diff --git a/.idea/lisa-on-gpu.iml b/.idea/lisa-on-gpu.iml
index 2bbb30bf5e0cb1adb986d0241cb3837e35cf5060..391071cd5bbef9bf426e71d65fbb62a4f7bd625f 100644
--- a/.idea/lisa-on-gpu.iml
+++ b/.idea/lisa-on-gpu.iml
@@ -4,7 +4,7 @@
     <content url="file://$MODULE_DIR$">
       <excludeFolder url="file://$MODULE_DIR$/venv" />
     </content>
-    <orderEntry type="inheritedJdk" />
+    <orderEntry type="jdk" jdkName="Python 3.10 (lisa_on_cuda)" jdkType="Python SDK" />
     <orderEntry type="sourceFolder" forTests="false" />
   </component>
   <component name="PyDocumentationSettings">
diff --git a/.idea/misc.xml b/.idea/misc.xml
index be60524394a702531607cee53bb4fc0b295a59f1..840557bcbec0547b9ced542410c3fe94015b0e44 100644
--- a/.idea/misc.xml
+++ b/.idea/misc.xml
@@ -3,5 +3,5 @@
   <component name="Black">
     <option name="sdkName" value="Python 3.10 (lisa-on-gpu)" />
   </component>
-  <component name="ProjectRootManager" version="2" project-jdk-name="Python 3.10 (lisa-on-gpu)" project-jdk-type="Python SDK" />
+  <component name="ProjectRootManager" version="2" project-jdk-name="Python 3.10 (lisa_on_cuda)" project-jdk-type="Python SDK" />
 </project>
\ No newline at end of file
diff --git a/model/LISA.py b/lisa_on_cuda/LISA.py
similarity index 100%
rename from model/LISA.py
rename to lisa_on_cuda/LISA.py
diff --git a/app/__init__.py b/lisa_on_cuda/__init__.py
similarity index 100%
rename from app/__init__.py
rename to lisa_on_cuda/__init__.py
diff --git a/model/__init__.py b/lisa_on_cuda/app/__init__.py
similarity index 100%
rename from model/__init__.py
rename to lisa_on_cuda/app/__init__.py
diff --git a/app/chat.py b/lisa_on_cuda/app/chat.py
similarity index 95%
rename from app/chat.py
rename to lisa_on_cuda/app/chat.py
index b0330afab52b54eda4eb4d8207aeb90ef6e7403c..358e46b9ae0e0b4d2c6bb33ba088c0cf6de1465a 100755
--- a/app/chat.py
+++ b/lisa_on_cuda/app/chat.py
@@ -7,11 +7,11 @@ import numpy as np
 import torch
 from transformers import AutoTokenizer, BitsAndBytesConfig, CLIPImageProcessor
 
-from model.LISA import LISAForCausalLM
-from model.llava import conversation as conversation_lib
-from model.llava.mm_utils import tokenizer_image_token
-from model.segment_anything.utils.transforms import ResizeLongestSide
-from utils import app_helpers, utils
+from lisa_on_cuda.LISA import LISAForCausalLM
+from lisa_on_cuda.llava import conversation as conversation_lib
+from lisa_on_cuda.llava.mm_utils import tokenizer_image_token
+from lisa_on_cuda.segment_anything.utils.transforms import ResizeLongestSide
+from ..utils import app_helpers, utils
 
 
 def main(args):
diff --git a/app/main.py b/lisa_on_cuda/app/main.py
similarity index 93%
rename from app/main.py
rename to lisa_on_cuda/app/main.py
index 1a65a38685e04de00a4f71f488916616c436f136..b6302b396cb47cdceac228b4c278e88025bee0a6 100755
--- a/app/main.py
+++ b/lisa_on_cuda/app/main.py
@@ -6,8 +6,8 @@ from fastapi import FastAPI
 from fastapi.staticfiles import StaticFiles
 from fastapi.templating import Jinja2Templates
 
-from app import routes
-from utils import app_helpers, session_logger
+from . import routes
+from ..utils import app_helpers, session_logger
 
 
 session_logger.change_logging(logging.DEBUG)
diff --git a/app/merge_lora_weights_and_save_hf_model.py b/lisa_on_cuda/app/merge_lora_weights_and_save_hf_model.py
similarity index 97%
rename from app/merge_lora_weights_and_save_hf_model.py
rename to lisa_on_cuda/app/merge_lora_weights_and_save_hf_model.py
index 1b41b7ac0cae033dc060e8a2ea39efbf5d7a8389..2f94f232190d1c088a5f863c59a07d67755f7a88 100755
--- a/app/merge_lora_weights_and_save_hf_model.py
+++ b/lisa_on_cuda/app/merge_lora_weights_and_save_hf_model.py
@@ -11,8 +11,8 @@ import transformers
 from peft import LoraConfig, get_peft_model
 from transformers import AutoTokenizer
 
