Index

A

access_count (kvboost.models.CachedChunk attribute) alignment_warning (kvboost.models.WarmResult attribute) apply() (kvboost.cacheblend.CacheBlendRecompute method) (kvboost.selective_recompute.SelectiveRecompute method)

approximate (kvboost.cache_manager.ChunkMatch attribute) AssembledPrompt (class in kvboost.models)

B

BASELINE (kvboost.engine.GenerationMode attribute) bits (kvboost.kv_quantize.QuantizedKV attribute)

broadcast_kv() (in module kvboost.batch) build_prefix_kv() (kvboost.cache_manager.KVCacheManager method)

C

cache_hit_ratio (kvboost.models.AssembledPrompt attribute) cache_stats() (kvboost.engine.InferenceEngine method) CACHEBLEND (kvboost.engine.RecomputeStrategy attribute) CacheBlendRecompute (class in kvboost.cacheblend) cached_length (kvboost.models.AssembledPrompt attribute) cached_past_kv (kvboost.models.AssembledPrompt attribute) cached_tokens (kvboost.engine.GenerationResult attribute) CachedChunk (class in kvboost.models) chained_hash() (in module kvboost.models) check_model_compatibility() (in module kvboost.compat) chunk (kvboost.cache_manager.ChunkMatch attribute) chunk_boundaries (kvboost.models.AssembledPrompt attribute)

chunk_boundary_aligned (kvboost.models.WarmResult attribute) chunk_id (kvboost.models.CachedChunk attribute) chunk_ids_for() (kvboost.chunk_registry.ChunkRegistry method) CHUNK_KV_REUSE (kvboost.engine.GenerationMode attribute) chunk_size (kvboost.models.WarmResult attribute) ChunkMatch (class in kvboost.cache_manager) ChunkRegistry (class in kvboost.chunk_registry) chunks_stored (kvboost.models.WarmResult attribute) ChunkStrategy (class in kvboost.chunk_registry) contains() (kvboost.disk_tier.DiskTier method) content_hash (kvboost.models.CachedChunk attribute) content_hash_from_tokens() (in module kvboost.models) created_at (kvboost.models.CachedChunk attribute)

D

dequantize_kv() (in module kvboost.kv_quantize)

DiskTier (class in kvboost.disk_tier) DOCUMENT (kvboost.chunk_registry.ChunkStrategy attribute)

F

find_common_chunk_prefix() (in module kvboost.batch) find_matching_chunks() (kvboost.cache_manager.KVCacheManager method)

FIXED (kvboost.chunk_registry.ChunkStrategy attribute) from_pretrained() (kvboost.engine.InferenceEngine class method) full_token_ids (kvboost.models.AssembledPrompt attribute)

G

generate() (kvboost.engine.InferenceEngine method) generate_batch() (kvboost.engine.InferenceEngine method) generate_many() (kvboost.engine.InferenceEngine method) generated_tokens (kvboost.engine.GenerationResult attribute)

GenerationMode (class in kvboost.engine) GenerationResult (class in kvboost.engine) get() (kvboost.cache_manager.KVCacheManager method) get_by_content() (kvboost.cache_manager.KVCacheManager method) group_by_prefix() (in module kvboost.batch)

H

has_approximate (kvboost.models.AssembledPrompt attribute)

I

InferenceEngine (class in kvboost.engine)

invalidate() (kvboost.cache_manager.KVCacheManager method)

K

kv_reuse_ratio (kvboost.engine.GenerationResult attribute) kv_seq_len() (kvboost.cache_manager.KVCacheManager static method) kvboost module kvboost.batch module kvboost.cache_manager module kvboost.cacheblend module kvboost.chunk_registry module

kvboost.compat module kvboost.disk_tier module kvboost.kv_quantize module kvboost.models module kvboost.selective_recompute module KVCacheManager (class in kvboost.cache_manager)

L

layers (kvboost.kv_quantize.QuantizedKV attribute) length (kvboost.models.CachedChunk property)

live_position_ids (kvboost.models.AssembledPrompt attribute) live_token_ids (kvboost.models.AssembledPrompt attribute) lookup() (kvboost.cache_manager.KVCacheManager method)

M

memory_bytes() (kvboost.kv_quantize.QuantizedKV method) (kvboost.models.CachedChunk method) merge_kv_list() (kvboost.cache_manager.KVCacheManager static method) mode (kvboost.engine.GenerationResult attribute) module kvboost kvboost.batch kvboost.cache_manager kvboost.cacheblend kvboost.chunk_registry kvboost.compat kvboost.disk_tier kvboost.kv_quantize kvboost.models kvboost.selective_recompute

N

NONE (kvboost.engine.RecomputeStrategy attribute)

O

original_dtype (kvboost.kv_quantize.QuantizedKV attribute)

output_text (kvboost.engine.GenerationResult attribute)

P

pad_and_mask() (in module kvboost.batch) partial_tail_tokens (kvboost.models.WarmResult attribute) past_key_values (kvboost.models.CachedChunk attribute) position_end (kvboost.models.CachedChunk attribute)

position_start (kvboost.models.CachedChunk attribute) PREFIX_CACHE (kvboost.engine.GenerationMode attribute) prefix_hash (kvboost.models.CachedChunk attribute) prompt (kvboost.engine.GenerationResult attribute) prompt_tokens (kvboost.engine.GenerationResult attribute)

Q

quantize_kv() (in module kvboost.kv_quantize)

QuantizedKV (class in kvboost.kv_quantize)

R

read() (kvboost.disk_tier.DiskTier method) recomputed (kvboost.models.CachedChunk attribute)

RecomputeStrategy (class in kvboost.engine) remove() (kvboost.disk_tier.DiskTier method)

S

SELECTIVE (kvboost.engine.RecomputeStrategy attribute) SelectiveRecompute (class in kvboost.selective_recompute) SEMANTIC (kvboost.chunk_registry.ChunkStrategy attribute) slice_kv() (kvboost.cache_manager.KVCacheManager static method)

split() (kvboost.chunk_registry.ChunkRegistry method) stats() (kvboost.cache_manager.KVCacheManager method) (kvboost.disk_tier.DiskTier method) store() (kvboost.cache_manager.KVCacheManager method) SUPPORTED_ARCHITECTURES (in module kvboost.compat)

T

text (kvboost.models.CachedChunk attribute) token_count (kvboost.models.WarmResult attribute) token_ids (kvboost.models.CachedChunk attribute) tokens_per_sec (kvboost.engine.GenerationResult attribute)

total_length (kvboost.models.AssembledPrompt property) total_ms (kvboost.engine.GenerationResult attribute) touch() (kvboost.models.CachedChunk method) ttft_ms (kvboost.engine.GenerationResult attribute)

U

UNSUPPORTED_ARCHITECTURES (in module kvboost.compat)

V

verify_correctness() (kvboost.engine.InferenceEngine method)

W

warm() (kvboost.engine.InferenceEngine method)

WarmResult (class in kvboost.models) write() (kvboost.disk_tier.DiskTier method)