451980.0493759997189044950.030860800016671420.02673599962145090.00708645313849972250.0204799994826316830.0309120006859302520.022747199889272450.0213599996641278270.00299668136506725050.00.00.00.00.00.7808960080146790.83014398813247680.803465598821640.80303999781608580.016801230312128875321281281standard_fused_topklogit_topksoftmax_renormTruestandalone_legacyvllm020_replicated_linear82048768True42560.3752.02.00.8593750.00.69033506357420383.06.59437470912181740.38067626953125uniform_random_logits20260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9610778571819444measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
540.023935999721288680.043423999100923540.0293807999230921260.0266880001872777940.00573747820005265740.0200319997966289520.0366079993546009060.0224879999645054350.0209119999781250950.00381352030621032350.00.00.00.00.00.86070400476455690.91952002048492430.87830080389976510.87510398030281070.01719059253115595641281281standard_fused_topklogit_topksoftmax_renormTruestandalone_legacyvllm020_replicated_linear82048768True45120.3754.04.00.97656250.00.494105884401309262.756.8144524743471340.271270751953125uniform_random_logits20260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9952941013961014measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
550.0254079997539520260.055103998631238940.0314303997904062240.027983999811112880.0073350502739827250.0204799994826316830.035615999251604080.022758399881422520.0215519992634654050.0035457183651658110.00.00.00.00.00.8339520096778870.8947520256042480.86199679970741270.8632159829139710.0187163788517971981281281281standard_fused_topklogit_topksoftmax_renormTruestandalone_legacyvllm020_replicated_linear82048768True410240.3758.08.01.00.250.336935291450747242.1256.91860752631555350.1867218017578125uniform_random_logits20260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9273256282883522measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
560.02441599965095520.053952001035213470.031889599841088060.0260319998487830160.009439959273874830.020511999726295470.0366079993546009060.0232479997910559170.021471999585628510.0039106238370696480.00.00.00.00.00.8348799943923950.88710397481918330.86515520215034480.86388799548149110.0152624308944009692561281281standard_fused_topklogit_topksoftmax_renormTruestandalone_legacyvllm020_replicated_linear82048768True420480.37516.016.01.00.3750.255672940186774561.81256.9524410491549370.14349365234375uniform_random_logits20260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone1.0380599882396606measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
570.0245120003819465640.046208001673221590.028846400231122970.0263200001791119580.0055165575833559250.020543999969959260.038463998585939410.0234016000293195240.0212639998644590380.00447429441982653740.00.00.00.00.00.85180801153182980.90975999832153320.88102720975875860.87510401010513310.0170058196332719065121281281standard_fused_topklogit_topksoftmax_renormTruestandalone_legacyvllm020_replicated_linear82048768True440960.37532.032.01.00.6250.17478013532185231.531256.9780695544827230.09820938110351562uniform_random_logits20260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9569603278386984measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
580.024127999320626260.069407999515533450.0308176004327833650.0268000001087784770.0100470851876529390.0201280005276203160.0440960004925727840.0246064000762999040.0223040003329515460.006228571668237140.00.00.00.00.00.84700798988342290.90102398395538330.86940159201622010.87169599533081050.01613888117119021610241281281standard_fused_topklogit_topksoftmax_renormTruestandalone_legacyvllm020_replicated_linear82048768True481920.37564.064.01.00.656250.11581224281541871.31256.99019081833588450.06445503234863281uniform_random_logits20260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9524274993623046measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
590.028319999575614930.0434879995882511140.0317440001294016850.029919999651610850.0039734158364289090.020703999325633050.0293439999222755430.0228864000178873530.0217439997941255570.00268730650450888730.00.00.00.00.01.16982400417327881.23113596439361571.18884799480438241.18907201290130620.01797879779449275820481281281standard_fused_topklogit_topksoftmax_renormTruestandalone_legacyvllm020_replicated_linear82048768True4163840.375128.0128.01.00.781250.083471818931086341.17968756.9949217725731540.046871185302734375uniform_random_logits20260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.8971818172100244measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
