Update documentation to reality

Signed-off-by: Ben Firshman <[email protected]>

Author: bfirsh

CONTRIBUTING.md

@@ -24,7 +24,7 @@ As much as possible, this is attempting to follow the [Standard Go Project Layou
 - `pkg/client/` - Client used by the CLI to communicate with the server.
 - `pkg/database/` - Used by the server to store metadata about models.
 - `pkg/docker/` - Various interfaces with Docker for building and running containers.
-- `pkg/model/` - Model, repo, and configs (`cog.yaml`).
+- `pkg/model/` - Versions, repos, and configs (`cog.yaml`).
 - `pkg/server/` - Server for storing data and building Docker images.
 - `pkg/serving/` - Runs inferences to test models.
 - `pkg/settings/` - Manages `.cog` directory in repos and `.config/cog/` directory for user settings.

README.md

@@ -40,18 +40,16 @@ environment:
   python_version: "3.8"
   python_requirements: "requirements.txt"
   system_packages:
-   - libgl1-mesa-glx
-   - libglib2.0-0
+    - libgl1-mesa-glx
+    - libglib2.0-0
 ```
 
 3. Push it to a repository and build it:
 
 ```
-$ cog build
+$ cog push
 --> Uploading '.' to repository http://10.1.2.3/colorization... done
---> Building CPU Docker image... done
---> Building GPU Docker image... done
---> Built model b6a2f8a2d2ff
+--> Created version b6a2f8a2d2ff
 ```
 
 This has:
@@ -72,7 +70,7 @@ $ cog infer b6a2f8a2d2ff -i @input.png -o @output.png
 It is also just a Docker image, so you can run it as an HTTP service wherever Docker runs:
 
 ```
-$ cog show b6a2f8a2d2ff 
+$ cog show b6a2f8a2d2ff
 ...
 Docker image (GPU):  registry.hooli.net/colorization:b6a2f8a2d2ff-gpu
 Docker image (CPU):  registry.hooli.net/colorization:b6a2f8a2d2ff-cpu

docs/getting-started.md

@@ -1,36 +1,46 @@
 # Getting started with Cog
 
-First step is to start a server. You'll need to point it at a Docker registry to store the Docker images For example, [create one on Google Cloud Container Registry](https://cloud.google.com/container-registry/docs/quickstart). 
+First step is to start a server. You'll need to point it at a Docker registry to store the Docker images For example, [create one on Google Cloud Container Registry](https://cloud.google.com/container-registry/docs/quickstart).
 
 Then, start the server, pointing at your Docker registry:
 
     cog server --port=8080 --docker-registry=gcr.io/some-project/cog
 
-
 Next, let's build a model. We have [some models you can play around with](https://github.com/replicate/cog-examples). Clone that repository (you'll need git-lfs) and hook up that directory to your Cog server:
 
     cd example-models/inst-colorization/
     cog repo set localhost:8080/examples/inst-colorization
 
-Then, let's build it:
+Take a look at `cog.yaml` and `model.py` to see what we're building.
+
+Then, let's push it:
+
+    cog push
 
-    cog build
+This has uploaded the currently directory to the server and the server has stored that as a version of your model. In the background, it is now building two Docker images (CPU and GPU variants) that will run your model. You can see the status of this build with `cog show`, replacing the ID with your model's ID:
 
-This will take a few minutes. In the meantime take a look at `cog.yaml` and `model.py` to see what we're building.
+    $ cog show b31f9f72d8f14f0eacc5452e85b05c957b9a8ed9
+    ...
+    CPU image: Building. Run `cog build logs 8f3a05c42ee5` to see its progress.
+    GPU image: Building. Run `cog build logs b087a0bb5b7a` to see its progress.
 
-When that has finished, you can run inferences on the built model from any machine that is pointed at the server. Replace the ID with your model's ID, and the file with an image on your disk you want to colorize:
+When the build has finished, you can run inferences on the built model from any machine that is pointed at the server. Replace the ID with your model's ID, and the file with an image on your disk you want to colorize:
 
     cog infer b31f9f72d8f14f0eacc5452e85b05c957b9a8ed9 -i @hotdog.jpg
 
 You can also list the models for this repo:
 
     cog list
 
-You can see more details about the model:
+## Deploying the model
+
+Cog builds Docker images for each version of your model. Those Docker images serve an HTTP inference API (to be documented).
+
+To get the Docker image, run this with the version you want to deploy:
 
     cog show b31f9f72d8f14f0eacc5452e85b05c957b9a8ed9
 
-In this output is the Docker image. You can run this anywhere a Docker image runs to deploy your model. For example:
+In this output is the name of the CPU or GPU Docker images, dependending on whether you are deploying on CPU or GPU. You can run these anywhere a Docker image runs. For example:
 
-    $ docker run -d -p 8000:8000 --gpus all registry.hooli.net/colorization:b6a2f8a2d2ff-gpu
-    $ curl http://localhost:8000/infer -F [email protected]
+    $ docker run -d -p 5000:5000 --gpus all registry.hooli.net/colorization:b6a2f8a2d2ff-gpu
+    $ curl http://localhost:5000/infer -F [email protected]