add ckl methods + modify files

joeljang · Apr 24, 2022 · 3f59221 · 3f59221
1 parent b63241f
commit 3f59221
Show file tree

Hide file tree

Showing 52 changed files with 780 additions and 1,006 deletions.
diff --git a/Datasets.py b/Datasets.py
@@ -59,33 +59,33 @@ def __init__(self, tokenizer, type_path, input_length, output_length, args, leng
             elif self.args.dataset=='IL':
                 self.dataset = pd.read_csv('data/TWiki_Probes/IL.csv')
             elif self.args.dataset=='data/wikipedia_09' or self.args.dataset=='wikipedia_0809' or self.args.dataset=='data/wikipedia_09_gpt2' or self.args.dataset=='wikipedia_0809_gpt2':
-                df1 = pd.read_csv('data/TWiki_Probes/aligned/0801-0901_unchanged.csv')
-                df2 = pd.read_csv('data/TWiki_Probes/aligned/0801-0901_updated.csv')
-                df3 = pd.read_csv('data/TWiki_Probes/aligned/0801-0901_new.csv')
+                df1 = pd.read_csv('data/twiki_probes/0801-0901_unchanged.csv')
+                df2 = pd.read_csv('data/twiki_probes/0801-0901_updated.csv')
+                df3 = pd.read_csv('data/twiki_probes/0801-0901_new.csv')
                 df4 = pd.read_csv('data/TWiki_Probes/IL.csv')
                 df1 = pd.concat([df1, df2])
                 df1 = pd.concat([df1, df3])
                 self.dataset = pd.concat([df1, df4])
             elif self.args.dataset=='data/wikipedia_10_gpt2' or self.args.dataset=='data/wikipedia_10' or self.args.dataset=='wikipedia_0910' or self.args.dataset=='wikipedia_0910_gpt2':
-                df1 = pd.read_csv('data/TWiki_Probes/aligned/0901-1001_unchanged.csv')
-                df2 = pd.read_csv('data/TWiki_Probes/aligned/0901-1001_updated.csv')
-                df3 = pd.read_csv('data/TWiki_Probes/aligned/0901-1001_new.csv')
+                df1 = pd.read_csv('data/twiki_probes/0901-1001_unchanged.csv')
+                df2 = pd.read_csv('data/twiki_probes/0901-1001_updated.csv')
+                df3 = pd.read_csv('data/twiki_probes/0901-1001_new.csv')
                 df4 = pd.read_csv('data/TWiki_Probes/IL.csv')
                 df1 = pd.concat([df1, df2])
                 df1 = pd.concat([df1, df3])
                 self.dataset = pd.concat([df1, df4])
             elif self.args.dataset=='data/wikipedia_11_gpt2' or self.args.dataset=='data/wikipedia_11' or self.args.dataset=='wikipedia_1011' or self.args.dataset=='wikipedia_1011_gpt2':
-                df1 = pd.read_csv('data/TWiki_Probes/aligned/1001-1101_unchanged.csv')
-                df2 = pd.read_csv('data/TWiki_Probes/aligned/1001-1101_updated.csv')
-                df3 = pd.read_csv('data/TWiki_Probes/aligned/1001-1101_new.csv')
+                df1 = pd.read_csv('data/twiki_probes/1001-1101_unchanged.csv')
+                df2 = pd.read_csv('data/twiki_probes/1001-1101_updated.csv')
+                df3 = pd.read_csv('data/twiki_probes/1001-1101_new.csv')
                 df4 = pd.read_csv('data/TWiki_Probes/IL.csv')