600.038304001092910770.062688000500202180.0433375997468829140.0405119992792606350.0058230162479461160.0231359992176294330.037471998482942580.0252063999883830530.023935999721288680.0032713812562311610.00.00.00.00.01.77027201652526861.80975997447967531.79191039800643931.79566401243209840.01264129567602155740961281281standard_fused_topklogit_topksoftmax_renormTruestandalone_legacyvllm020_replicated_linear82048768True4327680.375256.0256.01.00.781250.068667344778224841.207031256.9966025629387280.03801727294921875uniform_random_logits20260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.8113207890716184measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
610.056607998907566070.079328000545501710.062499200180172920.060399999842047690.0056364616914808450.029567999765276910.037471998482942580.0311263998970389350.0302879996597766880.0021575710066315410.00.00.00.00.02.9686720371246343.02780795097351072.98990077972412132.98247992992401120.0181641412263166781921281281standard_fused_topklogit_topksoftmax_renormTruestandalone_legacyvllm020_replicated_linear82048768True4655360.375512.0512.01.00.89843750.043995468339773761.1269531256.9986073149223620.024699926376342773uniform_random_logits20260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.883909666885606measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
620.025024000555276870.060927998274564740.0338399996981024740.0286720003932714460.0103152553437098180.0193600002676248550.03529600054025650.0234240002930164340.0220640003681182860.00388981022891945720.00.00.00.00.00.195743992924690250.25129601359367370.219392000138759620.21999999880790710.01753215641821256511281281standard_fused_topkfixed_hotset8softmax_renormFalsestandalone_legacyvllm020_replicated_linear82048768True480.3750.06250.06250.06250.03.87298334620741716.03.00.9375hotset820260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone1.0233364439829928measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
630.0252800006419420240.051327999681234360.0306415998376905930.0273439995944499970.0075624092327633040.0201600007712841030.0529600009322166440.0241456001996994030.0214240001514554020.0074501986455999850.00.00.00.00.00.204832002520561220.240063995122909550.222150400280952450.224335998296737670.0096963978613289281281281standard_fused_topkfixed_hotset8softmax_renormFalsestandalone_legacyvllm020_replicated_linear82048768True4640.3750.50.50.06250.03.87298334620741716.03.00.9375hotset820260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9806430689981738measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
640.0248319990932941440.0465600006282329560.0303712001070380220.027631999924778940.0058784872050286020.0203200001269578930.045600000768899920.026019200123846530.022704000584781170.007511299659067990.00.00.00.00.00.205599993467330930.247263997793197630.221267198026180280.222079999744892120.01328093478485499161281281standard_fused_topkfixed_hotset8softmax_renormFalsestandalone_legacyvllm020_replicated_linear82048768True41280.3751.01.00.06250.03.87298334620741716.03.00.9375hotset820260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9103623678483975measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
650.0242880005389451980.0493759997189044950.030860800016671420.02673599962145090.00708645313849972250.0204799994826316830.0309120006859302520.022747199889272450.0213599996641278270.00299668136506725050.00.00.00.00.00.200031995773315430.23011200129985810.214531199634075160.215984001755714420.010402239855151332321281281standard_fused_topkfixed_hotset8softmax_renormFalsestandalone_legacyvllm020_replicated_linear82048768True42560.3752.02.00.06250.03.87298334620741716.03.00.9375hotset820260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9610778571819444measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
660.023935999721288680.043423999100923540.0293807999230921260.0266880001872777940.00573747820005265740.0200319997966289520.0366079993546009060.0224879999645054350.0209119999781250950.00381352030621032350.00.00.00.00.00.195519998669624330.229727998375892640.212387199699878680.214880004525184630.010706095835489097641281281standard_fused_topkfixed_hotset8softmax_renormFalsestandalone_legacyvllm020_replicated_linear82048768True45120.3754.04.00.06250.03.87298334620741716.03.00.9375hotset820260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9952941013961014measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
