feat(platform): port Siracusa+RedMulE from pulp-platform/Deeploy#67#20

Open

runwangdl wants to merge 22 commits into

develfrom

feat/redmule

Commits on May 10, 2026

Add Siracusa_w_redmule platform (RedMulE accelerator integration)
runwangdl
committed
drop Float test fixtures
runwangdl
committed
Add CI skeleton for Siracusa_w_redmule (tiled)
runwangdl
committed
Run pre-commit autoformat over RedMule files
runwangdl
committed
ci: gate GAP9 workflows on upstream org to skip on forks
runwangdl
committed
ci: skip gh-pages publish on forks
runwangdl
committed
Populate Siracusa_w_redmule tiled kernel test matrix
runwangdl
committed
ci(redmule): force runwangdl/deeploy:redmule image on push/PR
runwangdl
committed
ci(redmule): authenticate container pulls for private ghcr.io image
runwangdl
committed
ci(redmule): add CCT_train tiled-L3 job to Siracusa+RedMulE CI

runwangdl
and
claude
committed
fix(redmule): lower bias-less Gemm to MatMul so CCT_train tiling stops crashing

runwangdl
and
claude
committed
fix(redmule+upstream-transpose): unblock CCT_train codegen end-to-end

runwangdl
and
claude
committed
fix(redmule): unmap Conv from RedMulE engine; drop weight-layout pass

runwangdl
and
claude
committed
style: pre-commit autoformat (yapf / isort / autoflake) + drop EOF blank line

runwangdl
and
claude
committed
ci(perf): expose GVSoC cycle counts for Siracusa(+RedMulE) training jobs

runwangdl
and
claude
committed
feat(redmule): RedMulE-accelerated FP32 Conv2d via im2col + matmul

runwangdl
and
claude
committed
style+ci: fix lint on 4517cc9; emit RedMulE cycle/speedup summary in Actions UI

runwangdl
and
claude
committed
feat(redmule): add ResNet8 + MobileNetV1 to training matrix; multi-model summary

runwangdl
and
claude
committed
feat(redmule): RedMulE-accelerated PWConvGradW / PWConvGradX kernels

runwangdl
and
claude
committed
fix(redmule): ship PWConvGradX-only path; defer PWConvGradW routing

runwangdl
and
claude
committed
feat(redmule): chunked PWConvGradW + scattered PWConvGradX; isolate kernel CI

runwangdl
and
claude
committed
fix(ci): empty L3_SINGLEBUFFER_TRAINING_MODELS broke pytest collection

runwangdl
and
claude
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat(platform): port Siracusa+RedMulE from pulp-platform/Deeploy#67#20

feat(platform): port Siracusa+RedMulE from pulp-platform/Deeploy#67#20
runwangdl wants to merge 22 commits into
develfrom
feat/redmule

Commits on May 10, 2026

Add Siracusa_w_redmule platform (RedMulE accelerator integration)

drop Float test fixtures

Add CI skeleton for Siracusa_w_redmule (tiled)

Run pre-commit autoformat over RedMule files

ci: gate GAP9 workflows on upstream org to skip on forks

ci: skip gh-pages publish on forks

Populate Siracusa_w_redmule tiled kernel test matrix

ci(redmule): force runwangdl/deeploy:redmule image on push/PR

ci(redmule): authenticate container pulls for private ghcr.io image

ci(redmule): add CCT_train tiled-L3 job to Siracusa+RedMulE CI

fix(redmule): lower bias-less Gemm to MatMul so CCT_train tiling stops crashing

fix(redmule+upstream-transpose): unblock CCT_train codegen end-to-end

fix(redmule): unmap Conv from RedMulE engine; drop weight-layout pass

style: pre-commit autoformat (yapf / isort / autoflake) + drop EOF blank line

ci(perf): expose GVSoC cycle counts for Siracusa(+RedMulE) training jobs

feat(redmule): RedMulE-accelerated FP32 Conv2d via im2col + matmul

style+ci: fix lint on 4517cc9; emit RedMulE cycle/speedup summary in Actions UI

feat(redmule): add ResNet8 + MobileNetV1 to training matrix; multi-model summary

feat(redmule): RedMulE-accelerated PWConvGradW / PWConvGradX kernels

fix(redmule): ship PWConvGradX-only path; defer PWConvGradW routing

feat(redmule): chunked PWConvGradW + scattered PWConvGradX; isolate kernel CI

fix(ci): empty L3_SINGLEBUFFER_TRAINING_MODELS broke pytest collection