Mounts on images

06_gpu_and_ml/comfyui/comfyapp.py

@@ -72,6 +72,11 @@
     .run_commands(  # use comfy-cli to install the ComfyUI repo and its dependencies
         "comfy --skip-prompt install --nvidia"
     )
+    .add_local_file(
+        Path(__file__).parent / "workflow_api.json",
+        "/root/workflow_api.json",
+        copy=True,
+    )
 )
 # ### Downloading custom nodes
 # We'll use `comfy-cli` to download custom nodes, in this case the popular WAS Node Suite pack.
@@ -86,7 +91,7 @@
 # ### Downloading models
 
 # You can also use comfy-cli to download models, but for this example we'll download the Flux models directly from Hugging Face into a Modal Volume.
-# Then on container start, we'll mount our models into the ComfyUI models directory.
+# Then on container start, we'll mount our models Volume into the ComfyUI models directory.
 # This allows us to avoid re-downloading the models every time you rebuild your image.
 
 image = (
@@ -182,12 +187,6 @@ def ui():
     allow_concurrent_inputs=10,
     container_idle_timeout=300,
     gpu="A10G",
-    mounts=[
-        modal.Mount.from_local_file(
-            Path(__file__).parent / "workflow_api.json",
-            "/root/workflow_api.json",
-        ),
-    ],
     volumes={"/root/comfy/ComfyUI/models": vol},
 )
 class ComfyUI:

06_gpu_and_ml/dreambooth/diffusers_lora_finetune.py

@@ -421,14 +421,17 @@ class AppConfig(SharedConfig):
     guidance_scale: float = 6
 
 
-assets_path = Path(__file__).parent / "assets"
+image_with_assets = image.add_local_dir(
+    # Add local web assets to the image
+    Path(__file__).parent / "assets",
+    remote_path="/assets",
+)
 
 
 @app.function(
-    image=image,
+    image=image_with_assets,
     concurrency_limit=1,
     allow_concurrent_inputs=1000,
-    mounts=[modal.Mount.from_local_dir(assets_path, remote_path="/assets")],
 )
 @modal.asgi_app()
 def fastapi_app():

06_gpu_and_ml/hyperparameter-sweep/hp_sweep_gpt.py

@@ -49,7 +49,7 @@
 import logging as L
 import urllib.request
 from dataclasses import dataclass
-from pathlib import Path
+from pathlib import Path, PosixPath
 
 import modal
 from pydantic import BaseModel
@@ -74,7 +74,7 @@
 volume = modal.Volume.from_name(
     "example-hp-sweep-gpt-volume", create_if_missing=True
 )
-volume_path = Path("/vol/data")
+volume_path = PosixPath("/vol/data")
 model_filename = "nano_gpt_model.pt"
 best_model_filename = "best_nano_gpt_model.pt"
 tb_log_path = volume_path / "tb_logs"
@@ -92,24 +92,22 @@
     "torch==2.1.2",
     "tensorboard==2.17.1",
     "numpy<2",
-)
+).add_local_dir(Path(__file__).parent / "src", remote_path="/root/src")
 
 # We also have some local dependencies that we'll need to import into the remote environment.
 # We mount them onto the remote container.
 
-mounts = [
-    modal.Mount.from_local_dir(
-        Path(__file__).parent / "src", remote_path=Path("/root/src")
-    )
-]
-
 # We'll serve a simple web endpoint
 web_image = base_image.pip_install(
     "fastapi[standard]==0.115.4", "starlette==0.41.2"
 )
 
 # And we'll deploy a web UI for interacting with our trained models using Gradio.
-ui_image = web_image.pip_install("gradio~=4.44.0")
+assets_path = Path(__file__).parent / "assets"
+ui_image = web_image.pip_install("gradio~=4.44.0").add_local_dir(
+    assets_path, remote_path="/assets"
+)
+
 
 # We can also "pre-import" libraries that will be used by the functions we run on Modal in a given image
 # using the `with image.imports` context manager.
@@ -139,7 +137,6 @@
 
 @app.function(
     image=torch_image,
-    mounts=mounts,
     volumes={volume_path: volume},
     gpu=gpu,
     timeout=1 * HOURS,
@@ -575,15 +572,13 @@ def web_generate(request: GenerationRequest):
 # The Gradio UI will look something like this:
 
 # ![Image of Gradio Web App. Top shows model selection dropdown. Left side shows input prompt textbox. Right side shows SLM generated output. Bottom has button for starting generation process](./gradio.png)
-assets_path = Path(__file__).parent / "assets"
 
 
 @app.function(
     image=ui_image,
     concurrency_limit=1,
     volumes={volume_path: volume},
     allow_concurrent_inputs=1000,
-    mounts=[modal.Mount.from_local_dir(assets_path, remote_path="/assets")],
 )
 @modal.asgi_app()
 def ui():

