[Bugfix] Max concurrency estimation and check_enough_kv_cache_memory for models with sliding window layers (vllm-project#19029)

Signed-off-by: Chen Zhang <[email protected]>

Author: heheda12345

tests/v1/core/test_kv_cache_utils.py

@@ -12,13 +12,11 @@
 from vllm.v1.core.kv_cache_manager import KVCacheManager
 # disable yapf here as it formats differently than isort such that both fail
 # yapf: disable
-from vllm.v1.core.kv_cache_utils import (FreeKVCacheBlockQueue, KVCacheBlock,
-                                         PrefixCachingMetrics,
-                                         estimate_max_model_len,
-                                         generate_block_hash_extra_keys,
-                                         hash_block_tokens,
-                                         hash_request_tokens,
-                                         unify_kv_cache_configs)
+from vllm.v1.core.kv_cache_utils import (
+    FreeKVCacheBlockQueue, KVCacheBlock, PrefixCachingMetrics,
+    estimate_max_model_len, generate_block_hash_extra_keys,
+    get_max_concurrency_for_kv_cache_config, hash_block_tokens,
+    hash_request_tokens, unify_kv_cache_configs)
 from vllm.v1.kv_cache_interface import (FullAttentionSpec, KVCacheConfig,
                                         KVCacheGroupSpec, KVCacheTensor,
                                         SlidingWindowSpec)
@@ -597,6 +595,84 @@ def test_estimate_max_model_len(model_id, max_model_len,
     assert estimated_max_len == want_estimated_max_len
 
 
+def test_get_max_concurrency_for_kv_cache_config():
+    # Create a VllmConfig
+    model_id = "Qwen/Qwen1.5-7B"
+    max_model_len = 16384
+    model_config = ModelConfig(
+        model_id,
+        task="generate",
+        tokenizer=model_id,
+        tokenizer_mode="auto",
+        trust_remote_code=False,
+        seed=0,
+        dtype="float16",
+        max_model_len=max_model_len,
+    )
+    scheduler_config = SchedulerConfig(max_num_batched_tokens=1024,
+                                       enable_chunked_prefill=True)
+
+    vllm_config = VllmConfig(
+        model_config=model_config,
+        scheduler_config=scheduler_config,
+    )
+
+    full_attention_spec = FullAttentionSpec(
+        block_size=16,
+        num_kv_heads=32,
+        head_size=128,
+        dtype=torch.float16,
+        use_mla=False,
+    )
+
+    sliding_window_spec = SlidingWindowSpec(
+        block_size=16,
+        num_kv_heads=32,
+        head_size=128,
+        dtype=torch.float16,
+        use_mla=False,
+        sliding_window=1024,
+    )
+
+    kv_cache_config_full_attention = KVCacheConfig(
+        num_blocks=int(1024 * 1.5),
+        tensors={},
+        kv_cache_groups=[
+            KVCacheGroupSpec([f"layer_{i}" for i in range(32)],
+                             full_attention_spec),
+        ],
+    )
+    max_concurrency_full_attention = get_max_concurrency_for_kv_cache_config(
+        vllm_config, kv_cache_config_full_attention)
+    assert max_concurrency_full_attention == 1.5
+
+    kv_cache_config_sliding_window = KVCacheConfig(
+        num_blocks=129 * 3,
+        tensors={},
+        kv_cache_groups=[
+            KVCacheGroupSpec([f"layer_{i}" for i in range(32)],
+                             sliding_window_spec),
+        ],
+    )
+    max_concurrency_sliding_window = get_max_concurrency_for_kv_cache_config(
+        vllm_config, kv_cache_config_sliding_window)
+    assert max_concurrency_sliding_window == 3
+
+    kv_cache_config_hybrid_model = KVCacheConfig(
+        num_blocks=(1024 + 129) * 3,
+        tensors={},
+        kv_cache_groups=[
+            KVCacheGroupSpec([f"layer_{i}" for i in range(32)],
+                             full_attention_spec),
+            KVCacheGroupSpec([f"layer_{i}" for i in range(32, 64)],
+                             sliding_window_spec),
+        ],
+    )
+    max_concurrency_hybrid_model = get_max_concurrency_for_kv_cache_config(
+        vllm_config, kv_cache_config_hybrid_model)
+    assert max_concurrency_hybrid_model == 3
+
+
 def test_allocate_with_lookahead():
     """Verify that lookahead tokens correctly affect block allocation"""
     block_size = 4

vllm/v1/core/kv_cache_utils.py

@@ -3,13 +3,13 @@
 """KV-Cache Utilities."""
 import os
 from collections import deque
-from collections.abc import Sequence
+from collections.abc import Iterable, Sequence
 from dataclasses import dataclass
 from typing import Any, Callable, NamedTuple, Optional
 
 from vllm.config import VllmConfig
 from vllm.logger import init_logger
-from vllm.utils import GiB_bytes, sha256
+from vllm.utils import GiB_bytes, cdiv, sha256
 from vllm.v1.kv_cache_interface import (FullAttentionSpec, KVCacheConfig,
                                         KVCacheGroupSpec, KVCacheSpec,
                                         KVCacheTensor, SlidingWindowSpec)
@@ -468,6 +468,15 @@ def hash_request_tokens(hash_function: Any, block_size: int,
     return ret
 