                 df1 = pd.concat([df1, df2])
                 df1 = pd.concat([df1, df3])
                 self.dataset = pd.concat([df1, df4])
             elif self.args.dataset=='data/wikipedia_12_gpt2' or self.args.dataset=='data/wikipedia_12' or self.args.dataset=='wikipedia_1011' or self.args.dataset=='wikipedia_1011_gpt2':
-                df1 = pd.read_csv('data/TWiki_Probes/aligned/1101-1201_unchanged.csv')
-                df2 = pd.read_csv('data/TWiki_Probes/aligned/1101-1201_updated.csv')
-                df3 = pd.read_csv('data/TWiki_Probes/aligned/1101-1201_new.csv')
+                df1 = pd.read_csv('data/twiki_probes/1101-1201_unchanged.csv')
+                df2 = pd.read_csv('data/twiki_probes/1101-1201_updated.csv')
+                df3 = pd.read_csv('data/twiki_probes/1101-1201_new.csv')
                 df4 = pd.read_csv('data/TWiki_Probes/IL.csv')
                 df1 = pd.concat([df1, df2])
                 df1 = pd.concat([df1, df3])

diff --git a/Datasets_.py b/Datasets_.py
diff --git a/configs/evaluation/GPT2/lora/0801-0901_changed.json b/configs/evaluation/GPT2/lora/0801-0901_changed.json
@@ -0,0 +1,16 @@
+{
+    "input_length" : 25,
+    "output_length" : 25,
+    "dataset" : "0801-0901_changed",
+    "dataset_version" : "full",
+    "train_batch_size" : 32,
+    "model" : "gpt2-large",
+    "method": "lora",
+    "ngpu" : 1,
+    "num_workers" : 40,
+    "CUDA_VISIBLE_DEVICES" : "2",
+    "check_validation" : true,
+    "mode" : "evaluate_ppl",
+    "checkpoint_path" : "outputs/09/GPT2_lora",
+    "output_log": "log/GPT2/lora/0801-0901_changed_ppl.csv"
+}
diff --git a/configs/evaluation/GPT2/lora/0801-0901_ppl_full.json b/configs/evaluation/GPT2/lora/0801-0901_ppl_full.json
@@ -0,0 +1,16 @@
+{
+    "input_length" : 512,
+    "output_length" : 512,
+    "dataset" : "wikipedia_09_gpt2",
+    "dataset_version" : "full",
+    "train_batch_size" : 4,
+    "model" : "gpt2-large",
+    "method": "lora",
+    "ngpu" : 1,
+    "num_workers" : 40,
+    "CUDA_VISIBLE_DEVICES" : "2",
+    "check_validation" : true,
+    "mode" : "evaluate_ppl_corpus",
+    "checkpoint_path" : "outputs/09/GPT2_lora",
+    "output_log": "log/GPT2/lora/0801-0901_ppl_full.csv"
+}
diff --git a/configs/evaluation/GPT2/lora/0801-0901_ppl_subset.json b/configs/evaluation/GPT2/lora/0801-0901_ppl_subset.json
@@ -0,0 +1,16 @@
+{
+    "input_length" : 512,
+    "output_length" : 512,
+    "dataset" : "wikipedia_0809_gpt2",
+    "dataset_version" : "full",
+    "train_batch_size" : 4,
+    "model" : "gpt2-large",
+    "method": "lora",
+    "ngpu" : 1,
+    "num_workers" : 40,
+    "CUDA_VISIBLE_DEVICES" : "2",
+    "check_validation" : true,
+    "mode" : "evaluate_ppl_corpus",
+    "checkpoint_path" : "outputs/09/GPT2_lora",
+    "output_log": "log/GPT2/lora/0801-0901_ppl_subset.csv"
+}
diff --git a/configs/evaluation/GPT2/lora/0801-0901_unchanged.json b/configs/evaluation/GPT2/lora/0801-0901_unchanged.json
