* Metal quantized modifications proposal.
- Add a device param, wherever needed.
- Create new QMetal storage thing that implements QuantizedType.
- Update everywhere needed.
Fix Python.
Fixing examples.
Fix: fmt + clippy + stub.
Moving everything around.
Only missing the actual implems.
Fixing everything + adding dequantized kernels.
More work.
Fixing matmul.
Fmt + Clippy
Some clippy fixes.
Working state.
Q2K Metal -> Bugged (also present in GGML).
Q4K CPU -> Bugged (present previously, new test catch it).
Q5K CPU -> Bugged (present previously).
Q8_1 Both -> Never really implemented it seems
Q8K metal -> Never implemented in metal
Fixing Q2K bug (present in ggml).
* Cleanup.
* Fix the rebase.
* Removing the fences speeds everything up and *is* correct this time...
* Cleanup the fence.
* After rebase.
* Bad code removal.
* Rebase after phi2 merge + fix replit default to CPU.
* Making the CI happy.
* More happy tests.
---------
Co-authored-by: Nicolas Patry <nicolas@Nicolass-MacBook-Pro.local>
* 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>
* Some first `Module` implementations
* Add `state_dict` and `load_state_dict` functionality
* Move modules around and create `candle.nn.Linear`
* Add `nn.Embedding` and `nn.LayerNorm`
* Add BERT implementation
* Batch q-matmul
* Automatically dequantize `QTensors` if a `Tensor` is expected
* Add Module `.to()`, `.cuda()`, `cpu()` and `.type()` functionality
* Unittests for `Module`, `Tensor` and `candle.utils`
* Add `pytorch` like slicing to `Tensor`
* Cleanup and BERT fixes
* `black` formatting + unit-test for `nn.Linear`
* Refactor slicing implementation
* Sketch a quantized llama using the pyo3 api.
* Add more ops.
* Expose a few more functions to use in the quantized model.
* Rope embeddings.
* Get the forward pass to work.
* Add more pyo3 support.
* Add some support for quantized tensors in pyo3.
* Add an arc layer on qmatmul.
* Add the quantized matmul.
* Quantization support.
* More quantization support.
* Test the python quantization.
* 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.