 
+def max_memory_usage_bytes(vllm_config: VllmConfig,
+                           kv_cache_specs: Iterable[KVCacheSpec]) -> int:
+    """
+    Get the maximum memory usage in bytes for the given KV cache specs.
+    """
+    return sum(
+        spec.max_memory_usage_bytes(vllm_config) for spec in kv_cache_specs)
+
+
 def estimate_max_model_len(vllm_config: VllmConfig,
                            kv_cache_spec: dict[str, KVCacheSpec],
                            available_memory: int) -> int:
@@ -489,11 +498,8 @@ def fits_in_memory(model_len: int) -> bool:
         # Modify the max_model_len for this calculation
         vllm_config.model_config.max_model_len = model_len
         # Calculate memory needed for the given model length
-        memory_needed = sum(
-            (layer_spec.max_memory_usage_bytes(vllm_config)
-             for layer_spec in kv_cache_spec.values()),
-            start=0,
-        )
+        memory_needed = max_memory_usage_bytes(vllm_config,
+                                               kv_cache_spec.values())
         return memory_needed <= available_memory
 
     # Binary search for the maximum model length
@@ -538,9 +544,7 @@ def check_enough_kv_cache_memory(vllm_config: VllmConfig,
                          "initializing the engine.")
 
     max_model_len = vllm_config.model_config.max_model_len
-    needed_memory = 0
-    for layer_spec in kv_cache_spec.values():
-        needed_memory += layer_spec.max_memory_usage_bytes(vllm_config)
+    needed_memory = max_memory_usage_bytes(vllm_config, kv_cache_spec.values())
 
     if needed_memory > available_memory:
         # Estimate the maximum model length that can fit in the available memory
@@ -606,6 +610,24 @@ def is_kv_cache_type_uniform(kv_cache_spec: dict[str, KVCacheSpec]) -> bool:
     return len(layer_keys) == 1
 
 
+def get_max_concurrency_for_kv_cache_config(
+        vllm_config: VllmConfig, kv_cache_config: KVCacheConfig) -> float:
+    """
+    Get the maximum concurrency for the given KV cache configuration.
+    """
+    num_layer_per_group = max(
+        len(group.layer_names) for group in kv_cache_config.kv_cache_groups)
+    max_memory_usage_per_request = num_layer_per_group * max_memory_usage_bytes(
+        vllm_config,
+        (group.kv_cache_spec for group in kv_cache_config.kv_cache_groups))
+    memory_per_block = kv_cache_config.kv_cache_groups[
+        0].kv_cache_spec.page_size_bytes * num_layer_per_group
+    num_block_per_request = cdiv(max_memory_usage_per_request,
+                                 memory_per_block)
+    max_concurrency = kv_cache_config.num_blocks / num_block_per_request
+    return max_concurrency
+
+
 def _get_kv_cache_config_uniform_type(vllm_config: VllmConfig,
                                       kv_cache_spec: dict[str, KVCacheSpec],
                                       available_memory: int) -> KVCacheConfig:
@@ -637,14 +659,6 @@ def _get_kv_cache_config_uniform_type(vllm_config: VllmConfig,
             "num_gpu_blocks_override=%d", num_blocks, num_gpu_blocks_override)
         num_blocks = num_gpu_blocks_override
 
-    num_tokens = num_blocks * vllm_config.cache_config.block_size
-    num_tokens_str = f"{num_tokens:,}"
-    logger.info("GPU KV cache size: %s tokens", num_tokens_str)
-    max_model_len_str = f"{vllm_config.model_config.max_model_len:,}"
-    max_concurrency = num_tokens / vllm_config.model_config.max_model_len
-    logger.info("Maximum concurrency for %s tokens per request: %.2fx",
-                max_model_len_str, max_concurrency)
-
     per_layer_size = page_size * num_blocks
     # All layers have the same KV cache spec, so we create one kv cache group
     # for all layers.
@@ -659,6 +673,15 @@ def _get_kv_cache_config_uniform_type(vllm_config: VllmConfig,
         kv_cache_groups=create_kv_cache_group_specs(kv_cache_spec,
                                                     grouped_layer_names),
     )
+
+    num_tokens = num_blocks * vllm_config.cache_config.block_size
+    num_tokens_str = f"{num_tokens:,}"
+    logger.info("GPU KV cache size: %s tokens", num_tokens_str)
+    max_model_len_str = f"{vllm_config.model_config.max_model_len:,}"
+    max_concurrency = get_max_concurrency_for_kv_cache_config(
+        vllm_config, kv_cache_config)
+    logger.info("Maximum concurrency for %s tokens per request: %.2fx",
+                max_model_len_str, max_concurrency)
     return kv_cache_config
 
 
@@ -705,8 +728,8 @@ def get_kv_cache_config(vllm_config: VllmConfig,
     Returns:
         The generated KVCacheConfigs
     """
-    check_enough_kv_cache_memory(vllm_config, kv_cache_spec, available_memory)
     unify_hybrid_kv_cache_specs(kv_cache_spec)
+    check_enough_kv_cache_memory(vllm_config, kv_cache_spec, available_memory)
     if is_kv_cache_type_uniform(kv_cache_spec):
         # KV cache of all layers are the same, which is true for
         # most models. Allocate the same amount of memory for