Revert "[Inference] Adapt to Fused rotary (#5348)" (#5373)

This reverts commit 9f4ab2eb92.
2025-09-12 20:54:35 +00:00 · 2024-02-07 14:27:04 +08:00
parent 9f4ab2eb92
commit 8106ede07f
5 changed files with 22 additions and 161 deletions
--- a/colossalai/kernel/triton/kvcache_copy.py
+++ b/colossalai/kernel/triton/kvcache_copy.py
@@ -75,6 +75,7 @@ def copy_kv_to_blocked_cache(
    block_size = k_cache.size(-2)

    num_warps = 8 if head_dim > 128 else 4
+
    grid = (bsz, num_kv_heads)
    _copy_to_kvcache_seqlen1_kernel[grid](
        k,