[Inference/opt] Fused KVCahce Memcopy (#5374)

* fused kv memcopy

* add TODO in test_kvcache_copy.py

colossalai/inference/modeling/models/nopadding_llama.py

@@ -301,8 +301,9 @@ class NopadLlamaAttention(LlamaAttention):
                 sm_scale=sm_scale,
             )
         else:
-            copy_kv_to_blocked_cache(key_states, k_cache, kv_lengths=sequence_lengths, block_tables=block_tables)
-            copy_kv_to_blocked_cache(value_states, v_cache, kv_lengths=sequence_lengths, block_tables=block_tables)
+            copy_kv_to_blocked_cache(
+                key_states, value_states, k_cache, v_cache, kv_lengths=sequence_lengths, block_tables=block_tables
+            )
             attn_output = flash_decoding_attention(
                 q=query_states,
                 k_cache=k_cache,

colossalai/inference/modeling/models/padding_llama.py

@@ -356,8 +356,9 @@ class PadLlamaAttention(LlamaAttention):
                 if attention_mask is not None:
                     attn_output = pad_input(attn_output, indices, bsz, q_len)
             else:
-                copy_kv_to_blocked_cache(key_states, k_cache, kv_lengths=sequence_lengths, block_tables=block_tables)
-                copy_kv_to_blocked_cache(value_states, v_cache, kv_lengths=sequence_lengths, block_tables=block_tables)
+                copy_kv_to_blocked_cache(
+                    key_states, value_states, k_cache, v_cache, kv_lengths=sequence_lengths, block_tables=block_tables
+                )
                 attn_output = flash_decoding_attention(
                     q=query_states,
                     k_cache=k_cache,

colossalai/kernel/triton/kvcache_copy.py

@@ -6,17 +6,26 @@ import triton.language as tl
 # Triton 2.1.0
 @triton.jit
 def _copy_to_kvcache_seqlen1_kernel(
-    KV,  # K or V
-    KVCache,  # KCache or VCache
+    K,  # K
+    V,  # V
+    KCache,  # KCache
+    VCache,  # VCache
     BLOCK_TABLES,
     context_lengths,
     stride_kt,
     stride_kh,
     stride_kd,
-    stride_cacheb,
-    stride_cacheh,
-    stride_cachebs,
-    stride_cached,
+    stride_vt,
+    stride_vh,
+    stride_vd,
+    stride_cachekb,
+    stride_cachekh,
+    stride_cachekbs,
+    stride_cachekd,
+    stride_cachevb,
+    stride_cachevh,
+    stride_cachevbs,
+    stride_cachevd,
     stride_bts,
     stride_btb,
     block_size,
@@ -32,20 +41,33 @@ def _copy_to_kvcache_seqlen1_kernel(
     offsets_in_last_block = past_kv_seq_len % block_size
     offsets_dmodel = tl.arange(0, HEAD_DIM)
     offsets_kv = cur_seq_idx * stride_kt + cur_kv_head_idx * stride_kh + offsets_dmodel * stride_kd
-    kv = tl.load(KV + offsets_kv)
+
+    k = tl.load(K + offsets_kv)
+    v = tl.load(V + offsets_kv)
+
     offsets_kvcache = (
-        block_id * stride_cacheb
-        + cur_kv_head_idx * stride_cacheh
-        + offsets_in_last_block * stride_cachebs
-        + offsets_dmodel * stride_cached
+        block_id * stride_cachekb
+        + cur_kv_head_idx * stride_cachekh
+        + offsets_in_last_block * stride_cachekbs
+        + offsets_dmodel * stride_cachekd
     )
-    tl.store(KVCache + offsets_kvcache, kv)
+    offsets_kvcache = (
+        block_id * stride_cachevb
+        + cur_kv_head_idx * stride_cachevh
+        + offsets_in_last_block * stride_cachevbs
+        + offsets_dmodel * stride_cachevd
+    )
+
+    tl.store(KCache + offsets_kvcache, k)
+    tl.store(VCache + offsets_kvcache, v)
     return
 
 
 def copy_kv_to_blocked_cache(
     k: torch.Tensor,
+    v: torch.Tensor,
     k_cache: torch.Tensor,
+    v_cache: torch.Tensor,
     kv_lengths: torch.Tensor,
     block_tables: torch.Tensor,
 ):
@@ -53,16 +75,23 @@ def copy_kv_to_blocked_cache(
     Copy keys or values to the blocked key/value cache during decoding stage.
 
