hide kv cache behind torch.compile

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

vllm/attention/layer.py

@@ -7,7 +7,8 @@
 
 from vllm.attention import AttentionMetadata, AttentionType
 from vllm.attention.selector import backend_name_to_enum, get_attn_backend
-from vllm.config import CacheConfig, get_current_vllm_config
+from vllm.config import (CacheConfig, LayerForwardContext,
+                         get_current_vllm_config)
 from vllm.forward_context import ForwardContext, get_forward_context
 from vllm.model_executor.layers.quantization.base_config import (
     QuantizationConfig)
@@ -117,7 +118,10 @@ def __init__(
         compilation_config = get_current_vllm_config().compilation_config
         if prefix in compilation_config.static_forward_context:
             raise ValueError(f"Duplicate layer name: {prefix}")
-        compilation_config.static_forward_context[prefix] = self
+        # use a placeholder kv cache tensor during init, which will be replaced
+        # after kv cache initialization
+        compilation_config.static_forward_context[
+            prefix] = LayerForwardContext(self, torch.tensor([]))
         self.layer_name = prefix
 
     def forward(
@@ -152,13 +156,11 @@ def forward(
             if value is not None:
                 value = value.view(-1, self.num_kv_heads, self.head_size)
             torch.ops.vllm.unified_attention_with_output(
-                query, key, value, output, kv_cache, attn_type,
-                self.layer_name)
+                query, key, value, output, attn_type, self.layer_name)
             return output.view(-1, hidden_size)
         else:
             return torch.ops.vllm.unified_attention(query, key, value,
-                                                    kv_cache, attn_type,
-                                                    self.layer_name)
+                                                    attn_type, self.layer_name)
 
     def extra_repr(self) -> str:
         s = f"head_size={self.impl.head_size}"  # type: ignore
@@ -236,17 +238,17 @@ def unified_attention(
     query: torch.Tensor,
     key: torch.Tensor,
     value: torch.Tensor,
-    kv_cache: torch.Tensor,
     attn_type: str,
     layer_name: str,
 ) -> torch.Tensor:
     forward_context: ForwardContext = get_forward_context()
-    attn_metadata = forward_context.dynamic_forward_context
-    self = forward_context.static_forward_context[layer_name]
+    attn_metadata = forward_context.attn_metadata
+    ctx = forward_context.layers[layer_name]
+    self = ctx.attn_module
     return self.impl.forward(query,
                              key,
                              value,
-                             kv_cache,
+                             ctx.kv_cache,
                              attn_metadata,
                              self._k_scale,
                              self._v_scale,
@@ -257,7 +259,6 @@ def unified_attention_fake(
     query: torch.Tensor,
     key: torch.Tensor,
     value: torch.Tensor,
-    kv_cache: torch.Tensor,
     attn_type: str,
     layer_name: str,
 ) -> torch.Tensor:
@@ -267,7 +268,7 @@ def unified_attention_fake(
 direct_register_custom_op(
     op_name="unified_attention",
     op_func=unified_attention,
-    mutates_args=["kv_cache"],
+    mutates_args=[],
     fake_impl=unified_attention_fake,
     dispatch_key=current_platform.dispatch_key,
 )
@@ -278,17 +279,17 @@ def unified_attention_with_output(
     key: torch.Tensor,
     value: torch.Tensor,
     output: torch.Tensor,
-    kv_cache: torch.Tensor,
     attn_type: str,
     layer_name: str,
 ) -> None:
     forward_context: ForwardContext = get_forward_context()
-    attn_metadata = forward_context.dynamic_forward_context
-    self = forward_context.static_forward_context[layer_name]
+    attn_metadata = forward_context.attn_metadata
+    ctx = forward_context.layers[layer_name]
+    self = ctx.attn_module
     self.impl.forward(query,
                       key,
                       value,
-                      kv_cache,
+                      ctx.kv_cache,
                       attn_metadata,
                       self._k_scale,
                       self._v_scale,
@@ -301,7 +302,6 @@ def unified_attention_with_output_fake(
     key: torch.Tensor,
     value: torch.Tensor,
     output: torch.Tensor,
-    kv_cache: torch.Tensor,
     attn_type: str,
     layer_name: str,
 ) -> None:
@@ -311,7 +311,7 @@ def unified_attention_with_output_fake(
 direct_register_custom_op(
     op_name="unified_attention_with_output",
     op_func=unified_attention_with_output,
-    mutates_args=["kv_cache", "output"],
+    mutates_args=["output"],
     fake_impl=unified_attention_with_output_fake,
     dispatch_key=current_platform.dispatch_key,
 )

