Revert "temporary decompose for decode (#353)"

This reverts commit 0e93b64.
nod-ai · Nov 1, 2024 · 0930126 · 0930126
1 parent 614dafa
commit 0930126
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,