-from model.LISA import LISAForCausalLM
-from utils.utils import DEFAULT_IM_END_TOKEN, DEFAULT_IM_START_TOKEN
+from lisa_on_cuda.LISA import LISAForCausalLM
+from ..utils.utils import DEFAULT_IM_END_TOKEN, DEFAULT_IM_START_TOKEN
 
 
 def parse_args(args):
diff --git a/app/routes.py b/lisa_on_cuda/app/routes.py
similarity index 91%
rename from app/routes.py
rename to lisa_on_cuda/app/routes.py
index fa24e078f260d83ec6ce75336ab74ca47df0e53f..6e2d72b8cfec99cc1197d497c8eaa70c5c432dfd 100644
--- a/app/routes.py
+++ b/lisa_on_cuda/app/routes.py
@@ -2,7 +2,7 @@ import json
 import logging
 from fastapi import APIRouter
 
-from utils import session_logger
+from ..utils import session_logger
 
 
 router = APIRouter()
diff --git a/app/train_ds.py b/lisa_on_cuda/app/train_ds.py
similarity index 98%
rename from app/train_ds.py
rename to lisa_on_cuda/app/train_ds.py
index 944e60b7affb9ec14aa85b876d824f7f2720ea1b..0f5a471161f288110c565ac6549ef891255fa4b0 100755
--- a/app/train_ds.py
+++ b/lisa_on_cuda/app/train_ds.py
@@ -13,10 +13,10 @@ import transformers
 from peft import LoraConfig, get_peft_model
 from torch.utils.tensorboard import SummaryWriter
 
-from model.LISA import LISAForCausalLM
-from model.llava import conversation as conversation_lib
-from utils.dataset import HybridDataset, ValDataset, collate_fn
-from utils.utils import (DEFAULT_IM_END_TOKEN, DEFAULT_IM_START_TOKEN,
+from lisa_on_cuda.LISA import LISAForCausalLM
+from lisa_on_cuda.llava import conversation as conversation_lib
+from ..utils.dataset import HybridDataset, ValDataset, collate_fn
+from ..utils.utils import (DEFAULT_IM_END_TOKEN, DEFAULT_IM_START_TOKEN,
                          AverageMeter, ProgressMeter, Summary, dict_to_cuda,
                          intersectionAndUnionGPU)
 
diff --git a/model/llava/__init__.py b/lisa_on_cuda/llava/__init__.py
similarity index 100%
rename from model/llava/__init__.py
rename to lisa_on_cuda/llava/__init__.py
diff --git a/model/llava/constants.py b/lisa_on_cuda/llava/constants.py
similarity index 100%
rename from model/llava/constants.py
rename to lisa_on_cuda/llava/constants.py
diff --git a/model/llava/conversation.py b/lisa_on_cuda/llava/conversation.py
similarity index 100%
rename from model/llava/conversation.py
rename to lisa_on_cuda/llava/conversation.py
diff --git a/model/llava/mm_utils.py b/lisa_on_cuda/llava/mm_utils.py
similarity index 100%
rename from model/llava/mm_utils.py
rename to lisa_on_cuda/llava/mm_utils.py
diff --git a/model/llava/model/__init__.py b/lisa_on_cuda/llava/model/__init__.py
similarity index 100%
rename from model/llava/model/__init__.py
rename to lisa_on_cuda/llava/model/__init__.py
diff --git a/model/llava/model/apply_delta.py b/lisa_on_cuda/llava/model/apply_delta.py
similarity index 96%
rename from model/llava/model/apply_delta.py
rename to lisa_on_cuda/llava/model/apply_delta.py
index 2f73809262b001a1f16ca3302cd75ab30893486a..9bc0be9e04242891ff6bddb807bd2301f891239d 100755
--- a/model/llava/model/apply_delta.py
+++ b/lisa_on_cuda/llava/model/apply_delta.py
@@ -5,10 +5,11 @@ python3 -m fastchat.model.apply_delta --base ~/model_weights/llama-7b --target ~
 import argparse
 
 import torch
-from llava import LlavaLlamaForCausalLM
 from tqdm import tqdm
 from transformers import AutoModelForCausalLM, AutoTokenizer
 
