fix(ci): resolve 6 real CI failures + DiT / weight-init vectorization#1156

Merged

ooples merged 29 commits into

masterfrom

fix/ci-master-test-failures

Apr 19, 2026

Commits on Apr 18, 2026

fix(stats): break BasicStats.CalculateStats recursion that crashed test host
ooples
committed
test(data): cross-platform retry trigger for RobustFileOps tests
ooples
committed
fix(serialization): match MultiHeadAttentionLayer 5-arg constructor in deserializer
ooples
committed
fix(optimizer): re-allocate Adam moments when cached shape mismatches param
ooples
committed
fix(serving): cross-platform sanitizer for AesGcm artifact filenames
ooples
committed
fix(layers): sparselinearlayer reports supportstraining true

ooples
and
claude
committed
perf(dit): vectorize Patchify/Unpatchify/AdaLN via Engine reshape+permute

ooples
and
claude
committed
perf(init): batched parallel Xavier normal weight initialization

ooples
and
claude
committed
chore(deps): bump aidotnet.tensors 0.46.0 -> 0.46.1

ooples
and
claude
committed
fix(stats): break EnsureFullStatsComputed recursion in errorstats/modelstats/predictionstats

ooples
and
claude
committed
fix(networks): resnet/vgg train adds batch dim for 3d input

ooples
and
claude
committed
fix(networks): mobilenetv2 handles 3d input in forward/train/namedactivations

ooples
and
claude
committed
test(networks): instructorembedding test shape matches 768-dim model

ooples
and
claude
committed
fix(networks): convolutionalneuralnetwork train adds batch dim for 3d input

ooples
and
claude
committed
fix(networks): unet3d decoder channel count + test output shape

ooples
and
claude
committed
fix(gp): escalating cholesky jitter for sparsegaussianprocess.fit

ooples
and
claude
committed
fix(generators): correct audio/video modeldomain ordinal in testscaffoldgenerator

ooples
and
claude
committed
test(neuralnetworks): align word2vec test shapes with softmax vocab head
ooples
committed
test(ner): emit 768-dim scaffolded shapes for transformer ner models
ooples
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

fix(ci): resolve 6 real CI failures + DiT / weight-init vectorization#1156

fix(ci): resolve 6 real CI failures + DiT / weight-init vectorization#1156
ooples merged 29 commits into
masterfrom
fix/ci-master-test-failures

Commits on Apr 18, 2026

fix(stats): break BasicStats.CalculateStats recursion that crashed test host

test(data): cross-platform retry trigger for RobustFileOps tests

fix(serialization): match MultiHeadAttentionLayer 5-arg constructor in deserializer

fix(optimizer): re-allocate Adam moments when cached shape mismatches param

fix(serving): cross-platform sanitizer for AesGcm artifact filenames

fix(layers): sparselinearlayer reports supportstraining true

perf(dit): vectorize Patchify/Unpatchify/AdaLN via Engine reshape+permute

perf(init): batched parallel Xavier normal weight initialization

chore(deps): bump aidotnet.tensors 0.46.0 -> 0.46.1

fix(stats): break EnsureFullStatsComputed recursion in errorstats/modelstats/predictionstats

fix(networks): resnet/vgg train adds batch dim for 3d input

fix(networks): mobilenetv2 handles 3d input in forward/train/namedactivations

test(networks): instructorembedding test shape matches 768-dim model

fix(networks): convolutionalneuralnetwork train adds batch dim for 3d input

fix(networks): unet3d decoder channel count + test output shape

fix(gp): escalating cholesky jitter for sparsegaussianprocess.fit

fix(generators): correct audio/video modeldomain ordinal in testscaffoldgenerator

test(neuralnetworks): align word2vec test shapes with softmax vocab head

test(ner): emit 768-dim scaffolded shapes for transformer ner models

Commits on Apr 19, 2026

fix(layers): default rnn head should use identityactivation, not relu-via-null

fix(memorynetwork): seed memory and wire training through the memory-aware flow

fix(quantumnn): migrate training to trainwithtape and use identity on final dense

fix(diffusion): upscaleavideo inputconv should match latent channels, not concat width

fix(diffusion): videounet spatial resblock must mix channels, not width

fix(serialization): register memoryread and memorywrite layers for deserialization

fix(gp): sparsegp ky solve falls back to svd pseudoinverse when cholesky gives up

fix(regression): poisson irls must not overwrite coefficients with nan/inf

fix(regression): rbf solve via tikhonov-damped svd instead of normal-equations inverse

fix: address 10 CodeRabbit review comments on PR #1156