latest peft makes model.save_pretrained in finetune.py save a 443 B adapter_model.bin which is clearly incorrect (normally adapter_model.bin should > 16 MB)

LLaMA_LoRA.ipynb

@@ -220,7 +220,7 @@
       "source": [
         "![ ! -d llama_lora ] && git clone -b {llama_lora_project_branch} --filter=tree:0 {llama_lora_project_url} llama_lora\n",
         "!cd llama_lora && git add --all && git stash && git fetch origin {llama_lora_project_branch} && git checkout {llama_lora_project_branch} && git reset origin/{llama_lora_project_branch} --hard\n",
-        "![ ! -f llama-lora-requirements-installed ] && cd llama_lora && pip install -r requirements.txt && touch ../llama-lora-requirements-installed"
+        "![ ! -f llama-lora-requirements-installed ] && cd llama_lora && pip install -r requirements.lock.txt && touch ../llama-lora-requirements-installed"
       ],
       "metadata": {
         "id": "JGYz2VDoAzC8"

README.md

@@ -56,7 +56,7 @@ file_mounts:
 # Clone the LLaMA-LoRA repo and install its dependencies.
 setup: |
   git clone https://github.com/zetavg/LLaMA-LoRA.git llama_lora
-  cd llama_lora && pip install -r requirements.txt
+  cd llama_lora && pip install -r requirements.lock.txt
   cd ..
   echo 'Dependencies installed.'
 
@@ -86,13 +86,13 @@ When you are done, run `sky stop <cluster_name>` to stop the cluster. To termina
   <summary>Prepare environment with conda</summary>
 
   ```bash
-  conda create -y -n llama-lora-multitool python=3.8
+  conda create -y python=3.8 -n llama-lora-multitool
   conda activate llama-lora-multitool
   ```
 </details>
 
 ```bash
-pip install -r requirements.txt
+pip install -r requirements.lock.txt
 python app.py --data_dir='./data' --base_model='decapoda-research/llama-7b-hf' --share
 ```