Improve efficiency of test/fully_fusible_subgraph_extractor_test.sh

graph_net/config/get_fusible_subgraph_sample_list.txt

@@ -0,0 +1,100 @@
+samples/timm/crossvit_small_240.in1k
+samples/timm/poolformerv2_s12.sail_in1k
+samples/timm/regnety_080.pycls_in1k
+samples/timm/dla46x_c.in1k
+samples/timm/mobilenetv1_100.ra4_e3600_r224_in1k
+samples/timm/efficientnetv2_rw_s.ra2_in1k
+samples/timm/vit_base_patch16_rope_ape_224.naver_in1k
+samples/timm/fastvit_t8.apple_dist_in1k
+samples/timm/test_byobnet.r160_in1k
+samples/timm/mambaout_base.in1k
+samples/timm/davit_small
+samples/timm/resnet61q.ra2_in1k
+samples/timm/coat_tiny
+samples/timm/regnetx_004.pycls_in1k
+samples/timm/convnextv2_large.fcmae
+samples/timm/regnety_640.seer
+samples/timm/repvit_m1_1.dist_300e_in1k
+samples/timm/tinynet_d.in1k
+samples/timm/resnetrs270.tf_in1k
+samples/timm/cait_m48_448
+samples/timm/legacy_seresnet50.in1k
+samples/timm/tinynet_a.in1k
+samples/timm/convnext_small.fb_in1k
+samples/timm/vit_huge_patch14_clip_quickgelu_224.dfn5b
+samples/timm/dpn131.mx_in1k
+samples/timm/convnextv2_large.fcmae_ft_in1k
+samples/timm/convnextv2_small
+samples/timm/repvit_m1.dist_in1k
+samples/timm/cs3darknet_s
+samples/timm/resnet50d.a1_in1k
+samples/timm/dm_nfnet_f6
+samples/timm/coatnet_1_rw_224
+samples/timm/lcnet_050.ra2_in1k
+samples/timm/efficientnet_em.ra2_in1k
+samples/timm/dpn48b
+samples/timm/semnasnet_075.rmsp_in1k
+samples/timm/skresnet34.ra_in1k
+samples/timm/crossvit_15_dagger_240.in1k
+samples/timm/mnasnet_100.rmsp_in1k
+samples/timm/mobilenetv3_rw.rmsp_in1k
+samples/timm/xception65p.ra3_in1k
+samples/timm/coatnet_0_rw_224
+samples/timm/eca_nfnet_l3
+samples/timm/deit3_base_patch16_224.fb_in1k
+samples/timm/mambaout_base_short_rw.sw_e500_in1k
+samples/timm/mobilenetv4_conv_small.e1200_r224_in1k
+samples/timm/xception71.tf_in1k
+samples/timm/dla60.in1k
+samples/timm/repghostnet_130.in1k
+samples/timm/mambaout_base_plus_rw.sw_e150_in12k
+samples/timm/poolformerv2_s36.sail_in1k
+samples/timm/deit3_huge_patch14_224.fb_in1k
+samples/timm/vit_base_patch32_clip_224.datacompxl
+samples/timm/poolformer_m48.sail_in1k
+samples/timm/regnety_006.pycls_in1k
+samples/timm/starnet_s4.in1k
+samples/timm/poolformer_m36.sail_in1k
+samples/timm/vit_huge_patch14_gap_224.in1k_ijepa
+samples/timm/efficientnet_b3.ra2_in1k
+samples/timm/mobilenetv3_large_150d.ra4_e3600_r256_in1k
+samples/timm/hgnetv2_b0.ssld_stage1_in22k_in1k
+samples/timm/convnextv2_huge.fcmae
+samples/timm/davit_huge
+samples/timm/regnetx_004_tv.tv2_in1k
+samples/timm/dla34.in1k
+samples/timm/convnext_xlarge.fb_in22k
+samples/timm/resmlp_12_224.fb_dino
+samples/timm/fasternet_t1.in1k
+samples/timm/resnetblur50.bt_in1k
+samples/timm/res2net50d.in1k
+samples/timm/vit_base_patch32_224.augreg_in1k
+samples/timm/mambaout_base_wide_rw.sw_e500_in1k
+samples/timm/vgg19_bn.tv_in1k
+samples/timm/vit_small_patch16_rope_ape_224.naver_in1k
+samples/timm/hardcorenas_b.miil_green_in1k
+samples/timm/vgg16.tv_in1k
+samples/timm/xception41p.ra3_in1k
+samples/timm/efficientnet_lite0.ra_in1k
+samples/timm/regnetv_064.ra3_in1k
+samples/timm/regnety_320.pycls_in1k
+samples/timm/convnext_pico.d1_in1k
+samples/timm/repvit_m1_0.dist_300e_in1k
+samples/timm/resnet50c.gluon_in1k
+samples/timm/mobileone_s4.apple_in1k
+samples/timm/ghostnet_100.in1k
+samples/timm/deit_base_distilled_patch16_384
+samples/timm/dpn68b.mx_in1k
+samples/timm/dla60_res2next
+samples/timm/resnet101d.gluon_in1k
+samples/timm/eva02_large_patch14_clip_224.merged2b
+samples/timm/fasternet_m.in1k
+samples/timm/mobilenetv2_110d.ra_in1k
+samples/timm/regnetx_064.pycls_in1k
+samples/timm/cspresnet50.ra_in1k
+samples/timm/resmlp_24_224.fb_dino
+samples/timm/mobileone_s3.apple_in1k
+samples/timm/mobileone_s2.apple_in1k
+samples/timm/res2net101d
+samples/timm/hardcorenas_f.miil_green_in1k
+samples/timm/hrnet_w18_ssld.paddle_in1k

