Profile replicated router without TP group

This commit is contained in:
2026-07-16 22:49:31 +08:00
parent fcca259475
commit 414838a799
2 changed files with 4 additions and 3 deletions

View File

@@ -115,6 +115,7 @@ def main() -> None:
bias=False,
quant_config=None,
prefix="model.layers.0.mlp.gate",
disable_tp=True,
).to(device=device, dtype=torch.bfloat16)
gate.weight.data.uniform_(-0.01, 0.01)