Minor fixes

Signed-off-by: Vishwesh Nath <[email protected]>
Project-MONAI · Oct 24, 2024 · 880b750 · 880b750
1 parent b69faa7
commit 880b750
Show file tree

Hide file tree

Showing 9 changed files with 42 additions and 6 deletions.
diff --git a/m3/data_prepare/README.md b/m3/data_prepare/README.md
@@ -3,7 +3,7 @@
 Preparing the datasets for VILA training and testing requires three steps:
 1. Downloading all the datasets (Information to download each dataset is provided in the readme.md for the `vqa`, `report` and `expert` directories)
 2. Generating the instruction data for all datasets (Information to generate the instruction data is provided in the readme.md for the `vqa`, `report` and `expert` directory)
-3. Adding the prepared datasets to VILA in a data mixture (More information can be found in the quickstart guide))
+3. Adding the prepared datasets to VILA in a data mixture (More information can be found in the [quickstart guide](../train/readme.md))
 
 ### VQA Datasets
 - PathVQA: Pathology-based VQA dataset with ~4,000 images and ~32,000 QA pairs, focusing on microscopic views of human tissue.

diff --git a/m3/data_prepare/vqa/README.md b/m3/data_prepare/vqa/README.md
@@ -7,7 +7,7 @@ To download the RadVQA dataset, please visit this [link](https://osf.io/89kps/)
 
 From the below screenshot, please ensure that the entire OSF Storage directory is downloaded.
 
-![image](../../images/radvqa_osf.png)
+![image](../../docs/images/radvqa_osf.png)
 
 An execution of the instruction data generation python script is required on the downloaded OSF Storage data directory.
 
@@ -16,7 +16,7 @@ To download the SLAKE dataset, please visit this [link](https://www.med-vqa.com/
 
 From the below screenshot, please note that we use the 'Google Drive' link to download the dataset.
 
-![image](../../images/slake_web.png)
+![image](../../docs/images/slake_web.png)
 
 Once the dataset is downloaded, please unzip the files and place them in a temporary data directory. Execution of instruction data generation python script is required on the unzipped data directory
 
@@ -25,7 +25,7 @@ All files from the huggingface PathVQA dataset from the data directory need to b
 
 Ensure that all 13 parquet files are present in the data directory before trying to generate the instruction data (see screenshot below). The python commands for running them are provided in the next section.
 
-![image](../../images/pathvqa_hf.png)
+![image](../../docs/images/pathvqa_hf.png)
 
 ## Generate Instruction Tuning Training Data
 ### RadVQA

diff --git a/m3/images/pathvqa_hf.png → m3/docs/images/pathvqa_hf.png b/m3/images/pathvqa_hf.png → m3/docs/images/pathvqa_hf.png
diff --git a/m3/images/radvqa_osf.png → m3/docs/images/radvqa_osf.png b/m3/images/radvqa_osf.png → m3/docs/images/radvqa_osf.png
diff --git a/m3/images/slake_web.png → m3/docs/images/slake_web.png b/m3/images/slake_web.png → m3/docs/images/slake_web.png
diff --git a/m3/train/m3-vila13b.sh b/m3/train/m3-vila13b.sh
@@ -1,5 +1,16 @@
 #!/bin/bash
 
+# Copyright (c) MONAI Consortium
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#     http://www.apache.org/licenses/LICENSE-2.0
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
 # Set master address and worker list
 master_addr="127.0.0.1"
 export MASTER_ADDR=${master_addr}
@@ -45,6 +56,7 @@ CURRENT_RANK=0
 n_node=1
 echo "JobID: N/A | Full list: $worker_list | rank $CURRENT_RANK of $n_node"
 
+# Upsampling datasets to balance the training data
 HEALTHCARE_DS=$(for i in {1..10}; do echo -n usmle+; done)
 HEALTHCARE_DS+=$(for i in {1..4}; do echo -n radvqa+; done)
 HEALTHCARE_DS+=$(for i in {1..4}; do echo -n slake+; done)

diff --git a/m3/train/m3-vila3b.sh b/m3/train/m3-vila3b.sh
@@ -1,5 +1,16 @@
 #!/bin/bash
 
+# Copyright (c) MONAI Consortium
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#     http://www.apache.org/licenses/LICENSE-2.0
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
 # User-defined variables
 NODES=4
 GPUS_PER_NODE=8
@@ -35,6 +46,7 @@ n_node=${NODES:-1}
 echo "JobID: $SLURM_JOB_ID | Full list: $worker_list | rank $CURRENT_RANK of $n_node"
 
 bs=32
+# Upsampling datasets to balance the training data
 HEALTHCARE_DS=$(for i in {1..10}; do echo -n usmle+; done)
 HEALTHCARE_DS+=$(for i in {1..4}; do echo -n radvqa+; done)
 HEALTHCARE_DS+=$(for i in {1..4}; do echo -n slake+; done)

diff --git a/m3/train/m3-vila8b.sh b/m3/train/m3-vila8b.sh
@@ -1,5 +1,16 @@
 #!/bin/bash
 
+# Copyright (c) MONAI Consortium
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#     http://www.apache.org/licenses/LICENSE-2.0
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
 # Set master address and worker list
 master_addr=$(hostname)
 export MASTER_ADDR=${master_addr:-"127.0.0.1"}
@@ -40,6 +51,7 @@ bs=16
 STAGE2_PATH=$STAGE2_PATH
 OUTPUT=provide_output_dir_name
 
+# Upsampling datasets to balance the training data
 HEALTHCARE_DS=$(for i in {1..10}; do echo -n usmle+; done)
 HEALTHCARE_DS+=$(for i in {1..4}; do echo -n radvqa+; done)
 HEALTHCARE_DS+=$(for i in {1..4}; do echo -n slake+; done)

diff --git a/m3/train/readme.md b/m3/train/readme.md
@@ -1,4 +1,4 @@
-# Training Recipe For M3
+# Quickstart Guide: Training Recipe For M3
 
 The M3-VILA is based on [VILA-v1.5](https://github.com/NVlabs/VILA).
 
@@ -15,7 +15,7 @@ For each model variant type, different large language models (LLM's) and vision
 Please address the below in the training bash scripts before trying to execute them:
 
 #### Define paths (replace with actual paths)
-1.) STAGE2_PATH="/path/to/your/model", these are VILA pre-trained checkpoints, the link to download them can be found [here](https://github.com/NVlabs/VILA#pre-trained-models).
+1.) STAGE2_PATH="/path/to/your/model", these are VILA pre-trained checkpoints, they can be found [here](https://github.com/NVlabs/VILA#pre-trained-models).
 
 2.) OUTPUT_DIR="/path/to/output/checkpoints"