670.0254079997539520260.055103998631238940.0314303997904062240.027983999811112880.0073350502739827250.0204799994826316830.035615999251604080.022758399881422520.0215519992634654050.0035457183651658110.00.00.00.00.00.194207996129989620.29039999842643740.22112319916486740.214767999947071080.0259550088861960041281281281standard_fused_topkfixed_hotset8softmax_renormFalsestandalone_legacyvllm020_replicated_linear82048768True410240.3758.08.00.06250.03.87298334620741716.03.00.9375hotset820260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9273256282883522measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
680.02441599965095520.053952001035213470.031889599841088060.0260319998487830160.009439959273874830.020511999726295470.0366079993546009060.0232479997910559170.021471999585628510.0039106238370696480.00.00.00.00.00.22905600070953370.2892479896545410.25433279871940610.249392002820968630.016632367535031152561281281standard_fused_topkfixed_hotset8softmax_renormFalsestandalone_legacyvllm020_replicated_linear82048768True420480.37516.016.00.06250.03.87298334620741716.03.00.9375hotset820260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone1.0380599882396606measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
690.0245120003819465640.046208001673221590.028846400231122970.0263200001791119580.0055165575833559250.020543999969959260.038463998585939410.0234016000293195240.0212639998644590380.00447429441982653740.00.00.00.00.00.308319985866546630.369536012411117550.33240000009536740.32886399328708650.0186179995721567075121281281standard_fused_topkfixed_hotset8softmax_renormFalsestandalone_legacyvllm020_replicated_linear82048768True440960.37532.032.00.06250.03.87298334620741716.03.00.9375hotset820260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9569603278386984measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
700.024127999320626260.069407999515533450.0308176004327833650.0268000001087784770.0100470851876529390.0201280005276203160.0440960004925727840.0246064000762999040.0223040003329515460.006228571668237140.00.00.00.00.00.4629119932651520.54979199171066280.48938560187816620.48169600963592530.02334815288717828610241281281standard_fused_topkfixed_hotset8softmax_renormFalsestandalone_legacyvllm020_replicated_linear82048768True481920.37564.064.00.06250.03.87298334620741716.03.00.9375hotset820260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.9524274993623046measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
710.028319999575614930.0434879995882511140.0317440001294016850.029919999651610850.0039734158364289090.020703999325633050.0293439999222755430.0228864000178873530.0217439997941255570.00268730650450888730.00.00.00.00.00.76620799303054810.87177598476409910.7884543955326080.77447998523712160.0317474648260481220481281281standard_fused_topkfixed_hotset8softmax_renormFalsestandalone_legacyvllm020_replicated_linear82048768True4163840.375128.0128.00.06250.03.87298334620741716.03.00.9375hotset820260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.8971818172100244measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
720.038304001092910770.062688000500202180.0433375997468829140.0405119992792606350.0058230162479461160.0231359992176294330.037471998482942580.0252063999883830530.023935999721288680.0032713812562311610.00.00.00.00.01.3639359474182131.41430401802062991.38127039670944221.37987202405929570.01477045007553000740961281281standard_fused_topkfixed_hotset8softmax_renormFalsestandalone_legacyvllm020_replicated_linear82048768True4327680.375256.0256.00.06250.03.87298334620741716.03.00.9375hotset820260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.8113207890716184measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize
730.056607998907566070.079328000545501710.062499200180172920.060399999842047690.0056364616914808450.029567999765276910.037471998482942580.0311263998970389350.0302879996597766880.0021575710066315410.00.00.00.00.02.55075192451477052.6807041168212892.5798592090606692.5662239789962770.0367328355836595581921281281standard_fused_topkfixed_hotset8softmax_renormFalsestandalone_legacyvllm020_replicated_linear82048768True4655360.375512.0512.00.06250.03.87298334620741716.03.00.9375hotset820260716FlashInfer CUTLASSCUDA_EVENTBF16genericnone0.883909666885606measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize