From 54d3560daf1dc50a390c722e5175aab62333d1ca Mon Sep 17 00:00:00 2001
From: Anmol Mekala <49127549+molereddy@users.noreply.github.com>
Date: Sat, 1 Mar 2025 09:13:50 -0500
Subject: [PATCH 1/6] Fix hyperlinks in README (#2)

* testing commit

* Fixes

* cleanup
---
 README.md | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/README.md b/README.md
index dce38e5..a04d486 100644
--- a/README.md
+++ b/README.md
@@ -44,7 +44,7 @@ We provide several variants for each of the components in the unlearning pipelin
 
 ## 📌 Table of Contents
 - 📖 [Overview](#-overview)
-- 🗃️ [Available Components](#-available-components)
+- 🗃️ [Available Components](#%EF%B8%8F-available-components)
 - ⚡ [Quickstart](#-quickstart)
   - 🛠️ [Environment Setup](#-environment-setup)
   - 💾 [Data Setup](#-data-setup)
@@ -56,7 +56,7 @@ We provide several variants for each of the components in the unlearning pipelin
 - ➕ [How to Add New Components](#-how-to-add-new-components)
 - 📚 [Further Documentation](#-further-documentation)
 - 🔗 [Support & Contributors](#-support--contributors)
-- 📝 [Citing this work](#-citating-this-work)
+- 📝 [Citing this work](#-citing-this-work)
 - 🤝 [Acknowledgements](#-acknowledgements)
 - 📄 [License](#-license)
 
@@ -198,7 +198,7 @@ If you use OpenUnlearning in your research, please cite:
 
 ---
 
-### 🤝 Acknowledgments
+### 🤝 Acknowledgements
 
 - This repo is inspired from [LLaMA-Factory](https://github.com/hiyouga/LLaMA-Factory). 
 - The [TOFU](https://github.com/locuslab/tofu) and [MUSE](https://github.com/jaechan-repo/muse_bench) benchmarks served as the foundation for our re-implementation. 

From 4c36e4f5a39d979280efd20452be7ba5ff54e40a Mon Sep 17 00:00:00 2001
From: Dornavineeth <vineethdorna@gmail.com>
Date: Sun, 2 Mar 2025 18:37:19 +0000
Subject: [PATCH 2/6] Fixed DPO command

---
 scripts/tofu_unlearn.sh | 56 ++++++++++++++++++++++++++++++++++++-----
 1 file changed, 50 insertions(+), 6 deletions(-)

diff --git a/scripts/tofu_unlearn.sh b/scripts/tofu_unlearn.sh
index a556bd1..eba38ff 100644
--- a/scripts/tofu_unlearn.sh
+++ b/scripts/tofu_unlearn.sh
@@ -9,12 +9,6 @@ models=(
     "Llama-3.2-3B-Instruct"
     "Llama-3.1-8B-Instruct"
 )
-trainers_experiments=(
-    "GradAscent unlearn/tofu/default.yaml"
-    "GradDiff unlearn/tofu/default.yaml"
-    "NPO unlearn/tofu/default.yaml"
-    "DPO unlearn/tofu/default.yaml"
-)
 forget_retain_splits=(
     "forget01 retain99"
     "forget05 retain95"
@@ -29,7 +23,57 @@ gradient_accumulation_steps=4
 ########################################### Unlearn TOFU models ########################################################
 ########################################################################################################################
 
+trainers_experiments=(
+    "GradAscent unlearn/tofu/default.yaml"
+    "GradDiff unlearn/tofu/default.yaml"
+    "NPO unlearn/tofu/default.yaml"
+)
 
