Skip to content
Open
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 6 additions & 0 deletions evaluation/evaluate_registry.py
Original file line number Diff line number Diff line change
Expand Up @@ -34,6 +34,7 @@
KVzapPress,
KVzipPress,
LagKVPress,
MergingPress,
ObservedAttentionPress,
PyramidKVPress,
QFilterPress,
Expand Down Expand Up @@ -121,4 +122,9 @@
"decoding_adakv_expected_attention_e2": DecodingPress(base_press=AdaKVPress(ExpectedAttentionPress(epsilon=1e-2))),
"decoding_adakv_snapkv": DecodingPress(base_press=AdaKVPress(SnapKVPress())),
"decoding_keydiff": DecodingPress(base_press=KeyDiffPress()),
# MergingPress: merge-on-evict during prefill (values-only merge preserves RoPE keys)
"merging_knorm": MergingPress(KnormPress()),
"merging_snapkv": MergingPress(SnapKVPress()),
"merging_adakv_snapkv": MergingPress(AdaKVPress(SnapKVPress())),
"merging_dms_kvzap_mlp": MergingPress(DMSPress(press=KVzapPress(model_type="mlp"))),
}
2 changes: 2 additions & 0 deletions kvpress/__init__.py
Original file line number Diff line number Diff line change
Expand Up @@ -29,6 +29,7 @@
from kvpress.presses.kvzip_press import KVzipPress
from kvpress.presses.lagkv_press import LagKVPress
from kvpress.presses.leverage_press import LeverageScorePress
from kvpress.presses.merging_press import MergingPress
from kvpress.presses.non_causal_attention_press import NonCausalAttnPress
from kvpress.presses.observed_attention_press import ObservedAttentionPress
from kvpress.presses.per_layer_compression_press import PerLayerCompressionPress
Expand Down Expand Up @@ -87,4 +88,5 @@
"DMSPress",
"FastKVzipPress",
"KVComposePress",
"MergingPress",
]
Loading