+from .language_model.llava_llama import LlavaLlamaForCausalLM
+
 
 def apply_delta(base_model_path, target_model_path, delta_path):
     print("Loading base model")
diff --git a/model/llava/model/builder.py b/lisa_on_cuda/llava/model/builder.py
similarity index 96%
rename from model/llava/model/builder.py
rename to lisa_on_cuda/llava/model/builder.py
index 0c841ab48b765184f05eac3326b36bfa1a7a4819..360c91be89d3dcfd47edba0d7f282eff4a8cae61 100755
--- a/model/llava/model/builder.py
+++ b/lisa_on_cuda/llava/model/builder.py
@@ -17,11 +17,11 @@ import os
 import shutil
 
 import torch
-from llava.constants import (DEFAULT_IM_END_TOKEN, DEFAULT_IM_START_TOKEN,
-                             DEFAULT_IMAGE_PATCH_TOKEN)
-from llava.model import *
-from transformers import (AutoConfig, AutoModelForCausalLM, AutoTokenizer,
-                          BitsAndBytesConfig)
+from transformers import AutoConfig, AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
+
+from .language_model.llava_llama import LlavaLlamaForCausalLM
+from .language_model.llava_mpt import LlavaMPTForCausalLM
+from ..constants import DEFAULT_IM_END_TOKEN, DEFAULT_IM_START_TOKEN, DEFAULT_IMAGE_PATCH_TOKEN
 
 
 def load_pretrained_model(
diff --git a/model/llava/model/consolidate.py b/lisa_on_cuda/llava/model/consolidate.py
similarity index 92%
rename from model/llava/model/consolidate.py
rename to lisa_on_cuda/llava/model/consolidate.py
index f1fd9b722bc2afe5338eb632dee2d09cc27367ca..3a81b7db5645ee1d35ef81b21d50ceaeb62990f5 100755
--- a/model/llava/model/consolidate.py
+++ b/lisa_on_cuda/llava/model/consolidate.py
@@ -5,8 +5,7 @@ python3 -m llava.model.consolidate --src ~/model_weights/llava-7b --dst ~/model_
 import argparse
 
 import torch
-from llava.model import *
-from llava.model.utils import auto_upgrade
+from .utils import auto_upgrade
 from transformers import AutoModelForCausalLM, AutoTokenizer
 
 
diff --git a/model/llava/model/language_model/llava_llama.py b/lisa_on_cuda/llava/model/language_model/llava_llama.py
similarity index 100%
rename from model/llava/model/language_model/llava_llama.py
rename to lisa_on_cuda/llava/model/language_model/llava_llama.py
diff --git a/model/llava/model/language_model/llava_mpt.py b/lisa_on_cuda/llava/model/language_model/llava_mpt.py
similarity index 100%
rename from model/llava/model/language_model/llava_mpt.py
rename to lisa_on_cuda/llava/model/language_model/llava_mpt.py
diff --git a/model/llava/model/language_model/mpt/adapt_tokenizer.py b/lisa_on_cuda/llava/model/language_model/mpt/adapt_tokenizer.py
similarity index 100%
rename from model/llava/model/language_model/mpt/adapt_tokenizer.py
rename to lisa_on_cuda/llava/model/language_model/mpt/adapt_tokenizer.py
diff --git a/model/llava/model/language_model/mpt/attention.py b/lisa_on_cuda/llava/model/language_model/mpt/attention.py
similarity index 100%
rename from model/llava/model/language_model/mpt/attention.py
rename to lisa_on_cuda/llava/model/language_model/mpt/attention.py
diff --git a/model/llava/model/language_model/mpt/blocks.py b/lisa_on_cuda/llava/model/language_model/mpt/blocks.py
similarity index 100%
rename from model/llava/model/language_model/mpt/blocks.py
rename to lisa_on_cuda/llava/model/language_model/mpt/blocks.py
diff --git a/model/llava/model/language_model/mpt/configuration_mpt.py b/lisa_on_cuda/llava/model/language_model/mpt/configuration_mpt.py
similarity index 100%
rename from model/llava/model/language_model/mpt/configuration_mpt.py
rename to lisa_on_cuda/llava/model/language_model/mpt/configuration_mpt.py
diff --git a/model/llava/model/language_model/mpt/custom_embedding.py b/lisa_on_cuda/llava/model/language_model/mpt/custom_embedding.py
similarity index 100%
rename from model/llava/model/language_model/mpt/custom_embedding.py
rename to lisa_on_cuda/llava/model/language_model/mpt/custom_embedding.py
diff --git a/model/llava/model/language_model/mpt/flash_attn_triton.py b/lisa_on_cuda/llava/model/language_model/mpt/flash_attn_triton.py
similarity index 100%
rename from model/llava/model/language_model/mpt/flash_attn_triton.py
rename to lisa_on_cuda/llava/model/language_model/mpt/flash_attn_triton.py
diff --git a/model/llava/model/language_model/mpt/hf_prefixlm_converter.py b/lisa_on_cuda/llava/model/language_model/mpt/hf_prefixlm_converter.py
similarity index 100%
rename from model/llava/model/language_model/mpt/hf_prefixlm_converter.py
rename to lisa_on_cuda/llava/model/language_model/mpt/hf_prefixlm_converter.py
diff --git a/model/llava/model/language_model/mpt/meta_init_context.py b/lisa_on_cuda/llava/model/language_model/mpt/meta_init_context.py
similarity index 100%
rename from model/llava/model/language_model/mpt/meta_init_context.py
rename to lisa_on_cuda/llava/model/language_model/mpt/meta_init_context.py
diff --git a/model/llava/model/language_model/mpt/modeling_mpt.py b/lisa_on_cuda/llava/model/language_model/mpt/modeling_mpt.py
similarity index 100%
rename from model/llava/model/language_model/mpt/modeling_mpt.py
rename to lisa_on_cuda/llava/model/language_model/mpt/modeling_mpt.py
diff --git a/model/llava/model/language_model/mpt/norm.py b/lisa_on_cuda/llava/model/language_model/mpt/norm.py
similarity index 100%
rename from model/llava/model/language_model/mpt/norm.py
rename to lisa_on_cuda/llava/model/language_model/mpt/norm.py
diff --git a/model/llava/model/language_model/mpt/param_init_fns.py b/lisa_on_cuda/llava/model/language_model/mpt/param_init_fns.py
similarity index 100%
rename from model/llava/model/language_model/mpt/param_init_fns.py
rename to lisa_on_cuda/llava/model/language_model/mpt/param_init_fns.py
diff --git a/model/llava/model/llava_arch.py b/lisa_on_cuda/llava/model/llava_arch.py
similarity index 98%
rename from model/llava/model/llava_arch.py
rename to lisa_on_cuda/llava/model/llava_arch.py
index 049ea7e6541b12ca921bfaf8a0944351bf062e62..4468579f185019a154419dcb0dfaf499f87133c9 100755
--- a/model/llava/model/llava_arch.py
+++ b/lisa_on_cuda/llava/model/llava_arch.py
@@ -18,10 +18,7 @@ from abc import ABC, abstractmethod
 import torch
 import torch.nn as nn
 
-# from llava.constants import IGNORE_INDEX, IMAGE_TOKEN_INDEX, DEFAULT_IMAGE_PATCH_TOKEN, DEFAULT_IM_START_TOKEN, DEFAULT_IM_END_TOKEN
-from utils.utils import (DEFAULT_IM_END_TOKEN, DEFAULT_IM_START_TOKEN,
-                         DEFAULT_IMAGE_PATCH_TOKEN, IGNORE_INDEX,
-                         IMAGE_TOKEN_INDEX)
+from lisa_on_cuda.utils.utils import IGNORE_INDEX, IMAGE_TOKEN_INDEX
 
 from .multimodal_encoder.builder import build_vision_tower
 
diff --git a/model/llava/model/make_delta.py b/lisa_on_cuda/llava/model/make_delta.py
similarity index 98%
rename from model/llava/model/make_delta.py
rename to lisa_on_cuda/llava/model/make_delta.py
index 26d73d2474e2da7f62955c6685c8812d6d94f6ad..1ac1a984c8c9560e010eedfb0c150c7e221abab2 100755
--- a/model/llava/model/make_delta.py
+++ b/lisa_on_cuda/llava/model/make_delta.py
@@ -5,7 +5,7 @@ python3 -m llava.model.make_delta --base ~/model_weights/llama-7b --target ~/mod
 import argparse
 
 import torch
-from llava.model.utils import auto_upgrade
+from .utils import auto_upgrade
 from tqdm import tqdm
 from transformers import AutoModelForCausalLM, AutoTokenizer
 