+for split in "${forget_retain_splits[@]}"; do
+    forget_split=$(echo $split | cut -d' ' -f1)
+    retain_split=$(echo $split | cut -d' ' -f2)
+    for model in "${models[@]}"; do
+        for trainer_experiment in "${trainers_experiments[@]}"; do
+            trainer=$(echo $trainer_experiment | cut -d' ' -f1)
+            experiment=$(echo $trainer_experiment | cut -d' ' -f2)
+            
+            task_name=tofu_${model}_${forget_split}_${trainer} 
+            model_path=open-unlearning/tofu_${model}_full
+            echo ${task_name}: Unlearning ${model_path} using ${trainer}
+
+            # Unlearn
+            CUDA_VISIBLE_DEVICES=0,1 accelerate launch --config_file configs/accelerate/default_config.yaml --main_process_port $MASTER_PORT \
+            src/train.py --config-name=unlearn.yaml \
+            experiment=${experiment} \
+            trainer=${trainer} \
+            task_name=${task_name} \
+            model=${model} \
+            forget_split=${forget_split} \
+            retain_split=${retain_split} \
+            model.model_args.pretrained_model_name_or_path=${model_path} \
+            retain_logs_path=saves/eval/tofu_${model}_${retain_split}/TOFU_EVAL.json \
+            trainer.args.per_device_train_batch_size=$per_device_train_batch_size \
+            trainer.args.gradient_accumulation_steps=$gradient_accumulation_steps \
+            trainer.args.ddp_find_unused_parameters=true \
+            trainer.args.gradient_checkpointing=true
+
+            # Eval
+            CUDA_VISIBLE_DEVICES=0 python src/eval.py \
+            experiment=eval/tofu/default.yaml \
+            forget_split=${forget_split} \
+            model=${model} \
+            task_name=${task_name} \
+            model.model_args.pretrained_model_name_or_path=saves/unlearn/${task_name} \
+            paths.output_dir=saves/unlearn/${task_name}/evals \
+            retain_logs_path=saves/eval/tofu_${model}_${retain_split}/TOFU_EVAL.json
+        done
+    done
+done
+
+
+trainers_experiments=(
+    "DPO unlearn/tofu/idk.yaml"
+)
 for split in "${forget_retain_splits[@]}"; do
     forget_split=$(echo $split | cut -d' ' -f1)
     retain_split=$(echo $split | cut -d' ' -f2)

From f7a69dee40cfe918e32250ce7c1ba7564205c9b4 Mon Sep 17 00:00:00 2001
From: Dornavineeth <vineethdorna@gmail.com>
Date: Sun, 2 Mar 2025 18:50:47 +0000
Subject: [PATCH 3/6] download idk

---
 setup_data.py | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/setup_data.py b/setup_data.py
index 48de0ad..358779c 100644
--- a/setup_data.py
+++ b/setup_data.py
@@ -1,8 +1,17 @@
 from huggingface_hub import snapshot_download
 
+# Setup retain model metrics
 snapshot_download(
     repo_id="open-unlearning/eval",
     allow_patterns="*.json",
     repo_type="dataset",
     local_dir="saves/eval",
 )
+
+# Setup data
+snapshot_download(
+    repo_id="open-unlearning/idk",
+    allow_patterns="*.jsonl",
+    repo_type="dataset",
+    local_dir="data",
+)
\ No newline at end of file

From 332af36c4772eb1e836b767adbd102951693b60c Mon Sep 17 00:00:00 2001
From: Vineeth <48151992+Dornavineeth@users.noreply.github.com>
Date: Sun, 2 Mar 2025 14:13:43 -0500
Subject: [PATCH 4/6] Revert "Dpo fix"

---
 scripts/tofu_unlearn.sh | 56 +++++------------------------------------
 setup_data.py           |  9 -------
 2 files changed, 6 insertions(+), 59 deletions(-)

diff --git a/scripts/tofu_unlearn.sh b/scripts/tofu_unlearn.sh
index eba38ff..a556bd1 100644
--- a/scripts/tofu_unlearn.sh
+++ b/scripts/tofu_unlearn.sh
@@ -9,6 +9,12 @@ models=(
     "Llama-3.2-3B-Instruct"
     "Llama-3.1-8B-Instruct"
 )
+trainers_experiments=(
+    "GradAscent unlearn/tofu/default.yaml"
+    "GradDiff unlearn/tofu/default.yaml"
+    "NPO unlearn/tofu/default.yaml"
+    "DPO unlearn/tofu/default.yaml"
+)
 forget_retain_splits=(
     "forget01 retain99"
     "forget05 retain95"
@@ -23,57 +29,7 @@ gradient_accumulation_steps=4
 ########################################### Unlearn TOFU models ########################################################
 ########################################################################################################################
 
-trainers_experiments=(
-    "GradAscent unlearn/tofu/default.yaml"
-    "GradDiff unlearn/tofu/default.yaml"
-    "NPO unlearn/tofu/default.yaml"
-)
 
