[Core] Support global prefix caching

examples/offline_inference_with_global_prefix.py

@@ -0,0 +1,124 @@
+from vllm import LLM, SamplingParams
+from vllm.distributed import cleanup_dist_env_and_memory
+
+# NOTE: This is just a running example. For benchmarking purpose,
+# please see benchmarks/benchmark_prefix_caching.py
+
+# Common prefix.
+prefix = (
+    "You are an expert school principal, skilled in effectively managing "
+    "faculty and staff. Draft 10-15 questions for a potential first grade "
+    "Head Teacher for my K-12, all-girls', independent school that emphasizes "
+    "community, joyful discovery, and life-long learning. The candidate is "
+    "coming in for a first-round panel interview for a 8th grade Math "
+    "teaching role. They have 5 years of previous teaching experience "
+    "as an assistant teacher at a co-ed, public school with experience "
+    "in middle school math teaching. Based on these information, fulfill "
+    "the following paragraph: ")
+
+# Sample prompts.
+prompts = [
+    "Hello, my name is",
+    "The president of the United States is",
+    "The capital of France is",
+    "The future of AI is",
+]
+
+generating_prompts = [prefix + prompt for prompt in prompts]
+
+# Create a sampling params object.
+sampling_params = SamplingParams(temperature=0.0)
+
+# Create an LLM without prefix caching as a baseline.
+regular_llm = LLM(model="facebook/opt-125m", gpu_memory_utilization=0.4)
+
+print("Results without `enable_prefix_caching`")
+
+# Generate texts from the prompts. The output is a list of RequestOutput objects
+# that contain the prompt, generated text, and other information.
+outputs = regular_llm.generate(generating_prompts, sampling_params)
+
+regular_generated_texts = []
+# Print the outputs.
+for output in outputs:
+    prompt = output.prompt
+    generated_text = output.outputs[0].text
+    regular_generated_texts.append(generated_text)
+    print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
+
+print("-" * 80)
+
+# Destroy the LLM object and free up the GPU memory.
+del regular_llm
+cleanup_dist_env_and_memory()
+
+# Create an LLM with prefix caching enabled.
+prefix_cached_llm = LLM(model="facebook/opt-125m",
+                        enable_prefix_caching=True,
+                        num_global_cache_blocks=5000,
+                        gpu_memory_utilization=0.4)
+
+# Warmup so that the shared prompt's KV cache is computed.
+prefix_cached_llm.generate(generating_prompts[0], sampling_params)
+
+# Generate with prefix caching.
+outputs = prefix_cached_llm.generate(generating_prompts, sampling_params)
+
+print("Results with `enable_prefix_caching`")
+
+cached_generated_texts = []
+# Print the outputs. You should see the same outputs as before.
+for output in outputs:
+    prompt = output.prompt
+    generated_text = output.outputs[0].text
+    cached_generated_texts.append(generated_text)
+    print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
+
+print("-" * 80)
+
+# Compare the results and display the speedup
+generated_same = all([
+    regular_generated_texts[i] == cached_generated_texts[i]
+    for i in range(len(prompts))
+])
+print(f"Generated answers are the same: {generated_same}")
+
+"""
+We can simulate the global prefix cache this way:
+1. Run the vllm instance with APC for some time, so some prompts may not 
+hit APC as they are old and evicted
+2. Delete the first vllm instance and start a new one. In this case, 
+global kv cache can be hit directly
+Here we demo the second option.
+"""
+# Destroy the LLM object and free up the GPU memory.
+del prefix_cached_llm
+cleanup_dist_env_and_memory()
+
+# Create an LLM with global prefix caching enabled.
+global_prefix_cached_llm = LLM(model="facebook/opt-125m",
+                        enable_prefix_caching=True,
+                        num_global_cache_blocks=5000,
+                        gpu_memory_utilization=0.4)
+
+# Generate with global prefix caching.
+outputs = global_prefix_cached_llm.generate(generating_prompts, sampling_params)
+
+print("Results with `enable_global_prefix_caching`")
+
+global_cached_generated_texts = []
+# Print the outputs. You should see the same outputs as before.
+for output in outputs:
+    prompt = output.prompt
+    generated_text = output.outputs[0].text
+    global_cached_generated_texts.append(generated_text)
+    print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
+
+print("-" * 80)
+
+# Compare the results and display the speedup
+generated_same = all([
+    regular_generated_texts[i] == global_cached_generated_texts[i]
+    for i in range(len(prompts))
+])
+print(f"Generated answers are the same: {generated_same}")

vllm/attention/backends/flash_attn.py

@@ -128,6 +128,7 @@ class FlashAttentionMetadata(AttentionMetadata):
     # 2nd dimensions are padded up to max_blocks_per_seq if it is cuda-graph
     # captured.
     block_tables: Optional[torch.Tensor]
+    block_hash_map: Optional[List[Dict[int, int]]]
 