diff --git a/model/llava/model/multimodal_encoder/builder.py b/lisa_on_cuda/llava/model/multimodal_encoder/builder.py
similarity index 100%
rename from model/llava/model/multimodal_encoder/builder.py
rename to lisa_on_cuda/llava/model/multimodal_encoder/builder.py
diff --git a/model/llava/model/multimodal_encoder/clip_encoder.py b/lisa_on_cuda/llava/model/multimodal_encoder/clip_encoder.py
similarity index 100%
rename from model/llava/model/multimodal_encoder/clip_encoder.py
rename to lisa_on_cuda/llava/model/multimodal_encoder/clip_encoder.py
diff --git a/model/llava/model/utils.py b/lisa_on_cuda/llava/model/utils.py
similarity index 100%
rename from model/llava/model/utils.py
rename to lisa_on_cuda/llava/model/utils.py
diff --git a/model/llava/train/llama_flash_attn_monkey_patch.py b/lisa_on_cuda/llava/train/llama_flash_attn_monkey_patch.py
similarity index 100%
rename from model/llava/train/llama_flash_attn_monkey_patch.py
rename to lisa_on_cuda/llava/train/llama_flash_attn_monkey_patch.py
diff --git a/model/llava/train/llava_trainer.py b/lisa_on_cuda/llava/train/llava_trainer.py
similarity index 100%
rename from model/llava/train/llava_trainer.py
rename to lisa_on_cuda/llava/train/llava_trainer.py
diff --git a/model/llava/train/train.py b/lisa_on_cuda/llava/train/train.py
similarity index 100%
rename from model/llava/train/train.py
rename to lisa_on_cuda/llava/train/train.py
diff --git a/model/llava/train/train_mem.py b/lisa_on_cuda/llava/train/train_mem.py
similarity index 74%
rename from model/llava/train/train_mem.py
rename to lisa_on_cuda/llava/train/train_mem.py
index f3940cf7fea248d055a9cb333a08ebca0f782885..2f4e94b5b0a23c13f0d1d7906a6b0136c5f3ffbd 100755
--- a/model/llava/train/train_mem.py
+++ b/lisa_on_cuda/llava/train/train_mem.py
@@ -3,12 +3,12 @@
 # Make it more memory efficient by monkey patching the LLaMA model with FlashAttn.
 
 # Need to call this before importing transformers.
-from llava.train.llama_flash_attn_monkey_patch import \
-    replace_llama_attn_with_flash_attn
+from .llama_flash_attn_monkey_patch import replace_llama_attn_with_flash_attn
 
 replace_llama_attn_with_flash_attn()
 
-from llava.train.train import train
+from .train import train
+
 
 if __name__ == "__main__":
     train()
diff --git a/model/llava/utils.py b/lisa_on_cuda/llava/utils.py
similarity index 99%
rename from model/llava/utils.py
rename to lisa_on_cuda/llava/utils.py
index 0a2d5fd533ded77352f5548a0ed027b700365ea4..b1527d0c82db1a0aa0f5a34c742865e8b3828bed 100755
--- a/model/llava/utils.py
+++ b/lisa_on_cuda/llava/utils.py
@@ -5,7 +5,7 @@ import os
 import sys
 
 import requests
-from llava.constants import LOGDIR
+from .constants import LOGDIR
 
