Change all is_inside to run_inside (#475)

* Change all is_inside to run_inside

* use images not assigned to the stub in some cases

02_building_containers/import_sklearn.py

@@ -18,12 +18,11 @@
     .pip_install("scikit-learn"),
 )
 
-# The `stub.is_inside()` lets us conditionally run code in the global scope.
+# The `stub.image.run_inside()` lets us conditionally run code in the global scope.
 # This is needed because we might not have sklearn and numpy installed locally,
-# but we know they are installed inside the custom image. `stub.is_inside()`
-# will return `False` when we run this locally, but `True` when it runs in the cloud.
+# but we know they are installed inside the custom image.
 
-if stub.is_inside():
+with stub.image.run_inside():
     import numpy as np
     from sklearn import datasets, linear_model
 

03_scaling_out/youtube_face_detection.py

@@ -52,7 +52,7 @@
 )
 stub = modal.Stub("example-youtube-face-detection", image=image)
 
-if stub.is_inside():
+with stub.image.run_inside():
     import cv2
     import moviepy.editor
     import pytube

06_gpu_and_ml/blender/blender_video.py

@@ -66,10 +66,9 @@
 #
 # We need various global configuration that we want to happen inside the containers (but not locally), such as
 # enabling the GPU device.
-# To do this, we use the `stub.is_inside()` conditional, which will evaluate to `False` when the script runs
-# locally, but to `True` when imported in the cloud.
+# To do this, we use the `stub.image.run.inside()` context manager.
 
-if stub.is_inside():
+with stub.image.run_inside():
     import bpy
 
     # NOTE: Blender segfaults if you try to do this after the other imports.

06_gpu_and_ml/stable_lm/main.py

@@ -215,7 +215,7 @@ def format_prompt(instruction: str) -> str:
     return f"<|USER|>{instruction}<|ASSISTANT|>"
 
 
-if stub.is_inside():
+with stub.image.run_inside():
     import uuid
 
     import msgspec

07_web_endpoints/chatbot_spa.py

@@ -40,15 +40,15 @@ def load_tokenizer_and_model():
     return tokenizer, model
 
 
-stub.gpu_image = (
+gpu_image = (
     Image.debian_slim()
     .pip_install("torch", find_links="https://download.pytorch.org/whl/cu116")
     .pip_install("transformers~=4.31", "accelerate")
     .run_function(load_tokenizer_and_model)
 )
 
 
-if stub.is_inside(stub.gpu_image):
+with gpu_image.run_inside():
     import torch
 
     tokenizer, model = load_tokenizer_and_model()
@@ -72,7 +72,7 @@ def chat(body: dict = fastapi.Body(...)):
     return app
 
 
-@stub.function(gpu="any", image=stub.gpu_image)
+@stub.function(gpu="any", image=gpu_image)
 def generate_response(
     message: str, id: Optional[str] = None
 ) -> Tuple[str, str]:

misc/news_summarizer.py

@@ -44,20 +44,20 @@ def fetch_model(local_files_only: bool = False):
     return model, tokenizer
 
 
-stub["deep_learning_image"] = (
+deep_learning_image = (
     modal.Image.debian_slim()
     .pip_install("transformers==4.16.2", "torch", "sentencepiece")
     .run_function(fetch_model)
 )
 
 # Defining the scraping image is very similar. This image only contains the packages required
 # to scrape the New York Times website, though; so it's much smaller.
-stub["scraping_image"] = modal.Image.debian_slim().pip_install(
+scraping_image = modal.Image.debian_slim().pip_install(
     "requests", "beautifulsoup4", "lxml"
 )
 
 
-if stub.is_inside(stub["scraping_image"]):
+with scraping_image.run_inside():
     import requests
     from bs4 import BeautifulSoup
 
@@ -83,9 +83,7 @@ class NYArticle:
 # Create an environment variable called `NYTIMES_API_KEY` with your API key.
 
 
-@stub.function(
-    secret=modal.Secret.from_name("nytimes"), image=stub["scraping_image"]
-)
+@stub.function(secret=modal.Secret.from_name("nytimes"), image=scraping_image)
 def latest_science_stories(n_stories: int = 5) -> List[NYArticle]:
     # query api for latest science articles
     params = {
@@ -120,7 +118,7 @@ def latest_science_stories(n_stories: int = 5) -> List[NYArticle]:
 # [Beautiful Soup](https://www.crummy.com/software/BeautifulSoup/bs4/doc/) for that.
 
 
-@stub.function(image=stub["scraping_image"])
+@stub.function(image=scraping_image)
 def scrape_nyc_article(url: str) -> str:
     print(f"Scraping article => {url}")
 
@@ -150,7 +148,7 @@ def scrape_nyc_article(url: str) -> str:
 
 
 @stub.function(
-    image=stub["deep_learning_image"],
+    image=deep_learning_image,
     gpu=False,
     memory=4096,
 )