@@ -0,0 +1,16 @@
+{
+    "input_length" : 25,
+    "output_length" : 25,
+    "dataset" : "0801-0901_unchanged",
+    "dataset_version" : "full",
+    "train_batch_size" : 32,
+    "model" : "gpt2-large",
+    "method": "lora",
+    "ngpu" : 1,
+    "num_workers" : 40,
+    "CUDA_VISIBLE_DEVICES" : "2",
+    "check_validation" : true,
+    "mode" : "evaluate_ppl",
+    "checkpoint_path" : "outputs/09/GPT2_lora",
+    "output_log": "log/GPT2/lora/0801-0901_unchanged_ppl.csv"
+}
diff --git a/configs/evaluation/GPT2/lora/0901-1001_changed.json b/configs/evaluation/GPT2/lora/0901-1001_changed.json
@@ -0,0 +1,16 @@
+{
+    "input_length" : 25,
+    "output_length" : 25,
+    "dataset" : "0901-1001_changed",
+    "dataset_version" : "full",
+    "train_batch_size" : 32,
+    "model" : "gpt2-large",
+    "method": "lora",
+    "ngpu" : 1,
+    "num_workers" : 40,
+    "CUDA_VISIBLE_DEVICES" : "2",
+    "check_validation" : true,
+    "mode" : "evaluate_ppl",
+    "checkpoint_path" : "outputs/10/GPT2_lora",
+    "output_log": "log/GPT2/lora/0901-1001_changed_ppl.csv"
+}
diff --git a/configs/evaluation/GPT2/lora/0901-1001_ppl_full.json b/configs/evaluation/GPT2/lora/0901-1001_ppl_full.json
@@ -0,0 +1,16 @@
+{
+    "input_length" : 512,
+    "output_length" : 512,
+    "dataset" : "wikipedia_10_gpt2",
+    "dataset_version" : "full",
+    "train_batch_size" : 4,
+    "model" : "gpt2-large",
+    "method": "lora",
+    "ngpu" : 1,
+    "num_workers" : 40,
+    "CUDA_VISIBLE_DEVICES" : "2",
+    "check_validation" : true,
+    "mode" : "evaluate_ppl_corpus",
+    "checkpoint_path" : "outputs/10/GPT2_lora",
+    "output_log": "log/GPT2/lora/0901-1001_ppl_full.csv"
+}
diff --git a/configs/evaluation/GPT2/lora/0901-1001_ppl_subset.json b/configs/evaluation/GPT2/lora/0901-1001_ppl_subset.json
@@ -0,0 +1,16 @@
+{
+    "input_length" : 512,
+    "output_length" : 512,
+    "dataset" : "wikipedia_0910_gpt2",
+    "dataset_version" : "full",
+    "train_batch_size" : 4,
+    "model" : "gpt2-large",
+    "method": "lora",
+    "ngpu" : 1,
+    "num_workers" : 40,
+    "CUDA_VISIBLE_DEVICES" : "2",
+    "check_validation" : true,
+    "mode" : "evaluate_ppl_corpus",
+    "checkpoint_path" : "outputs/10/GPT2_lora",
+    "output_log": "log/GPT2/lora/0901-1001_ppl_subset.csv"
+}
diff --git a/configs/evaluation/GPT2/lora/0901-1001_unchanged.json b/configs/evaluation/GPT2/lora/0901-1001_unchanged.json
@@ -0,0 +1,16 @@
+{
+    "input_length" : 25,
+    "output_length" : 25,
+    "dataset" : "0901-1001_unchanged",
+    "dataset_version" : "full",
+    "train_batch_size" : 32,
+    "model" : "gpt2-large",
+    "method": "lora",
+    "ngpu" : 1,
+    "num_workers" : 40,
+    "CUDA_VISIBLE_DEVICES" : "2",
+    "check_validation" : true,
+    "mode" : "evaluate_ppl",
+    "checkpoint_path" : "outputs/10/GPT2_lora",
+    "output_log": "log/GPT2/lora/0901-1001_unchanged_ppl.csv"
+}
