temporal-moe-extras / ablations /e6_per_layer_ranking.csv
ncylich's picture
Add files using upload-large-folder tool
94cd78c verified
Raw
History Blame Contribute Delete
1.2 kB
model,layer,hit_rate,swap_rate,lifetime_tokens
"8.1M active, coarse (6 of 64 experts)",2,0.29211,0.99979,5.986
"8.1M active, coarse (6 of 64 experts)",3,0.31882,0.99979,5.986
"8.1M active, coarse (6 of 64 experts)",4,0.38150,0.99875,5.993
"8.1M active, coarse (6 of 64 experts)",5,0.43823,0.99624,6.001
"8.1M active, coarse (6 of 64 experts)",6,0.47917,0.99453,6.003
"3.9M active, fine-grained (18 of 192 experts)",2,0.24859,1.00000,17.852
"3.9M active, fine-grained (18 of 192 experts)",3,0.33151,1.00000,17.852
"3.9M active, fine-grained (18 of 192 experts)",4,0.35685,1.00000,17.852
"3.9M active, fine-grained (18 of 192 experts)",5,0.45289,1.00000,17.852
"38.0M active, coarse (6 of 64 experts)",2,0.16537,1.00000,5.985
"38.0M active, coarse (6 of 64 experts)",3,0.25895,0.99994,5.986
"38.0M active, coarse (6 of 64 experts)",4,0.30528,0.99957,5.988
"38.0M active, coarse (6 of 64 experts)",5,0.33149,0.99896,5.992
"38.0M active, coarse (6 of 64 experts)",6,0.36196,0.99908,5.991
"38.0M active, coarse (6 of 64 experts)",7,0.33429,0.99933,5.989
"38.0M active, coarse (6 of 64 experts)",8,0.33904,0.99908,5.991
"38.0M active, coarse (6 of 64 experts)",9,0.33754,0.99896,5.992