06_gpu_and_ml/obj_detection_webcam/webcam.py

@@ -168,8 +168,9 @@ def detect(self, img_data_in):
 
 
 @app.function(
-    image=modal.Image.debian_slim().pip_install("fastapi[standard]"),
-    mounts=[modal.Mount.from_local_dir(static_path, remote_path="/assets")],
+    image=modal.Image.debian_slim()
+    .pip_install("fastapi[standard]")
+    .add_local_dir(static_path, remote_path="/assets")
 )
 @modal.asgi_app(label="example-webcam-object-detection")
 def fastapi_app():

06_gpu_and_ml/openai_whisper/pod_transcriber/app/main.py

@@ -93,9 +93,7 @@ def populate_podcast_metadata(podcast_id: str):
 
 
 @app.function(
-    mounts=[
-        modal.Mount.from_local_dir(config.ASSETS_PATH, remote_path="/assets")
-    ],
+    image=app_image.add_local_dir(config.ASSETS_PATH, remote_path="/assets"),
     network_file_systems={config.CACHE_DIR: volume},
     keep_warm=2,
 )
@@ -112,9 +110,7 @@ def fastapi_app():
     return web_app
 
 
-@app.function(
-    image=app_image,
-)
+@app.function()
 def search_podcast(name):
     from gql import gql
 

06_gpu_and_ml/stable_diffusion/text_to_image.py

@@ -221,14 +221,15 @@ def entrypoint(
 
 frontend_path = Path(__file__).parent / "frontend"
 
-web_image = modal.Image.debian_slim(python_version="3.12").pip_install(
-    "jinja2==3.1.4", "fastapi[standard]==0.115.4"
+web_image = (
+    modal.Image.debian_slim(python_version="3.12")
+    .pip_install("jinja2==3.1.4", "fastapi[standard]==0.115.4")
+    .add_local_dir(frontend_path, remote_path="/assets")
 )
 
 
 @app.function(
     image=web_image,
-    mounts=[modal.Mount.from_local_dir(frontend_path, remote_path="/assets")],
     allow_concurrent_inputs=1000,
 )
 @modal.asgi_app()

06_gpu_and_ml/text-to-pokemon/text_to_pokemon/config.py

@@ -206,5 +206,6 @@ def null_safety(images, **kwargs):
         "fastapi[standard]",
     )
     .run_function(load_stable_diffusion_pokemon_model)
+    .add_local_dir(local_path=ASSETS_PATH, remote_path="/assets")
 )
 app = modal.App(name="example-text-to-pokemon", image=image)

06_gpu_and_ml/text-to-pokemon/text_to_pokemon/main.py

@@ -129,13 +129,7 @@ def diskcached_text_to_pokemon(prompt: str) -> list[bytes]:
     return samples_data
 
 
-@app.function(
-    mounts=[
-        modal.Mount.from_local_dir(
-            local_path=config.ASSETS_PATH, remote_path="/assets"
-        )
-    ],
-)
+@app.function()
 @modal.asgi_app()
 def fastapi_app():
     import fastapi.staticfiles

07_web_endpoints/fasthtml-checkboxes/fasthtml_checkboxes.py

@@ -30,17 +30,14 @@
 db = modal.Dict.from_name("example-checkboxes-db", create_if_missing=True)
 
 css_path_local = Path(__file__).parent / "styles.css"
-css_path_remote = Path("/assets/styles.css")
+css_path_remote = "/assets/styles.css"
 
 
 @app.function(
-    image=modal.Image.debian_slim(python_version="3.12").pip_install(
-        "python-fasthtml==0.6.9", "inflect~=7.4.0"
-    ),
+    image=modal.Image.debian_slim(python_version="3.12")
+    .pip_install("python-fasthtml==0.6.9", "inflect~=7.4.0")
+    .add_local_file(css_path_local, remote_path=css_path_remote),
     concurrency_limit=1,  # we currently maintain state in memory, so we restrict the server to one worker
-    mounts=[
-        modal.Mount.from_local_file(css_path_local, remote_path=css_path_remote)
-    ],
     allow_concurrent_inputs=1000,
 )
 @modal.asgi_app()

09_job_queues/doc_ocr_webapp.py

@@ -75,28 +75,31 @@ async def poll_results(call_id: str):
     return result
 
 
-# Finally, we mount the static files for our front-end. We've made [a simple React
+# Specify a container image containing the version of fastapi we want to use to serve
+# our web app
+fast_api_image = modal.Image.debian_slim(python_version="3.12").pip_install(
+    "fastapi[standard]==0.115.4"
+)
+
+# Finally, we add the static files for our front-end. We've made [a simple React
 # app](https://github.com/modal-labs/modal-examples/tree/main/09_job_queues/doc_ocr_frontend)
-# that hits the two endpoints defined above. To package these files with our app, first
-# we get the local assets path, and then create a modal [`Mount`](https://modal.com/docs/guide/local-data#mounting-directories)
-# that mounts this directory at `/assets` inside our container. Then, we instruct FastAPI to [serve
+# that hits the two endpoints defined above. To package these files with our app, we use
+# add_local_dir with the local directory of the assets, and specify that we want them
+# in the `/assets` directory inside our container (the `remote_path`). Then, we instruct FastAPI to [serve
 # this static file directory](https://fastapi.tiangolo.com/tutorial/static-files/) at our root path.
 
