github/concrete
1
1
Fork 0
mirror of https://github.com/zama-ai/concrete.git synced 2026-02-09 03:55:04 -05:00
Code Issues Packages Projects Releases Wiki Activity
608 Commits 185 Branches 425 Tags
1b5785e058d033f2f0119769e7c1926ee4de8fc6
Commit Graph

234 Commits

Author SHA1 Message Date
Umut
1b5785e058 feat(extensions): add and test support for negative direct table lookups 2021-12-01 12:09:10 +03:00
Arthur Meyre
8b0a793cda feat: create torch-like API part deux 
- add function to get the FHECircuit

closes #233
 2021-12-01 09:36:47 +01:00
Benoit Chevallier-Mames
edefde9189 feat: adding a compile_torch_model function 
refs #898
 2021-11-30 16:24:21 +01:00
Arthur Meyre
e111891a41 test: more signed TLUs cases 
closes #1014
 2021-11-30 11:45:15 +01:00
jfrery
1625475897 feat: end-to-end compilation of a torch model 2021-11-30 11:30:13 +01:00
Arthur Meyre
13b9ff96f0 feat: create torch-like APIs part 1 
- work on generating OPGraph with a torch-like API

refs #233
 2021-11-30 11:06:46 +01:00
Umut
ac74e94e13 feat(mlir): implement mlir conversion of constant indexing 2021-11-30 12:40:34 +03:00
Umut
926597c3f6 feat(debugging): implement printing subgraphs within the graph during printing 2021-11-30 12:40:23 +03:00
jfrery
dfc762c2e2 feat: add QuantizedReLU6 as a supported activation function 2021-11-26 16:55:52 +01:00
Arthur Meyre
f53d374d1f feat: support TLUs with signed input 
- this is done with an absolute ugly hack

closes #923
refs #1001
 2021-11-25 12:01:40 +01:00
Benoit Chevallier-Mames
39b168dfd5 fix: fix issues for certain form of functions 
Table which was created contained too large values, for cells which were never accessed
closes #949
 2021-11-24 17:14:59 +01:00
Benoit Chevallier-Mames
33064efc06 chore: fix flaky tests 
closes #997
 2021-11-24 09:48:02 +01:00
Benoit Chevallier-Mames
ec396effb2 chore: seed torch as much as possible 
closes #877
 2021-11-24 09:47:37 +01:00
Arthur Meyre
60d8079303 feat: add clip support 
closes #965
closes #983
closes #984
 2021-11-23 17:08:24 +01:00
Arthur Meyre
f1ed07d580 refactor(tracing): add generic tracing functions for generic functions 
- remove nearly duplicated code when tracing GenericFunction for unary and
binary operators

refs #965
 2021-11-23 13:55:37 +01:00
Arthur Meyre
ba6207e71e refactor: remove workaround for scalar and tensors operations in tests 
- add Float16 for ldexp which sometimes returns float16 in numpy

closes #908
 2021-11-23 12:25:53 +01:00
Benoit Chevallier-Mames
446dfaf834 chore: let us use dtype=np.uint8 if we want 
closes #986
 2021-11-23 09:20:05 +01:00
Umut
7e65af3906 fix(representation): set correct output type for dot product 2021-11-23 10:22:17 +03:00
Umut
1d77816aa3 refactor(mlir): implement support for operations between tensors and scalars using string processing hacks 2021-11-23 10:22:17 +03:00
Arthur Meyre
a77c369daf feat: accept non tuple inputs ininputset for 1-parameter functions 
closes #952
 2021-11-22 13:38:17 +01:00
Benoit Chevallier-Mames
0fd8da8789 chore: fix a test 
refs #885
 2021-11-19 15:08:50 +01:00
Benoit Chevallier-Mames
247327b277 chore: fix #951 by making sure the dot product is never too large. 
closes #951
 2021-11-19 10:17:27 +01:00
Arthur Meyre
8a27525a64 refactor: separate some compilation steps to prepare torch-like API work 
refs #233
 2021-11-19 10:06:07 +01:00
jfrery
aa60d8ace6 fix: test_quantized_layer 2021-11-18 19:23:27 +01:00
Benoit Chevallier-Mames
2fdf22bf4e test(execution): add correctness tests of ndarray operators 
closes #931
 2021-11-18 18:16:33 +01:00
jfrery
a5b1d6232e feat: add signed intergers quantization 2021-11-18 17:21:28 +01:00
Benoit Chevallier-Mames
4c6e1661ec feat(tracing): add support for more arithmetic operators, when one input is a constant 
refs #218
refs #751
 2021-11-18 13:46:14 +01:00
Arthur Meyre
507ccd05c5 feat: static post training quantization and quantization module 2021-11-18 10:31:45 +01:00
Benoit Chevallier-Mames
bfa309a455 test: add tests with dot and negative tensors 
refs #891
 2021-11-18 09:55:22 +01:00
Arthur Meyre
f6aadd37e9 test: add tests for bivariate functions with tensor and multi TLU 
closes #558
closes #567
closes #665
 2021-11-18 09:08:55 +01:00
Arthur Meyre
ff03bc2220 feat(optimization): support more fusing topologies 
- corrected docstring that was mistaken on what was returned
- updated pyproject.toml to ignore warnings that happened naturally in
networkx and that was blocking proper test execution (no way around that
this is code from networkx that triggered the warning)
- add a test case for the newly supported fusing topology