 server_error_msg = (
     "**NETWORK ERROR DUE TO HIGH TRAFFIC. PLEASE REGENERATE OR REFRESH THIS PAGE.**"
diff --git a/model/segment_anything/__init__.py b/lisa_on_cuda/segment_anything/__init__.py
similarity index 100%
rename from model/segment_anything/__init__.py
rename to lisa_on_cuda/segment_anything/__init__.py
diff --git a/model/segment_anything/automatic_mask_generator.py b/lisa_on_cuda/segment_anything/automatic_mask_generator.py
similarity index 100%
rename from model/segment_anything/automatic_mask_generator.py
rename to lisa_on_cuda/segment_anything/automatic_mask_generator.py
diff --git a/model/segment_anything/build_sam.py b/lisa_on_cuda/segment_anything/build_sam.py
similarity index 100%
rename from model/segment_anything/build_sam.py
rename to lisa_on_cuda/segment_anything/build_sam.py
diff --git a/model/segment_anything/modeling/__init__.py b/lisa_on_cuda/segment_anything/modeling/__init__.py
similarity index 100%
rename from model/segment_anything/modeling/__init__.py
rename to lisa_on_cuda/segment_anything/modeling/__init__.py
diff --git a/model/segment_anything/modeling/common.py b/lisa_on_cuda/segment_anything/modeling/common.py
similarity index 100%
rename from model/segment_anything/modeling/common.py
rename to lisa_on_cuda/segment_anything/modeling/common.py
diff --git a/model/segment_anything/modeling/image_encoder.py b/lisa_on_cuda/segment_anything/modeling/image_encoder.py
similarity index 100%
rename from model/segment_anything/modeling/image_encoder.py
rename to lisa_on_cuda/segment_anything/modeling/image_encoder.py
diff --git a/model/segment_anything/modeling/mask_decoder.py b/lisa_on_cuda/segment_anything/modeling/mask_decoder.py
similarity index 100%
rename from model/segment_anything/modeling/mask_decoder.py
rename to lisa_on_cuda/segment_anything/modeling/mask_decoder.py
diff --git a/model/segment_anything/modeling/prompt_encoder.py b/lisa_on_cuda/segment_anything/modeling/prompt_encoder.py
similarity index 100%
rename from model/segment_anything/modeling/prompt_encoder.py
rename to lisa_on_cuda/segment_anything/modeling/prompt_encoder.py
diff --git a/model/segment_anything/modeling/sam.py b/lisa_on_cuda/segment_anything/modeling/sam.py
similarity index 100%
rename from model/segment_anything/modeling/sam.py
rename to lisa_on_cuda/segment_anything/modeling/sam.py
diff --git a/model/segment_anything/modeling/transformer.py b/lisa_on_cuda/segment_anything/modeling/transformer.py
similarity index 100%
rename from model/segment_anything/modeling/transformer.py
rename to lisa_on_cuda/segment_anything/modeling/transformer.py
diff --git a/model/segment_anything/predictor.py b/lisa_on_cuda/segment_anything/predictor.py
similarity index 100%
rename from model/segment_anything/predictor.py
rename to lisa_on_cuda/segment_anything/predictor.py
diff --git a/model/segment_anything/utils/__init__.py b/lisa_on_cuda/segment_anything/utils/__init__.py
similarity index 100%
rename from model/segment_anything/utils/__init__.py
rename to lisa_on_cuda/segment_anything/utils/__init__.py
diff --git a/model/segment_anything/utils/amg.py b/lisa_on_cuda/segment_anything/utils/amg.py
similarity index 100%
rename from model/segment_anything/utils/amg.py
rename to lisa_on_cuda/segment_anything/utils/amg.py
diff --git a/model/segment_anything/utils/onnx.py b/lisa_on_cuda/segment_anything/utils/onnx.py
similarity index 100%
rename from model/segment_anything/utils/onnx.py
rename to lisa_on_cuda/segment_anything/utils/onnx.py
diff --git a/model/segment_anything/utils/transforms.py b/lisa_on_cuda/segment_anything/utils/transforms.py
similarity index 100%
rename from model/segment_anything/utils/transforms.py
rename to lisa_on_cuda/segment_anything/utils/transforms.py
diff --git a/lisa_on_cuda/utils/__init__.py b/lisa_on_cuda/utils/__init__.py
new file mode 100644
index 0000000000000000000000000000000000000000..0710ce15051e354c4a3e168e001d779ff538f3cb
--- /dev/null
+++ b/lisa_on_cuda/utils/__init__.py
@@ -0,0 +1,3 @@
+from . import app_helpers
+from . import constants
+from . import utils
diff --git a/utils/ade20k_classes.json b/lisa_on_cuda/utils/ade20k_classes.json
similarity index 100%
rename from utils/ade20k_classes.json
rename to lisa_on_cuda/utils/ade20k_classes.json
diff --git a/utils/app_helpers.py b/lisa_on_cuda/utils/app_helpers.py
similarity index 97%
rename from utils/app_helpers.py
rename to lisa_on_cuda/utils/app_helpers.py
index 62ef9dcac1710f22a148ebb5d859041e94938377..f827ea99ca74b5fb23f1934396407a7e9e13365e 100644
--- a/utils/app_helpers.py
+++ b/lisa_on_cuda/utils/app_helpers.py
@@ -12,10 +12,11 @@ import torch.nn.functional as F
 from transformers import AutoTokenizer, BitsAndBytesConfig, CLIPImageProcessor
 
 from . import constants, session_logger, utils
-from model.LISA import LISAForCausalLM
-from model.llava import conversation as conversation_lib
-from model.llava.mm_utils import tokenizer_image_token
-from model.segment_anything.utils.transforms import ResizeLongestSide
+from lisa_on_cuda.LISA import LISAForCausalLM
+from lisa_on_cuda.llava import conversation as conversation_lib
+from lisa_on_cuda.llava.mm_utils import tokenizer_image_token
+from lisa_on_cuda.segment_anything.utils.transforms import ResizeLongestSide
+
 
 placeholders = utils.create_placeholder_variables()
 