-assets_path = Path(__file__).parent / "doc_ocr_frontend"
+local_assets_path = Path(__file__).parent / "doc_ocr_frontend"
+ocr_app_image = fast_api_image.add_local_dir(
+    local_assets_path, remote_path="/assets"
+)
 
 
-@app.function(
-    image=modal.Image.debian_slim(python_version="3.12").pip_install(
-        "fastapi[standard]==0.115.4"
-    ),
-    mounts=[modal.Mount.from_local_dir(assets_path, remote_path="/assets")],
-)
+@app.function(image=ocr_app_image)
 @modal.asgi_app()
 def wrapper():
     web_app.mount(
         "/", fastapi.staticfiles.StaticFiles(directory="/assets", html=True)
     )
-
     return web_app
 
 

10_integrations/dbt/dbt_duckdb.py

@@ -35,6 +35,12 @@
 PROJ_PATH = "/root/dbt"  # remote paths
 PROFILES_PATH = "/root/dbt_profile"
 TARGET_PATH = "/root/target"
+# Most of the DBT code and configuration is taken directly from the classic
+# [Jaffle Shop](https://github.com/dbt-labs/jaffle_shop) demo and modified to support
+# using `dbt-duckdb` with an S3 bucket.
+
+# The DBT `profiles.yml` configuration is taken from
+# [the `dbt-duckdb` docs](https://github.com/jwills/dbt-duckdb#configuring-your-profile).
 
 # We also define the environment our application will run in --
 # a container image, as in Docker.
@@ -56,27 +62,17 @@
             "DBT_TARGET_PATH": TARGET_PATH,
         }
     )
+    # Here we add all local code and configuration into the Modal Image
+    # so that it will be available when we run DBT on Modal.
+    .add_local_dir(LOCAL_DBT_PROJECT, remote_path=PROJ_PATH)
+    .add_local_file(
+        LOCAL_DBT_PROJECT / "profiles.yml",
+        remote_path=f"{PROFILES_PATH}/profiles.yml",
+    )
 )
 
 app = modal.App(name="example-dbt-duckdb-s3", image=dbt_image)
 
-# Most of the DBT code and configuration is taken directly from the classic
-# [Jaffle Shop](https://github.com/dbt-labs/jaffle_shop) demo and modified to support
-# using `dbt-duckdb` with an S3 bucket.
-
-# The DBT `profiles.yml` configuration is taken from
-# [the `dbt-duckdb` docs](https://github.com/jwills/dbt-duckdb#configuring-your-profile).
-
-# Here we mount all this local code and configuration into the Modal Function
-# so that it will be available when we run DBT on Modal.
-
-dbt_project = modal.Mount.from_local_dir(
-    LOCAL_DBT_PROJECT, remote_path=PROJ_PATH
-)
-dbt_profiles = modal.Mount.from_local_file(
-    local_path=LOCAL_DBT_PROJECT / "profiles.yml",
-    remote_path=Path(PROFILES_PATH, "profiles.yml"),
-)
 dbt_target = modal.Volume.from_name("dbt-target-vol", create_if_missing=True)
 
 # We'll also need to authenticate with AWS to store data in S3.
@@ -133,7 +129,6 @@
 
 
 @app.function(
-    mounts=[dbt_project],
     secrets=[s3_secret],
 )
 def create_source_data():
@@ -168,15 +163,14 @@ def create_source_data():
 # And DBT is a Python tool, so it's easy to run DBT with Modal:
 # below, we import the `dbt` library's `dbtRunner` to pass commands from our
 # Python code, running on Modal, the same way we'd pass commands on a command line.
-
-# Note that this Modal Function has access to our AWS Secret,
-# the `mount`ed local files with our DBT project and profiles,
+#
+# Note that this Modal Function has access to our AWS S3 Secret,
+# the local files associated with our DBT project and profiles,
 # and a remote Modal Volume that acts as a distributed file system.
 
 
 @app.function(
     secrets=[s3_secret],
-    mounts=[dbt_project, dbt_profiles],
     volumes={TARGET_PATH: dbt_target},
 )
 def run(command: str) -> None:
@@ -279,7 +273,6 @@ def serve_dbt_docs():
 @app.function(
     schedule=modal.Period(days=1),
     secrets=[s3_secret],
-    mounts=[dbt_project, dbt_profiles],
     volumes={TARGET_PATH: dbt_target},
 )
 def daily_build() -> None: