Init: add readme

READMD.md

@@ -0,0 +1,51 @@
+# TODO:
+- [x] Save LoRA separately
+- [ ] Load LoRA separately
+- [ ] Merge LoRA
+
+# How to use
+1. Setup the environment
+```bash
+conda create -n llava python=3.10 -y
+conda activate llava
+pip install --upgrade pip  # Enable PEP 660 support.
+pip install -e ".[train]"
+```
+
+2. Dataset
+The dataset is stored in a json file. Each item is in the following format:
+```json
+{
+        "id": "<id>",
+        "image": "/path/to/image",
+        "conversations": [
+            {
+                "from": "human",
+                "value": "<question>"
+            },
+            {
+                "from": "gpt",
+                "value": "<answer>"
+            }
+        ],
+        ...
+}
+```
+Modify the `--data_path` flag, which should be a folder containing `train.json` and `test.json`.
+
+3. Pre-trained ckpt
+Usually, huggingface will download the pretrained ckpt automatically. But sometimes it will take a lot of time if the server is in mainland. Alternatively, you can find the pretrained ckpt under `61.170.32.4:/mnt1/lyc/huggingface/hub`.
+
+4. Finetuning
+Sample bash
+```bash
+bash scripts/finetune/train_llava.sh
+```
+
+
+5. Testing
+Sample bash
+```bash
+bash scripts/finetune/test_llava.sh
+```
+