diff --git a/utils/cocostuff_classes.txt b/lisa_on_cuda/utils/cocostuff_classes.txt
similarity index 100%
rename from utils/cocostuff_classes.txt
rename to lisa_on_cuda/utils/cocostuff_classes.txt
diff --git a/utils/constants.py b/lisa_on_cuda/utils/constants.py
similarity index 100%
rename from utils/constants.py
rename to lisa_on_cuda/utils/constants.py
diff --git a/utils/conversation.py b/lisa_on_cuda/utils/conversation.py
similarity index 100%
rename from utils/conversation.py
rename to lisa_on_cuda/utils/conversation.py
diff --git a/utils/data_processing.py b/lisa_on_cuda/utils/data_processing.py
similarity index 100%
rename from utils/data_processing.py
rename to lisa_on_cuda/utils/data_processing.py
diff --git a/utils/dataset.py b/lisa_on_cuda/utils/dataset.py
similarity index 98%
rename from utils/dataset.py
rename to lisa_on_cuda/utils/dataset.py
index 30ab766743aeb9c013986aac133b276b93db6b9d..69c503c1ca7c8248d025091dc2ac07982e07741b 100755
--- a/utils/dataset.py
+++ b/lisa_on_cuda/utils/dataset.py
@@ -9,11 +9,11 @@ import torch.nn.functional as F
 from pycocotools import mask
 from transformers import CLIPImageProcessor
 
-from model.llava import conversation as conversation_lib
-from model.llava.constants import (DEFAULT_IMAGE_TOKEN, IGNORE_INDEX,
+from lisa_on_cuda.llava import conversation as conversation_lib
+from lisa_on_cuda.llava.constants import (DEFAULT_IMAGE_TOKEN, IGNORE_INDEX,
                                    IMAGE_TOKEN_INDEX)
-from model.llava.mm_utils import tokenizer_image_token
-from model.segment_anything.utils.transforms import ResizeLongestSide
+from lisa_on_cuda.llava.mm_utils import tokenizer_image_token
+from lisa_on_cuda.segment_anything.utils.transforms import ResizeLongestSide
 
 from .conversation import get_default_conv_template
 from .data_processing import get_mask_from_json
diff --git a/utils/grefcoco.py b/lisa_on_cuda/utils/grefcoco.py
similarity index 100%
rename from utils/grefcoco.py
rename to lisa_on_cuda/utils/grefcoco.py
diff --git a/utils/grefer.py b/lisa_on_cuda/utils/grefer.py
similarity index 100%
rename from utils/grefer.py
rename to lisa_on_cuda/utils/grefer.py
diff --git a/utils/reason_seg_dataset.py b/lisa_on_cuda/utils/reason_seg_dataset.py
similarity index 98%
rename from utils/reason_seg_dataset.py
rename to lisa_on_cuda/utils/reason_seg_dataset.py
index 62012092250f0279a179691c88d9cfa213b34463..25beb734bfa08e30284bcfe6b81eb795edafce57 100755
--- a/utils/reason_seg_dataset.py
+++ b/lisa_on_cuda/utils/reason_seg_dataset.py
@@ -9,8 +9,8 @@ import torch
 import torch.nn.functional as F
 from transformers import CLIPImageProcessor
 
-from model.llava import conversation as conversation_lib
-from model.segment_anything.utils.transforms import ResizeLongestSide
+from lisa_on_cuda.llava import conversation as conversation_lib
+from lisa_on_cuda.segment_anything.utils.transforms import ResizeLongestSide
 
 from .data_processing import get_mask_from_json
 from .utils import (ANSWER_LIST, DEFAULT_IMAGE_TOKEN,
diff --git a/utils/refer.py b/lisa_on_cuda/utils/refer.py
similarity index 100%
rename from utils/refer.py
rename to lisa_on_cuda/utils/refer.py
diff --git a/utils/refer_seg_dataset.py b/lisa_on_cuda/utils/refer_seg_dataset.py
similarity index 98%
rename from utils/refer_seg_dataset.py
rename to lisa_on_cuda/utils/refer_seg_dataset.py
index e28273d9d54a6882ee0f7388c4b85ebc974d6cbf..759e69d2a02ca2bac3fecae68a19ce8921993cab 100755
--- a/utils/refer_seg_dataset.py
+++ b/lisa_on_cuda/utils/refer_seg_dataset.py
@@ -8,8 +8,8 @@ import torch.nn.functional as F
 from pycocotools import mask
 from transformers import CLIPImageProcessor
 
