You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit b659712
Browse filesBrowse the repository at this point in the historyBrowse files
[Espressif] Fuse linear ReLU and HardTanh patterns
Quantize single-consumer linear followed by ReLU or HardTanh, including
ReLU6, at the pattern boundaries. Fold the final output qparams and clamp
bounds into the existing ESP-NN fully connected call, preserving per-channel
weights and derived int32 bias. Forked and separately quantized activations
retain their original quantization boundaries.
Read the quantizer and activation fold before the linear lowering and
schema/wrapper updates. Shared dialect and S3 cases cover bias/no bias,
forks, saturation, vector and higher-rank inputs, and narrower clamp grids
with asymmetric, positive-only and negative-only HardTanh bounds.
Authored with OpenAI Codex.
0 commit comments