closes #499
 2021-11-18 09:08:50 +01:00
Arthur Meyre
bc145e21e1 feat: allow tracing __eq__ while keeping hashing when not tracing 
closes #936
 2021-11-18 09:08:43 +01:00
jfrery
c978107124 feat: remove transpose from layers 2021-11-17 14:54:16 +01:00
Benoit Chevallier-Mames
a712b0573c feat: add support for comparators when one input is a constant 
closes #932
refs #751
 2021-11-17 13:50:18 +01:00
Umut
c733daa78c feat(mlir): implement MLIR conversion of MatMul 2021-11-17 15:01:29 +03:00
Benoit Chevallier-Mames
05cacc8744 feat(tracing): let's trace k<<x and k>>x 
closes #915
 2021-11-17 12:50:59 +01:00
Umut
1d691f232b feat(mlir): implement MLIR conversion of multi lookup tables 2021-11-17 14:27:47 +03:00
Benoit Chevallier-Mames
78f82fb9a1 chore: remove useless variables that make pcc didn't see 2021-11-17 11:01:57 +01:00
Benoit Chevallier-Mames
5448669e83 chore: split test_tracing 
closes #913
 2021-11-17 10:10:05 +01:00
Arthur Meyre
55a39d4c26 test: run single lut conformance on tensors 
closes #574
refs #558
refs #665
 2021-11-17 09:11:46 +01:00
Benoit Chevallier-Mames
a2a61a079f feat(tracing): add support for ndarray functions 
ndarray.flatten
ndarray.__abs__
ndarray.__neg__, __pos__ and __invert__
ndarray.__rshift__ and __lshift_

refs #751
refs #218
 2021-11-16 18:36:14 +01:00
Umut
db098fd3a3 refactor(debugging): improve error messages of generic functions with floating point inputs 2021-11-16 13:18:13 +03:00
Umut
46a018fd21 refactor: use a consistent short name for operation graph 2021-11-16 12:10:48 +03:00
Benoit Chevallier-Mames
5d31aa4d2c chore: factorize some data_gen. 2021-11-15 16:46:46 +01:00
Benoit Chevallier-Mames
50c1ceb6db test(execution): add tests with negative results 
for now, results are correct mod 128, but wrong without this modulo reduction. Will be later fixed
by the compiler.

closes #844
refs #845
 2021-11-15 16:46:46 +01:00
Umut
239f66eb46 refactor(mlir): re-write mlir conversion 2021-11-15 17:21:51 +03:00
Umut
6fec590e65 refactor(debugging): re-write graph formatting 2021-11-15 16:33:15 +03:00
Umut
b449ddc360 refactor(debugging): improve error message of unfused astype 2021-11-15 12:50:46 +03:00
Benoit Chevallier-Mames
e7e7a02425 test(execution): add tests with negative values 
closes #844
 2021-11-15 10:15:49 +01:00
Benoit Chevallier-Mames
1394dd6db5 chore: clarify these tests a bit 2021-11-15 10:10:58 +01:00