     # Whether or not if cuda graph is enabled.
     # Cuda-graph is currently enabled for decoding only.
@@ -234,6 +235,7 @@ def prefill_metadata(self) -> Optional["FlashAttentionMetadata"]:
             seq_start_loc=seq_start_loc,
             context_lens_tensor=context_lens_tensor,
             block_tables=block_tables,
+            block_hash_map=self.block_hash_map,
             use_cuda_graph=False,
             # Begin encoder & cross attn fields below...
             encoder_seq_lens=self.encoder_seq_lens,
@@ -284,6 +286,7 @@ def decode_metadata(self) -> Optional["FlashAttentionMetadata"]:
             if self.seq_start_loc is not None else None,
             context_lens_tensor=None,
             block_tables=block_tables,
+            block_hash_map=self.block_hash_map,
             use_cuda_graph=self.use_cuda_graph,
             # Begin encoder & cross attn fields below...
             encoder_seq_lens=self.encoder_seq_lens,
@@ -376,6 +379,7 @@ def __init__(self, input_builder: "ModelInputForGPUBuilder"):
         self.prefill_seq_lens: List[int] = []
         self.context_lens: List[int] = []
         self.block_tables: List[List[int]] = []
+        self.block_hash_map: List[Dict[int, int]] = []
         self.curr_seq_lens: List[int] = []
         self.multimodal_placeholder_maps: Dict[
             str,
@@ -440,6 +444,8 @@ def _add_seq_group(
                     block_table = block_tables[seq_id][
                         -curr_sliding_window_block:]
             self.block_tables.append(block_table)
+            if seq_id in inter_data.block_hash_map:
+                self.block_hash_map.append(inter_data.block_hash_map[seq_id])            
 
             # Compute slot mapping.
             is_profile_run = is_block_tables_empty(block_tables)
@@ -559,6 +565,7 @@ def build(self, seq_lens: List[int], query_lens: List[int],
             seq_start_loc=seq_start_loc_tensor,
             context_lens_tensor=context_lens_tensor,
             block_tables=block_tables,
+            block_hash_map=self.block_hash_map,
             use_cuda_graph=use_captured_graph,
         )
 

vllm/attention/backends/utils.py

@@ -126,6 +126,7 @@ def __init__(self, input_builder: "ModelInputForGPUBuilder"):
         self.prefill_seq_lens: List[int] = []
         self.context_lens: List[int] = []
         self.block_tables: List[List[int]] = []
+        self.block_hash_map: List[Dict[int, int]] = []
         self.curr_seq_lens: List[int] = []
         self.multimodal_placeholder_maps: Dict[
             str,
@@ -185,6 +186,8 @@ def _add_seq_group(
                     block_table = block_tables[seq_id][
                         -curr_sliding_window_block:]
             self.block_tables.append(block_table)
+            if seq_id in inter_data.block_hash_map:
+                self.block_hash_map.append(inter_data.block_hash_map[seq_id])
 
             # Compute slot mapping.
             is_profile_run = is_block_tables_empty(block_tables)
@@ -275,6 +278,7 @@ def build(self, seq_lens: List[int], query_lens: List[int],
             seq_start_loc=seq_start_loc_tensor,
             context_lens_tensor=context_lens_tensor,
             block_tables=block_tables,
+            block_hash_map=self.block_hash_map,
             use_cuda_graph=use_captured_graph,
         )
 
@@ -326,6 +330,7 @@ def graph_capture_get_metadata_for_batch(
             seq_start_loc=None,
             context_lens_tensor=None,
             block_tables=self._graph_block_tables[:batch_size],
+            block_hash_map={},
             use_cuda_graph=True,
         )
         if is_encoder_decoder_model:

vllm/attention/backends/xformers.py

@@ -223,6 +223,7 @@ def prefill_metadata(self) -> Optional["XFormersMetadata"]:
             query_start_loc=query_start_loc,
             context_lens_tensor=context_lens_tensor,
             block_tables=block_tables,
+            block_hash_map=self.block_hash_map,
             use_cuda_graph=False,
             # Begin encoder & cross attn fields below...
             encoder_seq_lens=self.encoder_seq_lens,
@@ -263,6 +264,7 @@ def decode_metadata(self) -> Optional["XFormersMetadata"]:
             max_prefill_seq_len=0,
             max_decode_seq_len=self.max_decode_seq_len,
             block_tables=block_tables,
+            block_hash_map=self.block_hash_map,
             use_cuda_graph=self.use_cuda_graph,
             # Begin encoder & cross attn fields below...
             encoder_seq_lens=self.encoder_seq_lens,

vllm/attention/ops/paged_attn.py

@@ -1,5 +1,5 @@
 from dataclasses import dataclass
-from typing import List, Optional, Tuple
+from typing import List, Optional, Tuple, Dict
 
 import torch
 
@@ -28,6 +28,7 @@ class PagedAttentionMetadata:
     # 2nd dimensions are padded up to max_blocks_per_seq if it is cuda-graph
     # captured.
     block_tables: Optional[torch.Tensor]
+    block_hash_map: Optional[List[Dict[int, int]]]
 
 
 class PagedAttention:

