support tome-llava

Author: zhangbilang

configs/sparsification/methods/ToMe/tome.yml

@@ -0,0 +1,25 @@
+base:
+    seed: &seed 42
+model:
+    type: Llava
+    path: model path
+    torch_dtype: auto
+eval:
+    eval_pos: [pretrain, transformed]
+    type: vqa
+    name: [mme]
+    download: False
+    path: MME dataset path
+    bs: 1
+    inference_per_block: False
+sparse:
+    vision:
+        method: TokenReduction
+        special:
+            method: ToMe
+            r: 16  # 16 or (16, -1.0) or [16,16,16,16,16,16]
+
+save:
+    save_trans: False
+    save_fake: False
+    save_path: /path/to/save/

llmc/compression/token_reduction/__init__.py

@@ -1,2 +1,3 @@
 from .base_blockwise_token_reduction import TokenReduction
 from .fastv import FastV
+from .tome import ToMe