File size: 4,657 Bytes
0f00646 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 | # Copyright (c) 2025 SandAI. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""
Base classes for cache management strategies.
"""
from abc import ABC, abstractmethod
from typing import Dict, Optional, Tuple
import torch
class CacheStrategy(ABC):
"""
Abstract base class for cache management strategies.
All cache implementations should inherit from this class and implement
the required methods.
"""
def __init__(self, enabled: bool = True):
"""
Initialize the cache strategy.
Args:
enabled: Whether this cache strategy is enabled
"""
self.enabled = enabled
@abstractmethod
def reset(self):
"""
Reset the cache state.
This method should clear all internal state and prepare the cache
for a new inference run.
"""
pass
def reset_if_enabled(self):
"""Reset the cache if it is enabled."""
if self.enabled:
self.reset()
class OutputCache(CacheStrategy):
"""
Abstract base class for output reuse strategies.
Output caching strategies determine when model outputs can be reused
based on input similarity metrics.
"""
@abstractmethod
def should_reuse(
self,
chunk_id: int,
step: int,
current_features: torch.Tensor,
**kwargs
) -> bool:
"""
Determine whether the output for a chunk should be reused.
Args:
chunk_id: The ID of the current chunk
step: The current denoising step
current_features: Feature tensor for the current input
**kwargs: Additional arguments specific to the implementation
Returns:
True if the output should be reused, False otherwise
"""
pass
@abstractmethod
def update_residual(
self,
chunk_id: int,
residual: torch.Tensor
):
"""
Update the residual for a chunk.
When outputs are reused, the residual from the previous step is
applied to the current input.
Args:
chunk_id: The ID of the chunk
residual: The residual tensor to store
"""
pass
@abstractmethod
def get_residual(self, chunk_id: int) -> Optional[torch.Tensor]:
"""
Get the stored residual for a chunk.
Args:
chunk_id: The ID of the chunk
Returns:
The residual tensor if available, None otherwise
"""
pass
class KVCompressor(CacheStrategy):
"""
Abstract base class for KV cache compression strategies.
KV cache compression manages memory usage by selectively compressing
KV caches from completed chunks.
"""
@abstractmethod
def should_compress(
self,
current_chunk_id: int,
cache_used: int,
cache_capacity: int,
**kwargs
) -> bool:
"""
Determine whether KV cache compression should be triggered.
Args:
current_chunk_id: The ID of the most recently completed chunk
cache_used: Current KV cache usage in tokens
cache_capacity: Total KV cache capacity in tokens
**kwargs: Additional arguments specific to the implementation
Returns:
True if compression should be performed, False otherwise
"""
pass
@abstractmethod
def compress(
self,
inference_params,
chunk_tracker,
clean_chunk_ids: list,
active_chunk_ids: list,
**kwargs
) -> Dict[int, Tuple[int, int]]:
"""
Compress KV caches for specified chunks.
Args:
inference_params: Inference parameters containing KV cache
chunk_tracker: Tracker managing chunk ranges
clean_chunk_ids: List of chunk IDs to compress
active_chunk_ids: List of chunk IDs to keep uncompressed
**kwargs: Additional arguments
Returns:
Dictionary mapping chunk_id to (start, end) ranges after compression
"""
pass
|