Guillaume Lagrange

Issue initially raised in #2923 https://github.com/tracel-ai/burn/blob/e53978f35ddc08881e4b720504685694635cded1/crates/burn-ndarray/src/backend.rs#L11 https://github.com/tracel-ai/burn/blob/e53978f35ddc08881e4b720504685694635cded1/crates/burn-cubecl/src/backend.rs#L12 After some discussions, the RNG should probably be per-device. ```rust pub trait Backend { /// Device type. type Device: DeviceOps; /// Seed the...

Trying to import a pytorch model into its corresponding Burn implementation can be a little difficult for new users, especially if the definition involves multiple enum modules to match the...

### Discussed in https://github.com/tracel-ai/burn/discussions/2643 Originally posted by **pmalex** December 26, 2024 Is it possible to use these metrics in burn? I'm not found any examples so far.

Make is_contiguous check common

1

### Checklist - [ ] Confirmed that `run-checks all` script has been executed. ### Related Issues/PRs Following [this comment](https://github.com/tracel-ai/burn/pull/3077#discussion_r2060270547) in #3077 ### Changes Migrated `is_contiguous` to common (including minor fix)....

Following #2389, we should add a test to avoid any future regressions for this target.

Fix remainder int

6

### Checklist - [ ] Confirmed that `cargo run-checks` command has been executed. - [ ] Made sure the book is up to date with changes in this PR. ###...

Add `module.train()` to move a module back to the autodiff backend

1

### Checklist - [ ] Confirmed that `cargo run-checks` command has been executed. ### Related Issues/PRs From [discord](https://discord.com/channels/1038839012602941528/1059209073784008835/1434660520828600451): ``` Is there a convenient way to get the model on the...

### Feature description Boolean mask indexing, similar to: ```py inds = scores > self.conf_threshold boxes, landmarks, scores = boxes[inds], landmarks[inds], scores[inds] ``` From [discord](https://discord.com/channels/1038839012602941528/1038839013735399547/1413497627772719317)

Expose backend supported types

4

Brought up on discord. Right now we can switch the backend types easily, which is nice, but you need prior knowledge on the supported dtypes (e.g. f16 is supported on...

There is currently no generic way to check for available devices. This requires outside knowledge for each backend. Related: https://github.com/tracel-ai/cubecl/issues/720 #1883