fix: import forward_attention

This commit is contained in:
2026-06-03 05:41:33 +00:00
parent 91dfac34d8
commit 6459fbca9a

View File

@@ -42,7 +42,7 @@ def main():
from single_shot_inference import (
load_all_weights, make_nvfp4_linear, get_nvfp4_weight,
rmsnorm, unweighted_rmsnorm, _apply_rope, build_rope_cache,
KVCache, Compressor, Indexer, forward_layer, moe_forward,
KVCache, Compressor, Indexer, forward_layer, forward_attention, moe_forward,
_load_moe_weights_stacked, _load_shared_expert_weights,
_cache_layer_weights_no_experts,
)