Skip to content

Commit

Permalink
Merge pull request #12 from instadeepai/fix/mkdocs-render-issue
Browse files Browse the repository at this point in the history
fix: mkdocs render issue of asterisk
  • Loading branch information
EdanToledo authored Feb 9, 2024
2 parents d23fb65 + bc47d3d commit 480ae0a
Showing 1 changed file with 1 addition and 1 deletion.
2 changes: 1 addition & 1 deletion README.md
Original file line number Diff line number Diff line change
Expand Up @@ -149,7 +149,7 @@ from CleanRLs DQN JAX example.
like Snake for our fully jitted examples.

## Vault 💾
Vault is an efficient mechanism for saving Flashbax buffers to persistent data storage, e.g. for use in offline reinforcement learning. Consider a Flashbax buffer which has experience data of dimensionality $(B, T, *E)$, where $B$ is a batch dimension (for the sake of recording independent trajectories synchronously), $T$ is a temporal/sequential dimension, and $*E$ indicates the one or more dimensions of the experience data itself. Since large quantities of data may be generated for a given environment, Vault extends the $T$ dimension to a virtually unconstrained degree by reading and writing slices of buffers along this temporal axis. In doing so, gigantic buffer stores can reside on disk, from which sub-buffers can be loaded into RAM/VRAM for efficient offline training. Vault has been tested with the item, flat, and trajectory buffers.
Vault is an efficient mechanism for saving Flashbax buffers to persistent data storage, e.g. for use in offline reinforcement learning. Consider a Flashbax buffer which has experience data of dimensionality $(B, T, \*E)$, where $B$ is a batch dimension (for the sake of recording independent trajectories synchronously), $T$ is a temporal/sequential dimension, and $\*E$ indicates the one or more dimensions of the experience data itself. Since large quantities of data may be generated for a given environment, Vault extends the $T$ dimension to a virtually unconstrained degree by reading and writing slices of buffers along this temporal axis. In doing so, gigantic buffer stores can reside on disk, from which sub-buffers can be loaded into RAM/VRAM for efficient offline training. Vault has been tested with the item, flat, and trajectory buffers.

For more information, see the demonstrative notebook: [![Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/instadeepai/flashbax/blob/main/examples/vault_demonstration.ipynb)

Expand Down

0 comments on commit 480ae0a

Please sign in to comment.