Few fixes.
Going back on remote metal-rs.
Reusing a single buffer (for now) to speed things up.
Adding some half kernels.
All tests are panicking instead of random failure.
Putting back f16 index select.
Add erf.
Working version for llama2-c.
Fixes + cache compute_pipeline_state.
BF16 metal fix.
Remove some prints.
new_owned -> new()..to_owned().
Better batched matmul.
Metal operational.
Reuse buffers on our own reference counts.
Tmp gemm.
Revert "Tmp gemm."
This reverts commit c65f68e988.
Interleave committing.
Speeding up copies using blit.
Fmt.
Fmt.
Remove the assert!
Fmt all.
Fixes after big rebase.
Add softmax for half and bfloat + tests
Fixing Llama example + accumulate softmax in float.
* Support the shape op in ONNX.
* Share the axis normalization bits.
* Add some limited support for gather.
* Unsqueeze.
* Comparison with broadcasting.
* Add Not + handle i32.
* Negative and `*args` shape handling
* Rename to `PyShapeWithHole` + validate that only one hole exists
* Regenerate stubs
---------
Co-authored-by: Laurent Mazare <laurent.mazare@gmail.com>
* Add some reinforcement learning example.
* Python initialization.
* Get the example to run.
* Vectorized gym envs for the atari wrappers.
* Get some simulation loop to run.
* Add some group parameter to convolutions.
* Avoid some unnecessary groups checks.
* Move the tensor convolution bits.
* Properh handling of groups.
* Bump the crate version.
* And add a changelog.
* Add a cudnn feature to be used for conv2d.
* Allocate the proper workspace.
* Only create a single cudnn handle per cuda device.
* Proper cudnn usage.
* Bugfix.
* Move the vision datasets to a separate crate.
* Move the batcher bits.
* Update the readme.
* Move the tiny-stories bits.
---------
Co-authored-by: Jane Doe <jane.doe@example.org>
- Loading with memmap
- Loading a sharded tensor
- Moved some snippets to `candle-examples/src/lib.rs` This is because
managing book specific dependencies is a pain https://github.com/rust-lang/mdBook/issues/706
- This causes a non aligned inclusion https://github.com/rust-lang/mdBook/pull/1856 which we have
to ignore fmt to remove.
mdbook might need some more love :)
* Rework the commands and run inference by default.
* Add the training module and load the training dataset.
* Random dataset iterator.
* Proper valid-loss computation.
* Compute the evaluation loss.
* Add more substance to the training loop.
* Add some flash-attn kernel, import the code for flash-attn v2 from Dao-AILab.
* More flash attn.
* Set up the flash attn parameters.
* Get things to compile locally.
* Move the flash attention files in a different directory.
* Build the static C library with nvcc.
* Add more flash attention.
* Update the build part.
* Better caching.
* Exclude flash attention from the default workspace.
* Put flash-attn behind a feature gate.
* Get the flash attn kernel to run.
* Move the flags to a more appropriate place.
* Enable flash attention in llama.
* Use flash attention in llama.
* Start adding llama2.c.
* Model loading.
* Add the llama-v2 model.
* Start converting the weights.
* Rotary embedding tweaks.
* Get the model to generate some tokens.
* Build example kernels.
* Add some sample custom kernel.
* Get the example kernel to compile.
* Add some cuda code.
* More cuda custom op.
* More cuda custom ops.