graph_net/config/small100_torch_samples_list.txt

@@ -0,0 +1,101 @@
+# shuf -n 100 torch_samples_list.txt > small_torch_samples_list.txt
+samples/ultralytics/yolov3-tinyu
+samples/transformers-auto-model/hf-tiny-model-private_tiny-random-Swinv2ForImageClassification
+samples/torchgeometric/RECT_L
+samples/transformers-auto-model/opus-mt-en-gv
+samples/transformers-auto-model/Visual-Attention-Network_van-tiny
+samples/timm/convnextv2_base.fcmae_ft_in1k
+samples/torchvision/vgg16_bn
+samples/timm/regnety_080_tv.tv2_in1k
+samples/transformers-auto-model/internetoftim_dinov2-base
+samples/timm/mobilenetv4_conv_aa_large.e230_r384_in12k
+samples/ultralytics/yolo12x
+samples/transformers-auto-model/opus-mt-eo-it
+samples/transformers-auto-model/opus-mt-es-csg
+samples/transformers-auto-model/google_byt5_base
+samples/transformers-auto-model/Fsoft-AIC_videberta-base
+samples/timm/mixer_l16_224.goog_in21k
+samples/transformers-auto-model/hf-tiny-model-private_tiny-random-EfficientFormerModel
+samples/transformers-auto-model/distilbert_distilbert-base-multilingual-cased
+samples/transformers-auto-model/ydshieh_tiny-random-ViTModel
+samples/transformers-auto-model/microsoft_beit-base-patch16-384
+samples/transformers-auto-model/hf-tiny-model-private_tiny-random-Data2VecVisionModel
+samples/transformers-auto-model/kanarya-750m
+samples/transformers-auto-model/Helsinki-NLP_opus-mt-zh-en
+samples/transformers-auto-model/opus-mt-de-tl
+samples/timm/vit_base_patch16_224_miil.in21k
+samples/mmseg/SegNeXt-b
+samples/transformers-auto-model/OFA-Sys_chinese-clip-vit-large-patch14
+samples/mmpose/ViTPose-B
+samples/transformers-auto-model/Sayan01_tiny-bert-sst2-distilled
+samples/timm/regnety_080.pycls_in1k
+samples/ultralytics/yolov9c
+samples/transformers-auto-model/opus-mt-bcl-fr
+samples/transformers-auto-model/opus-mt-de-ase
+samples/timm/eva02_base_patch16_clip_224.merged2b
+samples/mmpose/S-ViPNAS-Res50
+samples/transformers-auto-model/Lin-Chen_ShareGPT4V-7B_Pretrained_vit-large336-l12
+samples/transformers-auto-model/opus-mt-en-tw
+samples/timm/hgnet_tiny.paddle_in1k
+samples/transformers-auto-model/opus-mt-ase-es
+samples/transformers-auto-model/MIT_ast-finetuned-audioset-10-10-0.4593
+samples/transformers-auto-model/opus-mt-en-kqn
+samples/transformers-auto-model/opus-mt-en-guw
+samples/timm/vit_small_patch16_rope_mixed_ape_224.naver_in1k
+samples/timm/repvgg_b0.rvgg_in1k
+samples/transformers-auto-model/opus-mt-en-crs
+samples/transformers-auto-model/opus-mt-en-bat
+samples/transformers-auto-model/Francesco_resnet101-224-1k
+samples/timm/dpn107.mx_in1k
+samples/transformers-auto-model/google/t5-efficient-large-kv128
+samples/transformers-auto-model/aimarsg_prueba5
+samples/torchvision/resnext50_32x4d
+samples/timm/inception_v3.gluon_in1k
+samples/transformers-auto-model/mohsenfayyaz_BERT_Warmup
+samples/transformers-auto-model/MoritzLaurer_xtremedistil-l6-h256-mnli-fever-anli-ling-binary
+samples/transformers-auto-model/pipecat-ai_smart-turn-v2
+samples/timm/poolformerv2_m48.sail_in1k
+samples/timm/res2net101_26w_4s
+samples/transformers-auto-model/opus-mt-ar-es
+samples/timm/regnety_002.pycls_in1k
