Pablogps commited on
Commit
30f98d9
1 Parent(s): 1bcdbad

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +22 -0
README.md CHANGED
@@ -229,6 +229,28 @@ All models trained with max length 512 and batch size 8. The accuracy values in
229
  <caption>Table 4. Results for PAWS-X.</caption>
230
  </figure>
231
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
232
  # Conclusions
233
 
234
  With roughly 10 days worth of access to 3xTPUv3-8, we have achieved remarkable results surpassing previous state of the art in a few tasks, and even improving document classification on models trained in massive supercomputers with very large—private—and highly curated datasets.
 
229
  <caption>Table 4. Results for PAWS-X.</caption>
230
  </figure>
231
 
232
+ **CNLI**
233
+ All models trained with max length 256 and batch size 16.
234
+
235
+ <figure>
236
+
237
+ | Model | Accuracy |
238
+ |----------------------------------------------------|----------|
239
+ | bert-base-multilingual-cased | WIP |
240
+ | dccuchile/bert-base-spanish-wwm-cased | WIP |
241
+ | BSC-TeMU/roberta-base-bne | WIP |
242
+ | flax-community/bertin-roberta-large-spanish | WIP |
243
+ | bertin-project/bertin-roberta-base-spanish | WIP |
244
+ | bertin-project/bertin-base-random | 0.7745 |
245
+ | bertin-project/bertin-base-stepwise | 0.7820 |
246
+ | bertin-project/bertin-base-gaussian | **0.7942** |
247
+ | bertin-project/bertin-base-random-exp-512seqlen | 0.7723 |
248
+ | bertin-project/bertin-base-gaussian-exp-512seqlen | 0.7878 |
249
+
250
+
251
+ <caption>Table 5. Results for CNLI.</caption>
252
+ </figure>
253
+
254
  # Conclusions
255
 
256
  With roughly 10 days worth of access to 3xTPUv3-8, we have achieved remarkable results surpassing previous state of the art in a few tasks, and even improving document classification on models trained in massive supercomputers with very large—private—and highly curated datasets.