【训练营】小模型训练支持 - #232

Open

gavin-richie wants to merge 14 commits into

InfiniTensor:masterfrom

gavin-richie:feat/small-model-support

Commits on Sep 6, 2026

feat: add ReLU activation operator and module
gavin-richie
committed
76e6c98
feat: add Conv2d operator with im2col and GEMM kernels
gavin-richie
committed
0e76810
feat: add nn::Flatten module
gavin-richie
committed
5f4742f
fix: recompute MNIST image stride after float32 conversion
gavin-richie
committed
5f887c6
feat: add CNN training demo to the MNIST example
gavin-richie
committed
12aef40
test: cover Conv2d and ReLU forward and backward
gavin-richie
committed
e1cf0d6
build: add sm_89 to the CUDA architecture list
gavin-richie
committed
5805794
docs: document the MNIST CNN demo options
gavin-richie
committed
16c92e7

Commits on Sep 7, 2026

fix: sum the correct axis in the CUDA LinearBackwardBias kernel
gavin-richie
committed
f801f74
feat: add optional metrics logging to the mnist demo
gavin-richie
committed
ed7e912
chore: add swanlab metrics upload script
gavin-richie
committed
76c9600

Commits on Sep 8, 2026

build: add sm_86 to the CUDA architecture list
gavin-richie
committed
15a6265
feat: support DDP training in the mnist demo
gavin-richie
committed
bfb0333
docs: document mnist DDP launch
gavin-richie
committed
a8674d7