+samples/transformers-auto-model/simjo_dummy-model
+samples/mmseg/CCNet_R101
+samples/transformers-auto-model/opus-mt-tc-bible-big-deu_eng_fra_por_spa-bat
+samples/transformers-auto-model/facebook_sam-vit-large
+samples/transformers-auto-model/distilbert_distilbert-base-uncased
+samples/mmseg/PoolFormer_m36
+samples/timm/hgnetv2_b4.ssld_stage1_in22k_in1k
+samples/timm/regnetx_160.pycls_in1k
+samples/transformers-auto-model/google/bert_uncased_L-6_H-512_A-8
+samples/transformers-auto-model/ogoshi2000_stance-nystromformer
+samples/timm/resnetv2_18.ra4_e3600_r224_in1k
+samples/transformers-auto-model/TinyLlama/TinyLlama-1.1B-Chat-v0.4
+samples/timm/resnetaa50d.d_in12k
+samples/transformers-auto-model/bge-small-en-v1.5
+samples/timm/coatnet_rmlp_0_rw_224
+samples/transformers-auto-model/opus-mt-en-sal
+samples/mmseg/UPerNet_R101
+samples/transformers-auto-model/sentence-transformers/paraphrase-distilroberta-base-v1
+samples/transformers-auto-model/opus-mt-en-phi
+samples/transformers-auto-model/opus-mt-fi-niu
+samples/torchvision/vit_l_16
+samples/timm/res2net50_14w_8s.in1k
+samples/mmseg/Twins_pcpvt-b_uperhead
+samples/transformers-auto-model/google-t5_t5-large
+samples/transformers-auto-model/TrustSafeAI_RADAR_Vicuna_7B
+samples/torchaudio/wavlm_base
+samples/transformers-auto-model/opus-mt-NORTH_EU-NORTH_EU
+samples/transformers-auto-model/apple_aimv2-huge-patch14-224
+samples/timm/resnetrs270.tf_in1k
+samples/transformers-auto-model/opus-mt-de-no
+samples/transformers-auto-model/opus-mt-tc-bible-big-gmw-deu_eng_fra_por_spa
+samples/timm/repghostnet_111.in1k
+samples/mmseg/DeepLabV3+_R101
+samples/transformers-auto-model/microsoft_swin-base-patch4-window12-384-in22k
+samples/transformers-auto-model/opus-mt-es-de
+samples/transformers-auto-model/nvidia_mit-b4
+samples/transformers-auto-model/opus-mt-tc-bible-big-ine-deu_eng_fra_por_spa
+samples/mmpose/pose_swin_b
+samples/transformers-auto-model/opus-mt-bi-en
+samples/transformers-auto-model/all-mpnet-base-v2
+samples/transformers-auto-model/Neurora_opus-tatoeba-heb-eng

graph_net/config/small_sample_list_for_get_fusible_subgraph.txt

@@ -0,0 +1,10 @@
+#samples/timm/crossvit_small_240.in1k
+#samples/timm/poolformerv2_s12.sail_in1k
+#samples/timm/regnety_080.pycls_in1k
+#samples/timm/dla46x_c.in1k
+#samples/timm/mobilenetv1_100.ra4_e3600_r224_in1k
+samples/timm/efficientnetv2_rw_s.ra2_in1k
+samples/timm/vit_base_patch16_rope_ape_224.naver_in1k
+#samples/timm/fastvit_t8.apple_dist_in1k
+#samples/timm/test_byobnet.r160_in1k
+#samples/timm/mambaout_base.in1k

graph_net/subgraph_decompose_and_evaluation_step.py

@@ -186,7 +186,7 @@ def run_decomposer_for_single_model(
         "decorator_path": f"{graphnet_root}/graph_net/{framework}/extractor.py",
         "decorator_config": {
             "name": model_name,
-            "custom_extractor_path": f"{graphnet_root}/graph_net/{framework}/naive_graph_decomposer.py",
+            "custom_extractor_path": f"{graphnet_root}/graph_net/{framework}/graph_decomposer.py",
             "custom_extractor_config": {
                 "output_dir": output_dir,
                 "split_positions": split_positions,

