157 Commits

Author SHA1 Message Date
Christian Risi
b79521995c Last fixes 2025-10-17 22:17:24 +02:00
Christian Risi
540b78204c Added epochs 2025-10-17 17:06:42 +02:00
GassiGiuseppe
86a063591e Merge branch 'dev.train' of https://repositories.communitynotfound.work/PoliBa-DeepLearning/NanoSocrates into dev.train 2025-10-17 16:38:12 +02:00
GassiGiuseppe
f33d4f1db6 Added a loss_saver file to save the losses 2025-10-17 16:37:44 +02:00
Christian Risi
fe62b1edd5 Fixed evaluation 2025-10-16 20:05:35 +02:00
Christian Risi
892f91aad7 Fixes for evaluation 2025-10-16 19:20:23 +02:00
Christian Risi
4968d79403 Fixed a masking problem 2025-10-14 10:34:14 +02:00
GassiGiuseppe
80fd7fd600 evaluator WIP 2025-10-12 22:59:07 +02:00
GassiGiuseppe
7dedbc481b evaluator WIP 2025-10-12 18:18:20 +02:00
GassiGiuseppe
2ccec9efb8 typo 2025-10-12 16:41:06 +02:00
GassiGiuseppe
e2231eb3b9 Merge branch 'dev.train' of https://repositories.communitynotfound.work/PoliBa-DeepLearning/NanoSocrates into dev.train 2025-10-12 16:36:09 +02:00
GassiGiuseppe
144f8724d6 Update of the batcher to resolve a bug in the 4th construction 2025-10-12 16:35:42 +02:00
Christian Risi
07130ff489 Fixed several bugs for task 4 2025-10-12 16:30:30 +02:00
Christian Risi
4ca1d0a189 Activated Dropout to avoid overfitting 2025-10-12 12:28:06 +02:00
Christian Risi
f463f699cf Fixed a bug over task 4 2025-10-12 12:22:38 +02:00
Christian Risi
bcc2fe7368 Fixed bugs and added visibility 2025-10-11 21:49:29 +02:00
Christian Risi
160b7dbfc0 V0.0.1 Athene 2025-10-11 19:35:43 +02:00
GassiGiuseppe
49946727d8 updated decoder_input to work without embedder 2025-10-11 16:53:36 +02:00
GassiGiuseppe
1649cd7768 added decoder_input method to build the batch tensor to give in input to the deocder 2025-10-11 16:18:43 +02:00
GassiGiuseppe
443f54fffd WIP decoder with prefix mask 2025-10-11 15:31:43 +02:00
GassiGiuseppe
ff721107b9 typo 2025-10-11 15:26:58 +02:00
GassiGiuseppe
f1886e5be1 added builder for prefix mask 2025-10-11 15:19:09 +02:00
GassiGiuseppe
5e3878ea17 Merge branch 'dev' into dev.train 2025-10-11 11:51:58 +02:00
GassiGiuseppe
586f021276 Merge branch 'dev.train' of https://repositories.communitynotfound.work/PoliBa-DeepLearning/NanoSocrates into dev.train 2025-10-11 11:28:35 +02:00
GassiGiuseppe
82462078f8 WIP for the new prefix mask 2025-10-11 11:28:15 +02:00
Christian Risi
625f79f7c3 Fixed imports 2025-10-11 11:18:44 +02:00
GassiGiuseppe
3446870291 typo 2025-10-10 22:27:01 +02:00
GassiGiuseppe
e76dbeb9a7 typo 2025-10-10 22:26:06 +02:00
GassiGiuseppe
96610612fe Batcher added 2025-10-10 20:10:08 +02:00
Christian Risi
92ae40013d Added a way to detach models and create them standalone 2025-10-10 18:43:20 +02:00
GassiGiuseppe
ee253c39f4 Merge branch 'dev.embedder' of https://repositories.communitynotfound.work/PoliBa-DeepLearning/NanoSocrates into dev.embedder 2025-10-09 13:31:37 +02:00
GassiGiuseppe
2036b4015f added logistic collector 2025-10-09 13:31:16 +02:00
Christian Risi
aac7675b30 Pipeline fix and added a util to decode 2025-10-09 13:24:48 +02:00
Christian Risi
0158db2dce Fixed a bug where I took encoder embeddings rather than encoder output 2025-10-09 11:37:21 +02:00
Christian Risi
ba592c3480 Disabled Softmax 2025-10-09 11:36:56 +02:00
Christian Risi
1f9c30b531 Added Custom Learning Rate 2025-10-09 11:36:40 +02:00
GassiGiuseppe
b805dc538e learning sheduler as torch one 2025-10-08 16:05:22 +02:00
Christian Risi
c2e13bc9c6 Quick fix to architecture 2025-10-08 12:34:09 +02:00
GassiGiuseppe
14c3914571 Merge branch 'dev.embedder' of https://repositories.communitynotfound.work/PoliBa-DeepLearning/NanoSocrates into dev.embedder 2025-10-08 12:14:05 +02:00
Christian Risi
b9273b95e2 Merge branch 'dev.embedder' of https://repositories.communitynotfound.work/PoliBa-DeepLearning/NanoSocrates into dev.embedder 2025-10-08 12:13:46 +02:00
GassiGiuseppe
c263e2cf13 Custom learning rate sheduler from Attention is all you need 2025-10-08 12:13:02 +02:00
GassiGiuseppe
c9a50d50b7 typo in Batcher 2025-10-08 11:39:08 +02:00
GassiGiuseppe
9b0c57c238 Batcher ended, attention it returns list of tokenId, which later needs to be embedded 2025-10-08 11:26:47 +02:00
Christian Risi
24ea4d3ba4 Added a training model for NanoSocrates 2025-10-08 11:18:05 +02:00
GassiGiuseppe
159266a603 WIP Batcher added class to fourth task 2025-10-08 00:39:16 +02:00
GassiGiuseppe
7027414342 Merge branch 'dev.embedder' of https://repositories.communitynotfound.work/PoliBa-DeepLearning/NanoSocrates into dev.embedder 2025-10-07 23:16:20 +02:00
GassiGiuseppe
fc44929a7b moved spanned mask variables in init for better reliability, also tested 2025-10-07 23:15:50 +02:00
Christian Risi
0560bc439a Merge branch 'dev.embedder' of https://repositories.communitynotfound.work/PoliBa-DeepLearning/NanoSocrates into dev.embedder 2025-10-07 20:45:10 +02:00
Christian Risi
533347ee22 Added new special token 2025-10-07 20:44:54 +02:00
Christian Risi
d1ff88da82 Added small dataset 2025-10-07 20:44:40 +02:00