DecisionTransformer

class torchrl.modules.DecisionTransformer(state_dim, action_dim, config: dict | DTConfig = None, device: torch.device | None = None)

Online Decion Transformer.

Desdescribed in https://arxiv.org/abs/2202.05607 .

The transformer utilizes a default config to create the GPT2 model if the user does not provide a specific config. default_config = { … “n_embd”: 256, … “n_layer”: 4, … “n_head”: 4, … “n_inner”: 1024, … “activation”: “relu”, … “n_positions”: 1024, … “resid_pdrop”: 0.1, … “attn_pdrop”: 0.1, }

Parameters:

• state_dim (int) – dimension of the state space

• action_dim (int) – dimension of the action space

• config (DTConfig or dict, optional) – transformer architecture configuration, used to create the GPT2Config from transformers. Defaults to default_config.

Example

>>> config = DecisionTransformer.default_config()
>>> config.n_embd = 128
>>> print(config)
DTConfig(n_embd: 128, n_layer: 4, n_head: 4, n_inner: 1024, activation: relu, n_positions: 1024, resid_pdrop: 0.1, attn_pdrop: 0.1)
>>> # alternatively
>>> config = DecisionTransformer.DTConfig(n_embd=128)
>>> model = DecisionTransformer(state_dim=4, action_dim=2, config=config)
>>> batch_size = [3, 32]
>>> length = 10
>>> observation = torch.randn(*batch_size, length, 4)
>>> action = torch.randn(*batch_size, length, 2)
>>> return_to_go = torch.randn(*batch_size, length, 1)
>>> output = model(observation, action, return_to_go)
>>> output.shape
torch.Size([3, 32, 10, 128])

class DTConfig(n_embd: Any = 256, n_layer: Any = 4, n_head: Any = 4, n_inner: Any = 1024, activation: Any = 'relu', n_positions: Any = 1024, resid_pdrop: Any = 0.1, attn_pdrop: Any = 0.1)

Default configuration for DecisionTransformer.

forward(observation: Tensor, action: Tensor, return_to_go: Tensor)

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.