     Args:
-        k (torch.Tensor): [bsz, 1, num_kv_heads, head_dim]/[bsz, num_kv_heads, head_dim] - Keys or values during decoding with seq len 1.
-        k_cache (torch.Tensor): [num_blocks, num_kv_heads, block_size, head_dim] - Blocked key or value cache.
+        k (torch.Tensor): [bsz, 1, num_kv_heads, head_dim]/[bsz, num_kv_heads, head_dim] - Keys during decoding with seq len 1.
+        v (torch.Tensor): [bsz, 1, num_kv_heads, head_dim]/[bsz, num_kv_heads, head_dim] - Values during decoding with seq len 1.
+        k_cache (torch.Tensor): [num_blocks, num_kv_heads, block_size, head_dim] - Blocked key cache.
+        v_cache (torch.Tensor): [num_blocks, num_kv_heads, block_size, head_dim] - Blocked value cache.
         kv_lengths (torch.Tensor): [bsz] - Past key/value sequence lengths plus current sequence length for each sequence.
         block_tables (torch.Tensor): [bsz, max_blocks_per_sequence] - Block tables for each sequence.
     """
     assert k.size(-1) == k_cache.size(-1), "Incompatible head dim"
     assert k.dtype == k_cache.dtype, "Expected consistent dtype for tensor and cache."
-
     k = k.squeeze(1) if k.dim() == 4 else k
     assert k.dim() == 3, f"Incompatible k dim {k.dim()}"
+
+    assert v.size(-1) == v_cache.size(-1), "Incompatible head dim"
+    assert v.dtype == v_cache.dtype, "Expected consistent dtype for tensor and cache."
+    v = v.squeeze(1) if v.dim() == 4 else v
+    assert v.dim() == 3, f"Incompatible v dim {v.dim()}"
+
     bsz, num_kv_heads, head_dim = k.shape
 
     assert kv_lengths.shape[0] == block_tables.shape[0] == bsz, (
@@ -75,20 +104,28 @@ 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,
+        v,
         k_cache,
+        v_cache,
         block_tables,
         kv_lengths,
         k.stride(0),
         k.stride(1),
         k.stride(2),
+        v.stride(0),
+        v.stride(1),
+        v.stride(2),
         k_cache.stride(0),
         k_cache.stride(1),
         k_cache.stride(2),
         k_cache.stride(3),
+        v_cache.stride(0),
+        v_cache.stride(1),
+        v_cache.stride(2),
+        v_cache.stride(3),
         block_tables.stride(0),
         block_tables.stride(1),
         block_size,

tests/test_infer/test_ops/triton/test_kvcache_copy.py

@@ -44,18 +44,19 @@ def prepare_data(
     kv_unpad = torch.empty(size=kv_size, dtype=dtype, device=device).normal_(mean=0.0, std=0.5)
     k_unpad, v_unpad = torch.split(kv_unpad, [num_kv_heads, num_kv_heads], dim=-2)
 
-    k_cache, _, block_tables = generate_caches_and_block_tables_v2(
+    k_cache, v_cache, block_tables = generate_caches_and_block_tables_v2(
         k_unpad, v_unpad, past_kv_seq_lengths, bsz, max_num_blocks_per_seq, block_size, dtype=dtype, device=device
     )
     block_tables = block_tables.to(device=device)
 
     new_k = torch.randn((bsz, 1, num_kv_heads, head_dim), dtype=dtype, device=device)
+    new_v = torch.randn((bsz, 1, num_kv_heads, head_dim), dtype=dtype, device=device)
     # mock allocating blocks for the new k/v and update block tables
     mock_alloc_single_token(block_tables, past_kv_seq_lengths, block_size)
     # kv seq len = past kv seq len + seq len (1 during decoding stage)
     kv_seq_lengths = past_kv_seq_lengths + 1
 
