GassiGiuseppe
|
14b810c451
|
WIP NanoSocratesEmbedder for batching
|
2025-10-06 21:41:45 +02:00 |
|
GassiGiuseppe
|
56d438f01a
|
WIP NanoSocratesCore
|
2025-10-06 18:21:27 +02:00 |
|
GassiGiuseppe
|
745424a978
|
new special token for start sequence in decoder
|
2025-10-06 18:21:10 +02:00 |
|
GassiGiuseppe
|
e1549d4458
|
Modified decoder and decoder for sequential architecture
|
2025-10-06 18:20:46 +02:00 |
|
Christian Risi
|
456ce724fe
|
Added capability of returning target after truncating
|
2025-10-06 17:43:01 +02:00 |
|
Christian Risi
|
44307cd917
|
Added util to create padding mask
|
2025-10-06 17:29:05 +02:00 |
|
Christian Risi
|
ffdb312d58
|
Added a util to create truncated RDF lists
|
2025-10-06 17:22:13 +02:00 |
|
Christian Risi
|
0007c38212
|
Added a util to make masked inference
|
2025-10-06 17:02:06 +02:00 |
|
Christian Risi
|
9c1043e0ba
|
Added post tokenization utils
|
2025-10-06 17:01:18 +02:00 |
|
Christian Risi
|
ee8e56798c
|
Added new utils
|
2025-10-06 17:00:55 +02:00 |
|
Christian Risi
|
e93710af08
|
Fixed illegal tokens being added in target output
|
2025-10-06 16:16:47 +02:00 |
|
Christian Risi
|
b1e7af0607
|
Merge branch 'dev.embedder' of https://repositories.communitynotfound.work/PoliBa-DeepLearning/NanoSocrates into dev.embedder
|
2025-10-06 15:55:44 +02:00 |
|
Christian Risi
|
c217f5dec9
|
Added 2 types of masking
|
2025-10-06 15:45:45 +02:00 |
|
Christian Risi
|
49f0beb6ea
|
Updated imports
|
2025-10-06 15:45:28 +02:00 |
|
GassiGiuseppe
|
948c3fd7ac
|
update to batch attention mask
|
2025-10-06 13:03:03 +02:00 |
|
GassiGiuseppe
|
87409fecd5
|
added method fot batched attention_mask
|
2025-10-06 12:00:11 +02:00 |
|
GassiGiuseppe
|
7e40a36701
|
wip: NanoSocratesCore
|
2025-10-05 22:58:06 +02:00 |
|
GassiGiuseppe
|
d48815cca2
|
added task_type and updated init
|
2025-10-05 18:58:42 +02:00 |
|
GassiGiuseppe
|
0f243eaac2
|
added padding_mask entry to decoder and encoder
|
2025-10-05 18:46:06 +02:00 |
|
GassiGiuseppe
|
6f219f634f
|
Added attention_mask
|
2025-10-05 17:49:01 +02:00 |
|
Christian Risi
|
c60da8ba82
|
Refactoring
|
2025-10-05 15:40:29 +02:00 |
|
GassiGiuseppe
|
76200d936d
|
added first classes (Encoder, Decoder, Attention) for the model
|
2025-10-04 21:07:58 +02:00 |
|
Christian Risi
|
9a797a0485
|
Added embedder code for "Attention is all you need"
|
2025-10-04 19:43:25 +02:00 |
|
Christian Risi
|
3b274ad807
|
Added a way to take the default special token list
|
2025-10-04 19:43:02 +02:00 |
|
Christian Risi
|
da0bdf703b
|
Added a way to see vocabulary size
|
2025-10-04 19:42:29 +02:00 |
|
Christian Risi
|
03cdca1f00
|
Modified imports for BPE
|
2025-10-04 19:42:02 +02:00 |
|
Christian Risi
|
7188c8678a
|
Added imports for Embedder
|
2025-10-04 19:41:48 +02:00 |
|
Christian Risi
|
23d1eaf99e
|
Fixed a rare bug over training multiple times
|
2025-10-04 10:47:39 +02:00 |
|
Christian Risi
|
c6ac6df2c2
|
Added stubs for other libraries
|
2025-10-03 20:28:23 +02:00 |
|
Christian Risi
|
87f24878f4
|
Added shims for utils on using Pytorch
|
2025-10-03 20:11:14 +02:00 |
|
Christian Risi
|
d2a3dfe90f
|
Fixed bug
|
2025-10-03 17:59:46 +02:00 |
|
Christian Risi
|
0ee6e48004
|
Fixed the same bug as before, but this time is correct
|
2025-10-03 16:09:53 +02:00 |
|
Christian Risi
|
55e0d2ac23
|
Fixed a encoding bug
|
2025-10-03 16:08:11 +02:00 |
|
Christian Risi
|
9c5f42153f
|
fixed typos
|
2025-10-03 15:17:44 +02:00 |
|
Christian Risi
|
51f491d033
|
fixed typos
|
2025-10-03 13:27:17 +02:00 |
|
Christian Risi
|
c5c0c61f79
|
Fix of bugs and semantics
|
2025-10-03 13:26:58 +02:00 |
|
Christian Risi
|
6b9cb7cd35
|
Modified imports
|
2025-10-03 13:26:42 +02:00 |
|
Christian Risi
|
e8894504c6
|
Fixed a bug where a token (int) was yielded instead of a list of int
|
2025-10-03 11:44:44 +02:00 |
|
GassiGiuseppe
|
845d645348
|
added some stubs on special_regex_maker
|
2025-10-03 10:38:35 +02:00 |
|
GassiGiuseppe
|
070dc1b744
|
implemented token nano for the BPE encoding/decoding
|
2025-10-03 01:04:06 +02:00 |
|
GassiGiuseppe
|
8121c75a09
|
Updated NanoSocratesSplitter to split also token in decode phase
|
2025-10-03 01:00:36 +02:00 |
|
GassiGiuseppe
|
a5b8692a77
|
Updated NanoSocratesSpecial to work with TokeNano
|
2025-10-03 00:59:15 +02:00 |
|
GassiGiuseppe
|
7c935d2700
|
Update NanoSocratesBPE: corrected a minor bug about dictionary lenght,
added some comment to make the code more clear
|
2025-10-03 00:57:19 +02:00 |
|
GassiGiuseppe
|
0eef2148a9
|
in NanoSocratesBPE: encode() method rewritten and tested
|
2025-10-02 12:12:44 +02:00 |
|
Christian Risi
|
856bd8909c
|
Added treshold
|
2025-10-02 11:02:03 +02:00 |
|
Christian Risi
|
2e595a3a23
|
Changed training phase to take directly data instead of its encode
|
2025-10-02 09:56:44 +02:00 |
|
Christian Risi
|
1eae8582b2
|
Fixed decoding phase
|
2025-10-02 09:33:58 +02:00 |
|
Christian Risi
|
aa765b4555
|
Added time checking
|
2025-10-02 08:48:45 +02:00 |
|
Christian Risi
|
0975c19e69
|
added nwew method to encode from list of tokens
|
2025-10-02 08:48:13 +02:00 |
|
Christian Risi
|
3fe4e45ceb
|
Fixed a bug while joining frequencies
|
2025-10-02 01:50:37 +02:00 |
|