github/tfhe-rs
1
1
Fork 0
mirror of https://github.com/zama-ai/tfhe-rs.git synced 2026-01-09 14:47:56 -05:00
Code Issues Packages Projects Releases Wiki Activity

chore(gpu): test with parameters that match first tfhe paper

Browse Source
This commit is contained in:
Guillermo Oyarzun
2025-08-29 12:00:45 +02:00
parent ff29535eb0
commit a12fad9425
2 changed files with 16 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
tfhe/src/shortint/parameters/v1_4/multi_bit/gaussian/p_fail_2_minus_128/ks_pbs_gpu.rs
View File

@@ -20,9 +20,12 @@ pub const V1_4_PARAM_GPU_MULTI_BIT_GROUP_3_MESSAGE_3_CARRY_3_KS_PBS_GAUSSIAN_2M1
/// p-fail = 2^-131.073, algorithmic cost ~ 10902, 2-norm = 17
pub const V1_4_PARAM_GPU_MULTI_BIT_GROUP_3_MESSAGE_4_CARRY_4_KS_PBS_GAUSSIAN_2M128: MultiBitPBSParameters = crate::shortint::parameters::v1_1::V1_1_PARAM_GPU_MULTI_BIT_GROUP_3_MESSAGE_4_CARRY_4_KS_PBS_GAUSSIAN_2M128;
/// p-fail = 2^-128.034, algorithmic cost ~ 76, 2-norm = 3
/// Parameters found gaussian: {'l_ks': 4, 'l_bs': 1, 'k': 1, 'N': 2048, 'y': 164, 'b_ks': 8, 'b_bs': 4194304}
// Failure probability: (225113738.63843, 68.541808, 1.0584085770481941e-10)
// pfail -33
pub const V1_4_PARAM_GPU_MULTI_BIT_GROUP_4_MESSAGE_1_CARRY_1_KS_PBS_GAUSSIAN_2M128:
MultiBitPBSParameters = MultiBitPBSParameters {
lwe_dimension: LweDimension(736),
lwe_dimension: LweDimension(656),
glwe_dimension: GlweDimension(1),
polynomial_size: PolynomialSize(2048),
lwe_noise_distribution: DynamicDistribution::new_gaussian_from_std_dev(StandardDev(

17
tfhe/src/shortint/parameters/v1_4/multi_bit/tuniform/p_fail_2_minus_128/ks_pbs_gpu.rs
View File

@@ -1,7 +1,7 @@
use crate::shortint::parameters::{
CarryModulus, CiphertextModulus, DecompositionBaseLog, DecompositionLevelCount,
DynamicDistribution, EncryptionKeyChoice, GlweDimension, LweBskGroupingFactor, LweDimension,
MaxNoiseLevel, MessageModulus, MultiBitPBSParameters, PolynomialSize,
MaxNoiseLevel, MessageModulus, MultiBitPBSParameters, PolynomialSize, StandardDev,
};
/// p-fail = 2^-136.056, algorithmic cost ~ 63, 2-norm = 3
pub const V1_4_PARAM_GPU_MULTI_BIT_GROUP_2_MESSAGE_1_CARRY_1_KS_PBS_TUNIFORM_2M128: MultiBitPBSParameters = crate::shortint::parameters::v1_1::V1_1_PARAM_GPU_MULTI_BIT_GROUP_2_MESSAGE_1_CARRY_1_KS_PBS_TUNIFORM_2M128;
@@ -20,18 +20,25 @@ pub const V1_4_PARAM_GPU_MULTI_BIT_GROUP_3_MESSAGE_3_CARRY_3_KS_PBS_TUNIFORM_2M1
/// p-fail = 2^-131.906, algorithmic cost ~ 11197, 2-norm = 17
pub const V1_4_PARAM_GPU_MULTI_BIT_GROUP_3_MESSAGE_4_CARRY_4_KS_PBS_TUNIFORM_2M128: MultiBitPBSParameters = crate::shortint::parameters::v1_1::V1_1_PARAM_GPU_MULTI_BIT_GROUP_3_MESSAGE_4_CARRY_4_KS_PBS_TUNIFORM_2M128;
/// p-fail = 2^-145.020, algorithmic cost ~ 79, 2-norm = 3
/// Parameters found gaussian: {'l_ks': 4, 'l_bs': 1, 'k': 1, 'N': 2048, 'y': 168, 'b_ks': 8, 'b_bs': 4194304}
//Failure probability: (230602839.88691, 70.21296, 2.634949763872057e-17)
// pfail -54
pub const V1_4_PARAM_GPU_MULTI_BIT_GROUP_4_MESSAGE_1_CARRY_1_KS_PBS_TUNIFORM_2M128:
MultiBitPBSParameters = MultiBitPBSParameters {
lwe_dimension: LweDimension(760),
lwe_dimension: LweDimension(672),
glwe_dimension: GlweDimension(1),
polynomial_size: PolynomialSize(2048),
lwe_noise_distribution: DynamicDistribution::new_t_uniform(49),
glwe_noise_distribution: DynamicDistribution::new_t_uniform(17),
lwe_noise_distribution: DynamicDistribution::new_gaussian_from_std_dev(StandardDev(
1.9276917419403266e-05,
)),
glwe_noise_distribution: DynamicDistribution::new_gaussian_from_std_dev(StandardDev(
2.845267479601915e-15,
)),
pbs_base_log: DecompositionBaseLog(22),
pbs_level: DecompositionLevelCount(1),
ks_base_log: DecompositionBaseLog(3),
ks_level: DecompositionLevelCount(4),
message_modulus: MessageModulus(2),
message_modulus: MessageModulus(2),
carry_modulus: CarryModulus(2),
max_noise_level: MaxNoiseLevel::new(3),
log2_p_fail: -145.020,