davidhouseholter commited on
Commit
8956eb8
·
verified ·
1 Parent(s): 00e6b36

Update 17-way classical geometric optimizer ablation (add HuygensSymplecticAdam): eval_results.json

Browse files
Files changed (1) hide show
  1. eval_results.json +128 -0
eval_results.json CHANGED
@@ -140,6 +140,134 @@
140
  "stsb_test_pearson_after": 0.6564,
141
  "rocm_elapsed_sec": 243.28
142
  },
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
143
  "polar_natural_adam": {
144
  "stsb_dev_spearman_after": 0.7262,
145
  "stsb_dev_pearson_after": 0.702,
 
140
  "stsb_test_pearson_after": 0.6564,
141
  "rocm_elapsed_sec": 243.28
142
  },
143
+ "huygens_symplectic_adam": {
144
+ "s1_dev_spearman": 0.6282,
145
+ "stsb_dev_spearman_after": 0.7368,
146
+ "stsb_dev_pearson_after": 0.709,
147
+ "stsb_test_spearman_after": 0.6509,
148
+ "stsb_test_pearson_after": 0.6548,
149
+ "rocm_elapsed_sec": 236.41
150
+ },
151
+ "polar_natural_adam": {
152
+ "stsb_dev_spearman_after": 0.7262,
153
+ "stsb_dev_pearson_after": 0.702,
154
+ "stsb_test_spearman_after": 0.6471,
155
+ "stsb_test_pearson_after": 0.6494
156
+ },
157
+ "gyric_adam": {
158
+ "s1_dev_spearman": 0.6221,
159
+ "stsb_dev_spearman_after": 0.7304,
160
+ "stsb_dev_pearson_after": 0.7023,
161
+ "stsb_test_spearman_after": 0.6455,
162
+ "stsb_test_pearson_after": 0.6503,
163
+ "rocm_elapsed_sec": 419.85
164
+ },
165
+ "huygens_adam": {
166
+ "s1_dev_spearman": 0.6383,
167
+ "stsb_dev_spearman_after": 0.7295,
168
+ "stsb_dev_pearson_after": 0.7035,
169
+ "stsb_test_spearman_after": 0.642,
170
+ "stsb_test_pearson_after": 0.6469,
171
+ "rocm_elapsed_sec": 331.8
172
+ },
173
+ "hessian_pair_adam_1x_coupling": {
174
+ "stsb_dev_spearman_after": 0.7242,
175
+ "stsb_dev_pearson_after": 0.6876,
176
+ "stsb_test_spearman_after": 0.6337,
177
+ "stsb_test_pearson_after": 0.6369
178
+ },
179
+ "beltrami_adam": {
180
+ "s1_dev_spearman": 0.6337,
181
+ "stsb_dev_spearman_after": 0.7363,
182
+ "stsb_dev_pearson_after": 0.7068,
183
+ "stsb_test_spearman_after": 0.6333,
184
+ "stsb_test_pearson_after": 0.6328,
185
+ "rocm_elapsed_sec": 321.44
186
+ },
187
+ "hannay_adam": {
188
+ "s1_dev_spearman": 0.6115,
189
+ "stsb_dev_spearman_after": 0.7257,
190
+ "stsb_dev_pearson_after": 0.699,
191
+ "stsb_test_spearman_after": 0.6333,
192
+ "stsb_test_pearson_after": 0.633,
193
+ "rocm_elapsed_sec": 239.35
194
+ },
195
+ "rotor_adam": {
196
+ "stsb_dev_spearman_after": 0.7322,
197
+ "stsb_dev_pearson_after": 0.7099,
198
+ "stsb_test_spearman_after": 0.6329,
199
+ "stsb_test_pearson_after": 0.6309
200
+ },
201
+ "toroidal_lion": {
202
+ "stsb_dev_spearman_after": 0.7289,
203
+ "stsb_dev_pearson_after": 0.697,
204
+ "stsb_test_spearman_after": 0.6319,
205
+ "stsb_test_pearson_after": 0.6341
206
+ },
207
+ "adamw": {
208
+ "stsb_dev_spearman_after": 0.7289,
209
+ "stsb_test_spearman_after": 0.6317,
210
+ "stsb_test_pearson_after": 0.6372
211
+ },
212
+ "poncelet_adam": {
213
+ "s1_dev_spearman": 0.6262,
214
+ "stsb_dev_spearman_after": 0.7275,
215
+ "stsb_dev_pearson_after": 0.6987,
216
+ "stsb_test_spearman_after": 0.6315,
217
+ "stsb_test_pearson_after": 0.6374,
218
+ "rocm_elapsed_sec": 267.34
219
+ },
220
+ "hessian_pair_adam_3x_coupling": {
221
+ "stsb_dev_spearman_after": 0.7278,
222
+ "stsb_dev_pearson_after": 0.6917,
223
+ "stsb_test_spearman_after": 0.6288,
224
+ "stsb_test_pearson_after": 0.6316
225
+ },
226
+ "cyclide_adam": {
227
+ "s1_dev_spearman": 0.6255,
228
+ "stsb_dev_spearman_after": 0.7368,
229
+ "stsb_dev_pearson_after": 0.7179,
230
+ "stsb_test_spearman_after": 0.6287,
231
+ "stsb_test_pearson_after": 0.6476,
232
+ "rocm_elapsed_sec": 380.42
233
+ },
234
+ "rindler_adam": {
235
+ "s1_dev_spearman": 0.6208,
236
+ "stsb_dev_spearman_after": 0.7318,
237
+ "stsb_dev_pearson_after": 0.7069,
238
+ "stsb_test_spearman_after": 0.6264,
239
+ "stsb_test_pearson_after": 0.6353,
240
+ "rocm_elapsed_sec": 271.68
241
+ },
242
+ "pair_adam_3x_coupling": {
243
+ "stsb_dev_spearman_after": 0.7179,
244
+ "stsb_test_spearman_after": 0.6242,
245
+ "stsb_test_pearson_after": 0.633
246
+ },
247
+ "toroidal_muon": {
248
+ "stsb_dev_spearman_after": 0.7361,
249
+ "stsb_dev_pearson_after": 0.7105,
250
+ "stsb_test_spearman_after": 0.6197,
251
+ "stsb_test_pearson_after": 0.6304
252
+ }
253
+ },
254
+ "optimizer_ablations_17way": {
255
+ "symplectic_adam": {
256
+ "s1_dev_spearman": 0.6198,
257
+ "stsb_dev_spearman_after": 0.736,
258
+ "stsb_dev_pearson_after": 0.7087,
259
+ "stsb_test_spearman_after": 0.6516,
260
+ "stsb_test_pearson_after": 0.6564,
261
+ "rocm_elapsed_sec": 243.28
262
+ },
263
+ "huygens_symplectic_adam": {
264
+ "s1_dev_spearman": 0.6282,
265
+ "stsb_dev_spearman_after": 0.7368,
266
+ "stsb_dev_pearson_after": 0.709,
267
+ "stsb_test_spearman_after": 0.6509,
268
+ "stsb_test_pearson_after": 0.6548,
269
+ "rocm_elapsed_sec": 236.41
270
+ },
271
  "polar_natural_adam": {
272
  "stsb_dev_spearman_after": 0.7262,
273
  "stsb_dev_pearson_after": 0.702,