vllm/config.py

@@ -2616,6 +2616,12 @@ class CompilationLevel:
     PIECEWISE = 3
 
 
+@dataclass
+class LayerForwardContext:
+    attn_module: Any  # vllm.attention.layer.Attention
+    kv_cache: Any  # torch.Tensor
+
+
 class CompilationConfig(BaseModel):
     """
     Configuration for compilation.
@@ -2769,9 +2775,9 @@ def model_post_init(self, __context: Any) -> None:
     inductor_hash_cache: Any = PrivateAttr
 
     # Per-model forward context
-    # Mainly used to store attention cls
-    # Map from layer name to the attention cls
-    static_forward_context: Dict[str, Any] = PrivateAttr
+    # Map from layer name to the layer's forward context, which stores
+    # attention cls and kv_cache
+    static_forward_context: Dict[str, LayerForwardContext] = PrivateAttr
 
     def compute_hash(self) -> str:
         """

vllm/forward_context.py

@@ -2,14 +2,17 @@
 from collections import defaultdict
 from contextlib import contextmanager
 from dataclasses import dataclass
-from typing import Any, Dict, Optional
+from typing import TYPE_CHECKING, Any, Dict, Optional
 
 import torch
 
 import vllm.envs as envs
-from vllm.config import VllmConfig
+from vllm.config import LayerForwardContext, VllmConfig
 from vllm.logger import init_logger
 
+if TYPE_CHECKING:
+    from vllm.attention.backends.abstract import AttentionMetadata
+
 logger = init_logger(__name__)
 
 track_batchsize: bool = envs.VLLM_LOG_BATCHSIZE_INTERVAL >= 0
@@ -21,9 +24,10 @@
 
 @dataclass
 class ForwardContext:
-    static_forward_context: Dict[str, Any]
+    # copy from vllm_config.compilation_config.static_forward_context
+    layers: Dict[str, LayerForwardContext]
     # TODO: extend to support per-layer dynamic forward context
-    dynamic_forward_context: Any
+    attn_metadata: "AttentionMetadata"  # set dynamically for each forward pass
 
 
 _forward_context: Optional[ForwardContext] = None
@@ -38,34 +42,32 @@ def get_forward_context() -> ForwardContext:
 
 
 @contextmanager
-def set_forward_context(context: Any, vllm_config: VllmConfig):
+def set_forward_context(attn_metadata: Any, vllm_config: VllmConfig):
     """A context manager that stores the current forward context,
     can be attention metadata, etc.
     Here we can inject common logic for every model forward pass.
     """
     global forward_start_time
-    need_to_track_batchsize = track_batchsize and context is not None
+    need_to_track_batchsize = track_batchsize and attn_metadata is not None
     if need_to_track_batchsize:
         forward_start_time = time.perf_counter()
     global _forward_context
     prev_context = _forward_context
     _forward_context = ForwardContext(
-        static_forward_context=vllm_config.compilation_config.
-        static_forward_context,
-        dynamic_forward_context=context)
+        layers=vllm_config.compilation_config.static_forward_context,
+        attn_metadata=attn_metadata)
     try:
         yield
     finally:
-        global batchsize_counter
         global last_logging_time, batchsize_logging_interval
         if need_to_track_batchsize:
-            if hasattr(context, "num_prefill_tokens"):
+            if hasattr(attn_metadata, "num_prefill_tokens"):
                 # for v0 attention backends
-                batchsize = context.num_prefill_tokens + \
-                    context.num_decode_tokens
+                batchsize = attn_metadata.num_prefill_tokens + \
+                    attn_metadata.num_decode_tokens
             else:
                 # for v1 attention backends
-                batchsize = context.num_input_tokens
+                batchsize = attn_metadata.num_input_tokens
             # we use synchronous scheduling right now,
             # adding a sync point here should not affect
             # scheduling of the next batch

