feat(train): eval dataset pointed

AdityaNG · Mar 3, 2024 · 819cd8f · 819cd8f
1 parent 1fb637d
commit 819cd8f
Show file tree

Hide file tree

Showing 3 changed files with 22 additions and 7 deletions.
diff --git a/LLaVA/llava/train/train.py b/LLaVA/llava/train/train.py
@@ -780,8 +780,12 @@ def make_supervised_data_module(tokenizer: transformers.PreTrainedTokenizer,
                                 data_path=data_args.data_path,
                                 data_args=data_args)
     data_collator = DataCollatorForSupervisedDataset(tokenizer=tokenizer)
+
+    eval_dataset = LazySupervisedDataset(tokenizer=tokenizer,
+                                data_path=data_args.validation_data_path,
+                                data_args=data_args)
     return dict(train_dataset=train_dataset,
-                eval_dataset=None,
+                eval_dataset=eval_dataset,
                 data_collator=data_collator)
 
 

diff --git a/drivellava/scripts/compile_jsons.py b/drivellava/scripts/compile_jsons.py
@@ -10,9 +10,11 @@
 from drivellava.constants import ENCODED_JSON, VAL_ENCODED_JSON
 from drivellava.sparse_llava_dataset import get_drivellava_prompt
 from drivellava.trajectory_encoder import (
-    TrajectoryEncoder, NUM_TRAJECTORY_TEMPLATES
+    NUM_TRAJECTORY_TEMPLATES,
+    TrajectoryEncoder,
 )
 
+
 def load_json_dataset(
     json_list: List[str],
     trajectory_encoder: TrajectoryEncoder,
@@ -57,8 +59,12 @@ def main():
     random.shuffle(train)
     random.shuffle(val)
 
-    new_train_json_path = os.path.abspath(f"checkpoints/train_{str(NUM_TRAJECTORY_TEMPLATES)}.json")
-    new_val_json_path = os.path.abspath(f"checkpoints/val_{NUM_TRAJECTORY_TEMPLATES}.json")
+    new_train_json_path = os.path.abspath(
+        f"checkpoints/train_{str(NUM_TRAJECTORY_TEMPLATES)}.json"
+    )
+    new_val_json_path = os.path.abspath(
+        f"checkpoints/val_{NUM_TRAJECTORY_TEMPLATES}.json"
+    )
 
     # Save train to a temp file
     with open(new_train_json_path, "w", encoding="utf-8") as f:

diff --git a/drivellava/scripts/train.py b/drivellava/scripts/train.py
@@ -74,8 +74,12 @@ def load_json_dataset_balanced(
 
 def main():
 
-    train_json_path = os.path.join(COMMAVQ_DIR, f"train_{str(NUM_TRAJECTORY_TEMPLATES)}.json")
-    val_json_path = os.path.join(COMMAVQ_DIR, f"val_{str(NUM_TRAJECTORY_TEMPLATES)}.json")
+    train_json_path = os.path.join(
+        COMMAVQ_DIR, f"train_{str(NUM_TRAJECTORY_TEMPLATES)}.json"
+    )
+    val_json_path = os.path.join(
+        COMMAVQ_DIR, f"val_{str(NUM_TRAJECTORY_TEMPLATES)}.json"
+    )
 
     train = load_json_dataset_balanced(
         [
@@ -113,7 +117,7 @@ def main():
     DEEPSPEED_JSON = os.path.abspath("./config/zero3.json")
     MODEL_NAME = "liuhaotian/llava-v1.5-7b"
     DATA_PATH = new_train_json_path  # Replace with your JSON data path
-    # VAL_DATA_PATH = new_val_json_path
+    VAL_DATA_PATH = new_val_json_path
     IMAGE_FOLDER = os.path.expanduser(
         "~/Datasets/commavq"
     )  # Replace with your image folder path
@@ -131,6 +135,7 @@ def main():
         --model_name_or_path {MODEL_NAME} \
         --version llava_llama_2 \
         --data_path {DATA_PATH} \
+        --validation_data_path {VAL_DATA_PATH} \
         --image_folder {IMAGE_FOLDER} \
         --vision_tower {VISION_TOWER} \
         --mm_projector_type mlp2x_gelu \