-    return new_k, k_cache, kv_seq_lengths, block_tables
+    return new_k, new_v, k_cache, v_cache, kv_seq_lengths, block_tables
 
 
 @pytest.mark.skipif(not (HAS_TRITON and TRITON_CUDA_SUPPORT), reason="requires triton")
@@ -80,7 +81,7 @@ def test_copy_kv_to_caches(
     dtype = torch.float16
     device = get_current_device()
 
-    new_k, k_cache, kv_seq_lengths, block_tables = prepare_data(
+    new_k, new_v, k_cache, v_cache, kv_seq_lengths, block_tables = prepare_data(
         bsz,
         num_kv_heads,
         HEAD_DIM,
@@ -93,16 +94,20 @@ def test_copy_kv_to_caches(
     )
     # k_cache_torch = k_cache.clone().detach()
     # copy_to_cache(new_k, k_cache_torch, lengths=kv_seq_lengths, block_tables=block_tables, type="decoding")
-    copy_kv_to_blocked_cache(new_k, k_cache, kv_seq_lengths, block_tables)
+    copy_kv_to_blocked_cache(new_k, new_v, k_cache, v_cache, kv_seq_lengths, block_tables)
 
     past_kv_seq_len = kv_seq_lengths - 1
     target_block_ids = block_tables[range(0, block_tables.size(0)), past_kv_seq_len // block_size]
     offsets_in_block = past_kv_seq_len % block_size
-    target = k_cache[target_block_ids, :, offsets_in_block, :]
-    source = new_k.squeeze()
+    k_target = k_cache[target_block_ids, :, offsets_in_block, :]
+    k_source = new_k.squeeze()
+    v_target = v_cache[target_block_ids, :, offsets_in_block, :]
+    v_source = new_v.squeeze()
 
-    assert target.shape == source.shape
-    assert torch.equal(target, source)
+    assert k_target.shape == k_source.shape
+    assert torch.equal(k_target, k_source)
+    assert v_target.shape == v_source.shape
+    assert torch.equal(v_target, v_source)
     # target_torch = k_cache_copy[target_block_ids, :, offsets_in_block, :]
     # assert target_torch.shape == source.shape
     # assert torch.equal(target_torch, source)
@@ -143,7 +148,7 @@ def benchmark_kvcache_copy(
 
     assert KV_SEQ_LEN <= max_seq_len, "Assigned maximum kv length must be smaller or equal to maximum seq len"
 
-    new_k, k_cache, context_lengths, block_tables = prepare_data(
+    new_k, new_v, k_cache, v_cache, context_lengths, block_tables = prepare_data(
         bsz,
         num_kv_heads,
         HEAD_DIM,
@@ -156,10 +161,11 @@ def benchmark_kvcache_copy(
     )
 
     quantiles = [0.5, 0.2, 0.8]
+    # TODO copy_to_cache needs to support copying both k and v at the same time in the future.
     if provider == "torch_copy_func":
         fn = lambda: copy_to_cache(new_k, k_cache, lengths=context_lengths, block_tables=block_tables, type="decoding")
     if provider == "triton_copy_func":
-        fn = lambda: copy_kv_to_blocked_cache(new_k, k_cache, context_lengths, block_tables)
+        fn = lambda: copy_kv_to_blocked_cache(new_k, new_v, k_cache, v_cache, context_lengths, block_tables)
 
     ms, min_ms, max_ms = triton.testing.do_bench(fn, warmup=WARM_UPS, rep=REPS, quantiles=quantiles)
     return ms, min_ms, max_ms