diff --git a/configs/evaluation/GPT2/lora/1001-1101_changed.json b/configs/evaluation/GPT2/lora/1001-1101_changed.json
@@ -0,0 +1,16 @@
+{
+    "input_length" : 25,
+    "output_length" : 25,
+    "dataset" : "1001-1101_changed",
+    "dataset_version" : "full",
+    "train_batch_size" : 32,
+    "model" : "gpt2-large",
+    "method": "lora",
+    "ngpu" : 1,
+    "num_workers" : 40,
+    "CUDA_VISIBLE_DEVICES" : "2",
+    "check_validation" : true,
+    "mode" : "evaluate_ppl",
+    "checkpoint_path" : "outputs/11/GPT2_lora",
+    "output_log": "log/GPT2/lora/1001-1101_changed_ppl.csv"
+}
diff --git a/configs/evaluation/GPT2/lora/1001-1101_ppl_full.json b/configs/evaluation/GPT2/lora/1001-1101_ppl_full.json
@@ -0,0 +1,16 @@
+{
+    "input_length" : 512,
+    "output_length" : 512,
+    "dataset" : "wikipedia_11_gpt2",
+    "dataset_version" : "full",
+    "train_batch_size" : 4,
+    "model" : "gpt2-large",
+    "method": "lora",
+    "ngpu" : 1,
+    "num_workers" : 40,
+    "CUDA_VISIBLE_DEVICES" : "2",
+    "check_validation" : true,
+    "mode" : "evaluate_ppl_corpus",
+    "checkpoint_path" : "outputs/11/GPT2_lora",
+    "output_log": "log/GPT2/lora/1001-1101_ppl_full.csv"
+}
diff --git a/configs/evaluation/GPT2/lora/1001-1101_ppl_subset.json b/configs/evaluation/GPT2/lora/1001-1101_ppl_subset.json
@@ -0,0 +1,16 @@
+{
+    "input_length" : 512,
+    "output_length" : 512,
+    "dataset" : "wikipedia_1011_gpt2",
+    "dataset_version" : "full",
+    "train_batch_size" : 4,
+    "model" : "gpt2-large",
+    "method": "lora",
+    "ngpu" : 1,
+    "num_workers" : 40,
+    "CUDA_VISIBLE_DEVICES" : "2",
+    "check_validation" : true,
+    "mode" : "evaluate_ppl_corpus",
+    "checkpoint_path" : "outputs/11/GPT2_lora",
+    "output_log": "log/GPT2/lora/1001-1101_ppl_subset.csv"
+}
diff --git a/configs/evaluation/GPT2/lora/1001-1101_unchanged.json b/configs/evaluation/GPT2/lora/1001-1101_unchanged.json
@@ -0,0 +1,16 @@
+{
+    "input_length" : 25,
+    "output_length" : 25,
+    "dataset" : "1001-1101_unchanged",
+    "dataset_version" : "full",
+    "train_batch_size" : 32,
+    "model" : "gpt2-large",
+    "method": "lora",
+    "ngpu" : 1,
+    "num_workers" : 40,
+    "CUDA_VISIBLE_DEVICES" : "2",
+    "check_validation" : true,
+    "mode" : "evaluate_ppl",
+    "checkpoint_path" : "outputs/11/GPT2_lora",
+    "output_log": "log/GPT2/lora/1001-1101_unchanged_ppl.csv"
+}
diff --git a/configs/evaluation/GPT2/lora/1101-1201_changed.json b/configs/evaluation/GPT2/lora/1101-1201_changed.json
@@ -0,0 +1,16 @@
+{
+    "input_length" : 25,
+    "output_length" : 25,
+    "dataset" : "1101-1201_changed",
+    "dataset_version" : "full",
+    "train_batch_size" : 32,
+    "model" : "gpt2-large",
+    "method": "lora",
+    "ngpu" : 1,
+    "num_workers" : 40,
+    "CUDA_VISIBLE_DEVICES" : "2",
+    "check_validation" : true,
+    "mode" : "evaluate_ppl",
+    "checkpoint_path" : "outputs/12/GPT2_lora",
+    "output_log": "log/GPT2/lora/1101-1201_changed_ppl.csv"
+}