graph_net/tensor_meta.py

@@ -58,6 +58,11 @@ def serialize_to_py_str(self) -> str:
         lines = [
             (f"class {self.record_class_name}:"),
             (f'\tname = "{self.name}"'),
+            *(
+                [f'\toriginal_name = "{self.original_name}"']
+                if self.original_name is not None
+                else []
+            ),
             (f"\tshape = {self.shape}"),
             (f'\tdtype = "{self.dtype}"'),
             (f'\tdevice = "{self.device}"'),

graph_net/test/chain_naive_graph_decomposer_test.sh

@@ -11,7 +11,7 @@ decorator_config_json_str=$(cat <<EOF
     "decorator_path": "$GRAPH_NET_ROOT/torch/extractor.py",
     "decorator_config": {
         "name": "$MODEL_NAME",
-        "custom_extractor_path": "$GRAPH_NET_ROOT/torch/naive_graph_decomposer.py",
+        "custom_extractor_path": "$GRAPH_NET_ROOT/torch/graph_decomposer.py",
         "custom_extractor_config": {
             "output_dir": "/tmp/chain_naive_decompose_workspace",
             "split_positions": [8, 16, 32],

graph_net/test/decomposer_validator_test.sh

@@ -16,7 +16,7 @@ extractor_config_json_str=$(cat <<EOF
     "decorator_path": "$GRAPH_NET_ROOT/torch/extractor.py",
     "decorator_config": {
         "name": "$MODEL_NAME",
-        "custom_extractor_path": "$GRAPH_NET_ROOT/torch/naive_graph_decomposer.py",
+        "custom_extractor_path": "$GRAPH_NET_ROOT/torch/graph_decomposer.py",
         "custom_extractor_config": {
             "output_dir": "$OUTPUT_DIR/${MODEL_NAME}_decomposed",
             "split_positions": [8, 16, 32],

graph_net/test/fully_fusible_subgraph_extractor_test.sh

@@ -0,0 +1,33 @@
+#!/bin/bash
+
+GRAPH_NET_ROOT=$(python3 -c "import graph_net; import os; print(
+os.path.dirname(graph_net.__file__))")
+
+# input model path
+MODEL_NAME=resnet18
+MODEL_PATH_IN_SAMPLES=/timm/$MODEL_NAME
+# INPUT_MODEL_LIST=$GRAPH_NET_ROOT/config/get_fusible_subgraph_sample_list.txt
+INPUT_MODEL_LIST=$GRAPH_NET_ROOT/config/small_sample_list_for_get_fusible_subgraph.txt
+
+OUTPUT_DIR="/tmp/find_fully_fusible_output"
+config_json_str=$(cat <<EOF
+{
+    "handler_path": "$GRAPH_NET_ROOT/torch/fully_fusible_subgraph_extractor.py",
+    "handler_class_name":"FullyFusibleSubgraphExtractor",
+    "handler_config": {
+        "resume": false,
+        "model_path_prefix": "$GRAPH_NET_ROOT/../",
+        "output_dir": "$OUTPUT_DIR",
+        "nn_module_fully_fusible_decorator_path": "$GRAPH_NET_ROOT/torch/count_kernels_util.py",
+        "nn_module_fully_fusible_decorator_class_name": "TorchSubModuleFullyFusibleDecorator",
+        "max_step": 3,
+        "min_step": 2,
+        "max_nodes": 4
+    }
+}
+EOF
+)
+CONFIG=$(echo $config_json_str | base64 -w 0)
+
+# python3 -m graph_net.model_path_handler --model-path $GRAPH_NET_ROOT/../samples/$MODEL_PATH_IN_SAMPLES --handler-config=$CONFIG
+python3 -m graph_net.model_path_handler --model-path-list $INPUT_MODEL_LIST --handler-config=$CONFIG