vllm/v1/worker/gpu_model_runner.py

@@ -13,6 +13,7 @@
 from vllm.inputs import INPUT_REGISTRY
 from vllm.logger import init_logger
 from vllm.model_executor.model_loader import get_model
+from vllm.model_executor.models.utils import extract_layer_index
 from vllm.multimodal import MULTIMODAL_REGISTRY, MultiModalKwargs
 from vllm.sampling_params import SamplingType
 from vllm.utils import (STR_DTYPE_TO_TORCH_DTYPE, DeviceMemoryProfiler,
@@ -752,3 +753,13 @@ def initialize_kv_cache(self, num_blocks: int) -> None:
                 torch.zeros(kv_cache_shape,
                             dtype=self.kv_cache_dtype,
                             device=self.device))
+        # register kv_cache for forward_context
+        if self.vllm_config.parallel_config.pipeline_parallel_size > 1:
+            # TODO(Chen): In pipeline parallelism, layer_name 'layers.i.xxx'
+            # is mapped to kv_caches[i - start_layer_idx]. Need to implement
+            # and verify after supporting PP in v1
+            raise NotImplementedError("Pipeline parallelism is not supported.")
+        ctx = self.vllm_config.compilation_config.static_forward_context
+        for layer_name, forward_ctx in ctx.items():
+            layer_id = extract_layer_index(layer_name)
+            forward_ctx.kv_cache = self.kv_caches[layer_id]

vllm/worker/cache_engine.py

@@ -1,11 +1,13 @@
 """CacheEngine class for managing the KV cache."""
-from typing import List
+from typing import Dict, List
 
 import torch
 
 from vllm.attention import get_attn_backend
-from vllm.config import CacheConfig, DeviceConfig, ModelConfig, ParallelConfig
+from vllm.config import (CacheConfig, CompilationConfig, DeviceConfig,
+                         LayerForwardContext, ModelConfig, ParallelConfig)
 from vllm.logger import init_logger
+from vllm.model_executor.models.utils import extract_layer_index
 from vllm.utils import (STR_DTYPE_TO_TORCH_DTYPE, LayerBlockType,
                         get_dtype_size, is_pin_memory_available)
 
@@ -26,6 +28,7 @@ def __init__(
         model_config: ModelConfig,
         parallel_config: ParallelConfig,
         device_config: DeviceConfig,
+        compilation_config: CompilationConfig,
     ) -> None:
         self.cache_config = cache_config
         self.model_config = model_config
@@ -62,6 +65,7 @@ def __init__(
         self.gpu_cache = self._allocate_kv_cache(
             self.num_gpu_blocks, self.device_config.device_type)
         self.cpu_cache = self._allocate_kv_cache(self.num_cpu_blocks, "cpu")
+        self._register_gpu_kv_cache(compilation_config.static_forward_context)
 
     def _allocate_kv_cache(
         self,
@@ -84,6 +88,14 @@ def _allocate_kv_cache(
                             device=device))
         return kv_cache
 
+    def _register_gpu_kv_cache(self, ctx: Dict[str,
+                                               LayerForwardContext]) -> None:
+        if self.parallel_config.pipeline_parallel_size > 1:
+            raise NotImplementedError
+        for layer_name, forward_ctx in ctx.items():
+            layer_id = extract_layer_index(layer_name)
+            forward_ctx.kv_cache = self.gpu_cache[layer_id]
+
     def swap_in(self, src_to_dst: torch.Tensor) -> None:
         for i in range(self.num_attention_layers):
             self.attn_backend.swap_blocks(self.cpu_cache[i], self.gpu_cache[i],

vllm/worker/hpu_worker.py

@@ -208,7 +208,8 @@ def _init_cache_engine(self):
         assert self.cache_config.num_gpu_blocks is not None
         self.cache_engine = [
             HPUCacheEngine(self.cache_config, self.model_config,
-                           self.parallel_config, self.device_config)
+                           self.parallel_config, self.device_config,
+                           self.compilation_config)
             for _ in range(self.parallel_config.pipeline_parallel_size)
         ]
         self.hpu_cache = [

vllm/worker/worker.py

@@ -278,7 +278,8 @@ def _init_cache_engine(self):
         assert self.cache_config.num_gpu_blocks is not None
         self.cache_engine = [
             CacheEngine(self.cache_config, self.model_config,
-                        self.parallel_config, self.device_config)
+                        self.parallel_config, self.device_config,
+                        self.compilation_config)
             for _ in range(self.parallel_config.pipeline_parallel_size)
         ]
         self.gpu_cache = [

vllm/worker/worker_base.py

@@ -43,6 +43,7 @@ def __init__(
         self.prompt_adapter_config = vllm_config.prompt_adapter_config
         self.observability_config = vllm_config.observability_config
         self.kv_transfer_config = vllm_config.kv_transfer_config
+        self.compilation_config = vllm_config.compilation_config
         from vllm.platforms import current_platform
         self.current_platform = current_platform