Update readme

Remove examples as redundant with tutorial
Add link to API docs

README.md

@@ -2,99 +2,19 @@
 _GLoW, but very light_
 
 This repository is the barebones API to build a global workspace.
-It contains:
-- the GlobalWorkspace module (using LightningModule) in two flavors: deterministic or variational;
-- an interface for your expert domains that are connected to the GlobalWorkspace;
-- a generic and barebone VAE class;
-
 
 ## Install
-See [Installation instructions](docs/installation.md)
+See [Installation instructions](docs/installation.md).
 
 ## Examples
 You can find code examples in the `examples` folder. Each subfolder is a different
 example.
 
-You can also follow a tutorial on the basics of shimmer [here](docs/shimmer_basics.md)
-
-## Make a domain
-
-```python
-from typing import Any
-
-import torch
-import torch.nn.functional as F
-from shimmer import DomainModule, LossOutput
-
-
-class MyDomain(DomainModule):
-    def __init__(self, latent_dim: int):
-        super().__init__(latent_dim)
-
-    def encode(self, x: Any) -> torch.Tensor:
-        # encode the input x into a latent representation
-        # provided to the GW
-        ...
-
-    def decode(self, z: torch.Tensor) -> Any:
-        # decode the latent representation back into the input form
-        ...
-
-    def compute_loss(self, pred: torch.Tensor, target: torch.Tensor) -> LossOutput:
-        return LossOutput(
-            loss=F.mse_loss(pred, target),  # loss used for training
-            metrics={},  # additional metrics to log (not required)
-        )
-```
-
-## Use a GW
-
-### GWInterface
-To link each domain module with the global workspace, we need a GWInterface.
-It encodes the unimodal representations into a GW representation, or decodes a
-GW representation into a unimodal representation.
-
-```python
-from shimmer import GWInterface
-
-my_domain = MyDomain(latent_dim=32)
-my_domain_gw_interface = GWInterface(
-    my_domain,
-    workspace_dim=12,  # latent dim of the global workspace
-    encoder_hidden_dim=32,  # hidden dimension for the GW encoder
-    encoder_n_layers=3,  # n layers to use for the GW encoder
-    decoder_hidden_dim=32,  # hidden dimension for the GW decoder
-    decoder_n_layers=3,  # n layers to use for the GW decoder
-)
-```
-
-
-### GW
-To load a global workspace, use: 
-```python
-from shimmer import GlobalWorkspace, LossCoefs
-
-domain_modules = {"my_domain": my_domain}
-
-gw_interfaces = {"my_domain": my_domain_gw_interface}
-
-workspace_dim = 32
-
-loss_coefs: LossCoefs = {
-    "translations": 1.0,
-    "demi_cycles": 0.0,
-    "cycles": 1.0,
-    "contrastives": 0.1,
-}
-
-model = GlobalWorkspace(
-    domain_modules,
-    gw_interfaces,
-    workspace_dim,
-    loss_coefs,
-)
-```
+You can also follow a tutorial on the basics of shimmer [here](docs/shimmer_basics.md).
 
+## API Documentation
+The full API documentation is available here:
+[https://bdvllrs.github.io/shimmer/](https://bdvllrs.github.io/shimmer/).
 
 # Contributing
 See [CONTRIBUTING.md](CONTRIBUTING.md).