graph_net/test/graph_variable_rename_test.sh

@@ -0,0 +1,27 @@
+#!/bin/bash
+
+GRAPH_NET_ROOT=$(python3 -c "import graph_net; import os; print(
+os.path.dirname(graph_net.__file__))")
+
+# input model path
+MODEL_NAME=resnet18
+MODEL_PATH_IN_SAMPLES=/timm/$MODEL_NAME
+config_json_str=$(cat <<EOF
+{
+    "handler_path": "$GRAPH_NET_ROOT/torch/graph_variable_renamer.py",
+    "handler_class_name": "GraphVariableRenamer",
+    "handler_config": {
+        "model_path_prefix": "$GRAPH_NET_ROOT/../",
+        "data_input_predicator_filepath": "$GRAPH_NET_ROOT/torch/constraint_util.py",
+        "data_input_predicator_class_name": "NaiveDataInputPredicator",
+        "model_runnable_predicator_filepath": "$GRAPH_NET_ROOT/torch/constraint_util.py",
+        "model_runnable_predicator_class_name": "ModelRunnablePredicator",
+        "output_dir": "/tmp/graph_variable_rename_workspace"
+    }
+}
+EOF
+)
+CONFIG=$(echo $config_json_str | base64 -w 0)
+
+python3 -m graph_net.model_path_handler --model-path samples/$MODEL_PATH_IN_SAMPLES --handler-config=$CONFIG
+# python3 -m graph_net.model_path_handler --model-path-list $GRAPH_NET_ROOT/config/decomposition_error_tmp_torch_samples_list.txt --handler-config=$CONFIG

graph_net/test/naive_check_if_fully_fusionabale.sh

@@ -9,7 +9,7 @@ MODEL_PATH_IN_SAMPLES=/timm/$MODEL_NAME
 checker_config_json_str=$(cat <<EOF
 {
     "post_extract_process_config": {
-        "post_extract_process_path":"$GRAPH_NET_ROOT/torch/post_extract_process_count_kernels.py",
+        "post_extract_process_path":"$GRAPH_NET_ROOT/torch/count_kernels_util.py",
         "post_extract_process_class_name": "GraphFullyFusionable"
     }
 }

graph_net/test/naive_decomposer_and_post_extract_process_test.sh

@@ -12,14 +12,14 @@ decorator_config_json_str=$(cat <<EOF
     "decorator_path": "$GRAPH_NET_ROOT/torch/extractor.py",
     "decorator_config": {
         "name": "$MODEL_NAME",
-        "custom_extractor_path": "$GRAPH_NET_ROOT/torch/naive_graph_decomposer.py",
+        "custom_extractor_path": "$GRAPH_NET_ROOT/torch/graph_decomposer.py",
         "custom_extractor_config": {
             "output_dir": "/tmp/naive_decompose_workspace",
             "split_positions": [8, 16, 32],
             "group_head_and_tail": true,
             "filter_path":"$GRAPH_NET_ROOT/torch/naive_subgraph_filter.py",
             "filter_config": {},
-            "post_extract_process_path":"$GRAPH_NET_ROOT/torch/post_extract_process_count_kernels.py",
+            "post_extract_process_path":"$GRAPH_NET_ROOT/torch/count_kernels_util.py",
             "post_extract_process_class_name": "GraphFullyFusible"
         }
     }

graph_net/test/naive_graph_decomposer_test.sh

@@ -8,7 +8,7 @@ MODEL_NAME=resnet18
 MODEL_PATH_IN_SAMPLES=/timm/$MODEL_NAME
 config_json_str=$(cat <<EOF
 {
-    "handler_path": "$GRAPH_NET_ROOT/torch/naive_graph_decomposer.py",
+    "handler_path": "$GRAPH_NET_ROOT/torch/graph_decomposer.py",
     "handler_class_name": "NaiveDecomposerExtractor",
     "handler_config": {
         "model_path_prefix": "$GRAPH_NET_ROOT/../",
@@ -23,4 +23,4 @@ EOF
 CONFIG=$(echo $config_json_str | base64 -w 0)
 
 # python3 -m graph_net.model_path_handler --model-path $GRAPH_NET_ROOT/../samples/$MODEL_PATH_IN_SAMPLES --handler-config=$CONFIG
-python3 -m graph_net.model_path_handler --model-path-list $GRAPH_NET_ROOT/config/decomposition_error_tmp_torch_samples_list.txt --handler-config=$CONFIG
+python3 -m graph_net.model_path_handler --model-path-list $GRAPH_NET_ROOT/test/dev_model_list/decomposition_error_tmp_torch_samples_list.txt --handler-config=$CONFIG

graph_net/test/shape_prop_batch_init_input_tensor_constraints_test.sh

@@ -26,4 +26,4 @@ EOF
 )
 CONFIG=$(echo $config_json_str | base64 -w 0)
 
-python3 -m graph_net.model_path_handler --model-path-list $GRAPH_NET_ROOT/config/small_torch_samples_list.txt --handler-config=$CONFIG
+python3 -m graph_net.model_path_handler --model-path-list $GRAPH_NET_ROOT/config/small100_torch_samples_list.txt --handler-config=$CONFIG