vllm/config.py

@@ -886,6 +886,7 @@ def __init__(
         num_gpu_blocks_override: Optional[int] = None,
         sliding_window: Optional[int] = None,
         enable_prefix_caching: bool = False,
+        num_global_cache_blocks: int = 0,
         cpu_offload_gb: float = 0,
     ) -> None:
         self.block_size = block_size
@@ -896,6 +897,7 @@ def __init__(
         self.is_attention_free = is_attention_free
         self.sliding_window = sliding_window
         self.enable_prefix_caching = enable_prefix_caching
+        self.num_global_cache_blocks = num_global_cache_blocks
         self.cpu_offload_gb = cpu_offload_gb
 
         self._verify_args()

vllm/core/block/block_table.py

@@ -1,5 +1,5 @@
 import math
-from typing import List, Optional
+from typing import List, Optional, Dict
 
 from vllm.core.block.common import BlockList
 from vllm.core.block.interfaces import Block, DeviceAwareBlockAllocator
@@ -257,6 +257,14 @@ def physical_block_ids(self) -> List[int]:
         """
         return self._blocks.ids()
 
+    @property
+    def block_hashes(self) -> Dict[int, int]:
+        return self._blocks.hashes()
+
+    @property
+    def global_computed_list(self) -> List[int]:
+        return self._blocks.global_computed_list()
+
     def get_unseen_token_ids(self, sequence_token_ids: List[int]) -> List[int]:
         """Get the number of "unseen" tokens in the sequence.
 

vllm/core/block/common.py

@@ -235,25 +235,58 @@
     def __init__(self, blocks: List[Block]):
         self._blocks: List[Block] = []
         self._block_ids: List[int] = []
+        self._block_hashes: Dict[int, int] = {}
+        self._block_glb_computed: List[int] = []
 
         self.update(blocks)
 
     def _add_block_id(self, block_id: Optional[BlockId]) -> None:
         assert block_id is not None
         self._block_ids.append(block_id)
 
+    def _add_block_hash(self,
+                        block_id: Optional[BlockId],
+                        block_hash: Optional[int]) -> None:
+        assert block_id is not None
+        self._block_hashes[block_id] = block_hash
+
+    def _add_block_glb_computed(self,
+                                block_id: Optional[BlockId]) -> None:
+        assert block_id is not None
+        self._block_glb_computed.append(block_id)
+
     def _update_block_id(self, block_index: int,
                          new_block_id: Optional[BlockId]) -> None:
         assert new_block_id is not None
         self._block_ids[block_index] = new_block_id
 
+    def _update_block_hash(self,
+                            prev_block_id: Optional[BlockId],
+                            new_block_id: Optional[BlockId],
+                            new_block_hash: Optional[int]) -> None:
+        assert new_block_id is not None
+        del self._block_hashes[prev_block_id]
+        self._block_hashes[new_block_id] = new_block_hash
+
+    def _update_block_glb_computed(self,
+                                    prev_block_id: Optional[BlockId],
+                                    new_block_id: Optional[BlockId]) -> None:
+        assert new_block_id is not None
+        for idx, block_id in enumerate(self._block_glb_computed):
+            if block_id == prev_block_id:
+                self._block_glb_computed[idx] = new_block_id
+                break
+
     def update(self, blocks: List[Block]):
         self._blocks = blocks
 
         # Cache block ids for fast query
         self._block_ids = []
         for block in self._blocks:
             self._add_block_id(block.block_id)
+            self._add_block_hash(block.block_id, block.content_hash)
+            if block.global_computed:
+                self._add_block_glb_computed(block.block_id)
 
     def append_token_ids(self, block_index: int, token_ids: List[int]) -> None:
         block = self._blocks[block_index]
@@ -264,10 +297,17 @@
         # CoW or promotion may update the internal block_id
         if prev_block_id != block.block_id:
             self._update_block_id(block_index, block.block_id)
+            self._update_block_hash(
+                prev_block_id, block.block_id, block.content_hash)
+            if block.global_computed:
+                self._update_block_glb_computed(prev_block_id, block.block_id)
 
     def append(self, new_block: Block):
         self._blocks.append(new_block)
         self._add_block_id(new_block.block_id)
+        self._add_block_hash(new_block.block_id, new_block.content_hash)
+        if new_block.global_computed:
+            self._add_block_glb_computed(new_block.block_id)
 
     def __len__(self) -> int:
         return len(self._blocks)
@@ -282,13 +322,21 @@
     def reset(self):
         self._blocks = []
         self._block_ids = []
+        self._block_hashes = []
+        self._block_glb_computed = []
 
     def list(self) -> List[Block]:
         return self._blocks
 
     def ids(self) -> List[int]:
         return self._block_ids
 
+    def hashes(self) -> Dict[int, int]:
+        return self._block_hashes
+
+    def global_computed_list(self) -> List[int]:
+        return self._block_glb_computed
+
 
 @dataclass
 class CacheMetricData:

vllm/core/block/cpu_gpu_block_allocator.py

@@ -416,6 +416,14 @@ def computed(self):
     def computed(self, value):
         self._proxy.computed = value
 
+    @property
+    def global_computed(self):
+        return self._proxy.global_computed
+
+    @global_computed.setter
+    def global_computed(self, value):
+        self._proxy.global_computed = value        
+
     @property
     def last_accessed(self) -> float:
         return self._proxy.last_accessed