-for split in "${forget_retain_splits[@]}"; do
-    forget_split=$(echo $split | cut -d' ' -f1)
-    retain_split=$(echo $split | cut -d' ' -f2)
-    for model in "${models[@]}"; do
-        for trainer_experiment in "${trainers_experiments[@]}"; do
-            trainer=$(echo $trainer_experiment | cut -d' ' -f1)
-            experiment=$(echo $trainer_experiment | cut -d' ' -f2)
-            
-            task_name=tofu_${model}_${forget_split}_${trainer} 
-            model_path=open-unlearning/tofu_${model}_full
-            echo ${task_name}: Unlearning ${model_path} using ${trainer}
-
-            # Unlearn
-            CUDA_VISIBLE_DEVICES=0,1 accelerate launch --config_file configs/accelerate/default_config.yaml --main_process_port $MASTER_PORT \
-            src/train.py --config-name=unlearn.yaml \
-            experiment=${experiment} \
-            trainer=${trainer} \
-            task_name=${task_name} \
-            model=${model} \
-            forget_split=${forget_split} \
-            retain_split=${retain_split} \
-            model.model_args.pretrained_model_name_or_path=${model_path} \
-            retain_logs_path=saves/eval/tofu_${model}_${retain_split}/TOFU_EVAL.json \
-            trainer.args.per_device_train_batch_size=$per_device_train_batch_size \
-            trainer.args.gradient_accumulation_steps=$gradient_accumulation_steps \
-            trainer.args.ddp_find_unused_parameters=true \
-            trainer.args.gradient_checkpointing=true
-
-            # Eval
-            CUDA_VISIBLE_DEVICES=0 python src/eval.py \
-            experiment=eval/tofu/default.yaml \
-            forget_split=${forget_split} \
-            model=${model} \
-            task_name=${task_name} \
-            model.model_args.pretrained_model_name_or_path=saves/unlearn/${task_name} \
-            paths.output_dir=saves/unlearn/${task_name}/evals \
-            retain_logs_path=saves/eval/tofu_${model}_${retain_split}/TOFU_EVAL.json
-        done
-    done
-done
-
-
-trainers_experiments=(
-    "DPO unlearn/tofu/idk.yaml"
-)
 for split in "${forget_retain_splits[@]}"; do
     forget_split=$(echo $split | cut -d' ' -f1)
     retain_split=$(echo $split | cut -d' ' -f2)
diff --git a/setup_data.py b/setup_data.py
index 358779c..48de0ad 100644
--- a/setup_data.py
+++ b/setup_data.py
@@ -1,17 +1,8 @@
 from huggingface_hub import snapshot_download
 
-# Setup retain model metrics
 snapshot_download(
     repo_id="open-unlearning/eval",
     allow_patterns="*.json",
     repo_type="dataset",
     local_dir="saves/eval",
 )
-
-# Setup data
-snapshot_download(
-    repo_id="open-unlearning/idk",
-    allow_patterns="*.jsonl",
-    repo_type="dataset",
-    local_dir="data",
-)
\ No newline at end of file

From f468efb9eaa0c737e6e8b4e64abb411131ff7a99 Mon Sep 17 00:00:00 2001
From: Dornavineeth <vineethdorna@gmail.com>
Date: Sun, 2 Mar 2025 19:22:21 +0000
Subject: [PATCH 5/6] download idk data

---
 setup_data.py | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/setup_data.py b/setup_data.py
index 48de0ad..358779c 100644
--- a/setup_data.py
+++ b/setup_data.py
@@ -1,8 +1,17 @@
 from huggingface_hub import snapshot_download
 
+# Setup retain model metrics
 snapshot_download(
     repo_id="open-unlearning/eval",
     allow_patterns="*.json",
     repo_type="dataset",
     local_dir="saves/eval",
 )
+
+# Setup data
+snapshot_download(
+    repo_id="open-unlearning/idk",
+    allow_patterns="*.jsonl",
+    repo_type="dataset",
+    local_dir="data",
+)
\ No newline at end of file

From ca8d5038b07ca8b8bbf0a71bf8a9a5502899f154 Mon Sep 17 00:00:00 2001
From: Dornavineeth <vineethdorna@gmail.com>
Date: Sun, 2 Mar 2025 19:22:45 +0000
Subject: [PATCH 6/6] fix dpo experiment config

---
 scripts/tofu_unlearn.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/scripts/tofu_unlearn.sh b/scripts/tofu_unlearn.sh
index a556bd1..1794c9b 100644
--- a/scripts/tofu_unlearn.sh
+++ b/scripts/tofu_unlearn.sh
@@ -13,7 +13,7 @@ trainers_experiments=(
     "GradAscent unlearn/tofu/default.yaml"
     "GradDiff unlearn/tofu/default.yaml"
     "NPO unlearn/tofu/default.yaml"
-    "DPO unlearn/tofu/default.yaml"
+    "DPO unlearn/tofu/idk.yaml"
 )
 forget_retain_splits=(
     "forget01 retain99"