Revert "temporary decompose for decode" (nod-ai#409)

Reverts nod-ai#353
renxida · Nov 2, 2024 · ce8b77f · ce8b77f
1 parent ad48f9d
commit ce8b77f
Showing 1 changed file with 0 additions and 1 deletion.
diff --git a/sharktank/sharktank/models/llama/llama.py b/sharktank/sharktank/models/llama/llama.py
@@ -269,7 +269,6 @@ def decode(
         for block_idx, block in enumerate(self.attn_blocks):
             if block_idx == 0:
                 self.trace_tensor(f"llama.attn_block.{block_idx}.input", h)
-            block.attn.attention_kernel = "decomposed"
             h = block(
                 h,
                 start_positions=start_positions,