-from model.llava import conversation as conversation_lib
-from model.segment_anything.utils.transforms import ResizeLongestSide
+from lisa_on_cuda.llava import conversation as conversation_lib
+from lisa_on_cuda.segment_anything.utils.transforms import ResizeLongestSide
 
 from .grefer import G_REFER
 from .refer import REFER
diff --git a/utils/sem_seg_dataset.py b/lisa_on_cuda/utils/sem_seg_dataset.py
similarity index 98%
rename from utils/sem_seg_dataset.py
rename to lisa_on_cuda/utils/sem_seg_dataset.py
index db2ba063cdf4014c26b46d5cff4e9eaba26bc8ec..0825125ae594496a895be6c36d6fef7c1ce7b0ce 100755
--- a/utils/sem_seg_dataset.py
+++ b/lisa_on_cuda/utils/sem_seg_dataset.py
@@ -11,8 +11,8 @@ from PIL import Image
 from pycocotools.coco import COCO
 from transformers import CLIPImageProcessor
 
-from model.llava import conversation as conversation_lib
-from model.segment_anything.utils.transforms import ResizeLongestSide
+from lisa_on_cuda.llava import conversation as conversation_lib
+from lisa_on_cuda.segment_anything.utils.transforms import ResizeLongestSide
 
 from .utils import ANSWER_LIST, SHORT_QUESTION_LIST
 
diff --git a/utils/session_logger.py b/lisa_on_cuda/utils/session_logger.py
similarity index 100%
rename from utils/session_logger.py
rename to lisa_on_cuda/utils/session_logger.py
diff --git a/utils/utils.py b/lisa_on_cuda/utils/utils.py
similarity index 99%
rename from utils/utils.py
rename to lisa_on_cuda/utils/utils.py
index 80a4baa254f075a740358738470edee5d18817f4..b6eb198123c82db54b4431876d78016bd047ec65 100755
--- a/utils/utils.py
+++ b/lisa_on_cuda/utils/utils.py
@@ -43,7 +43,7 @@ ANSWER_LIST = [
     "Sure, the segmentation result is [SEG].",
     "[SEG].",
 ]
-ROOT = Path(__file__).parent.parent
+ROOT = Path(__file__).parent.parent.parent
 
 
 class Summary(Enum):
diff --git a/utils/vqa_dataset.py b/lisa_on_cuda/utils/vqa_dataset.py
similarity index 96%
rename from utils/vqa_dataset.py
rename to lisa_on_cuda/utils/vqa_dataset.py
index dd9c6e909f25d558cbf3a5de604405ddf617f6af..bb7cf0aacf4001d475999441f1b13377cdeaf6d5 100755
--- a/utils/vqa_dataset.py
+++ b/lisa_on_cuda/utils/vqa_dataset.py
@@ -7,8 +7,8 @@ import torch
 import torch.nn.functional as F
 from transformers import CLIPImageProcessor
 
-from model.llava import conversation as conversation_lib
-from model.segment_anything.utils.transforms import ResizeLongestSide
+from lisa_on_cuda.llava import conversation as conversation_lib
+from lisa_on_cuda.segment_anything.utils.transforms import ResizeLongestSide
 
 from .utils import DEFAULT_IMAGE_TOKEN
 
diff --git a/tests/test_app_helpers.py b/tests/test_app_helpers.py
index a7fa4a3e135b4aa0938dceb912a4497fd3545642..9d1376d96fb22bf9eb83d634136ae9a24a1cd49d 100644
--- a/tests/test_app_helpers.py
+++ b/tests/test_app_helpers.py
@@ -5,7 +5,7 @@ import unittest
 class TestAppBuilders(unittest.TestCase):
 
     def test_default_creation(self):
-        from utils import utils
+        from lisa_on_cuda.utils import utils
 
         placeholders = utils.create_placeholder_variables()
         self.assertIsInstance(placeholders, dict)
@@ -13,7 +13,7 @@ class TestAppBuilders(unittest.TestCase):
         assert placeholders["error_happened"].shape == (512, 512, 3)
 
     def test_parse_args(self):
-        from utils import app_helpers
+        from lisa_on_cuda.utils import app_helpers
 
         test_args_parse = app_helpers.parse_args([])
         assert vars(test_args_parse) == {
@@ -34,7 +34,7 @@ class TestAppBuilders(unittest.TestCase):
     def test_inference(self):
         import cv2
         import numpy as np
-        from utils import app_helpers, constants, utils
+        from lisa_on_cuda.utils import app_helpers, constants, utils
 
         max_diff = 0.02
 
diff --git a/utils/__init__.py b/utils/__init__.py
deleted file mode 100644
index e69de29bb2d1d6434b8b29ae775ad8c2e48c5391..0000000000000000000000000000000000000000