From f7bfd4ca9e7da34e4a670f09319c22102cfefee9 Mon Sep 17 00:00:00 2001
From: Charles Frye <charles@modal.com>
Date: Thu, 8 Aug 2024 13:52:53 -0400
Subject: [PATCH] fix small nits (#76)

* remove deprecated background commit flag

* move yaml import out of global scope

* rename huggingface secret to default
---
 README.md        | 2 +-
 src/common.py    | 2 +-
 src/inference.py | 2 +-
 src/train.py     | 2 --
 4 files changed, 3 insertions(+), 5 deletions(-)

diff --git a/README.md b/README.md
index e8e110fa..e6897ff4 100644
--- a/README.md
+++ b/README.md
@@ -36,7 +36,7 @@ Inference on the fine-tuned model displays conformity to the output structure (`
    1. Create a [Modal](https://modal.com/) account.
    2. Install `modal` in your current Python virtual environment (`pip install modal`)
    3. Set up a Modal token in your environment (`python3 -m modal setup`)
-   4. You need to have a [secret](https://modal.com/docs/guide/secrets#secrets) named `huggingface` in your workspace. You can [create a new secret](https://modal.com/secrets) with the HuggingFace template in your Modal dashboard, using the key from HuggingFace (in settings under API tokens) to populate `HF_TOKEN` and changing the name from `my-huggingface-secret` to `huggingface`.
+   4. You need to have a [secret](https://modal.com/docs/guide/secrets#secrets) named `my-huggingface-secret` in your workspace. You can [create a new secret](https://modal.com/secrets) with the HuggingFace template in your Modal dashboard, using the key from HuggingFace (in settings under API tokens) to populate `HF_TOKEN`.
    5. For some LLaMA models, you need to go to the Hugging Face page (e.g. [this page for LLaMA 3 8B](https://huggingface.co/meta-llama/Meta-Llama-3-8B)\_ and agree to their Terms and Conditions for access (granted instantly).
    6. If you want to use [Weights & Biases](https://wandb.ai) for logging, you need to have a secret named `wandb` in your workspace as well. You can also create it [from a template](https://modal.com/secrets). Training is hard enough without good logs, so we recommend you try it or look into `axolotl`'s integration with [MLFlow](https://mlflow.org/)!
    </details>
diff --git a/src/common.py b/src/common.py
index 71abb02a..d535c367 100644
--- a/src/common.py
+++ b/src/common.py
@@ -45,7 +45,7 @@
 app = modal.App(
     APP_NAME,
     secrets=[
-        modal.Secret.from_name("huggingface"),
+        modal.Secret.from_name("my-huggingface-secret"),
         modal.Secret.from_dict({"ALLOW_WANDB": os.environ.get("ALLOW_WANDB", "false")}),
         *([modal.Secret.from_name("wandb")] if ALLOW_WANDB else []),
     ],
diff --git a/src/inference.py b/src/inference.py
index 700a59ac..6e2aa0a0 100644
--- a/src/inference.py
+++ b/src/inference.py
@@ -1,6 +1,5 @@
 import os
 import time
-import yaml
 from pathlib import Path
 
 import modal
@@ -21,6 +20,7 @@
     from vllm.engine.async_llm_engine import AsyncLLMEngine
     from vllm.sampling_params import SamplingParams
     from vllm.utils import random_uuid
+    import yaml
 
 
 def get_model_path_from_run(path: Path) -> Path:
diff --git a/src/train.py b/src/train.py
index 3aa412b8..40fec5b3 100644
--- a/src/train.py
+++ b/src/train.py
@@ -23,7 +23,6 @@
     gpu=GPU_CONFIG,
     volumes=VOLUME_CONFIG,
     timeout=24 * HOURS,
-    _allow_background_volume_commits=True,
 )
 def train(run_folder: str, output_dir: str):
     import torch
@@ -48,7 +47,6 @@ def train(run_folder: str, output_dir: str):
     gpu=SINGLE_GPU_CONFIG,
     volumes=VOLUME_CONFIG,
     timeout=24 * HOURS,
-    _allow_background_volume_commits=True,
 )
 def preproc_data(run_folder: str):
     print("Preprocessing data.")