Commit Graph

28 Commits

Author SHA1 Message Date
718671a0d5 Use BufferOffset in metal backend ops. (#2029)
* Use BufferOffset in the metal backend.

* More BufferOffset usage.

* Use in where-cond.
2024-04-08 09:37:25 +02:00
c5fe4a7f89 Rework the buffer offset logic for metal kernels (#2028)
* Move the metal kernels utils in a separate module.

* Use the BufferOffset for unary ops.

* Fix clippy lints.

* Use the new BufferOffset.

* Adapt the binary ops.

* Affine.

* More ops (powf, elu, cast).
2024-04-07 22:37:53 +02:00
fee33b45c2 Add support for strided index-select on Metal (#1909)
* initial implementation

* use correct index, but still not breaking like it should have...

* fix test
2024-03-22 07:30:02 +01:00
2a8679509e Add support for conv_transpose1d for metal backend (#1874)
* first attempt

* progress

* integrate into metal backend

* finish and get test passing

* add other dtype support

* update transpose1d dtypes supported
2024-03-19 08:46:58 +01:00
04a61a9c72 Add avg_pool2d metal implementation for the metal backend (#1869)
* implement metal avg pool 2d

* fixX

* add suggested precision workaround for the accumulator
2024-03-18 18:50:14 +01:00
754fa1e813 Add support for max_pool2d for Metal backend (#1863)
* first pass at implementation of maxpool2d

* Add definitions for other dtypes

* add tests for other dtypes

* Cosmetic tweaks + re-enable maxpool2d tests for metal.

---------

Co-authored-by: Laurent <laurent.mazare@gmail.com>
2024-03-18 08:33:30 +01:00
184105792f add test for index add and add missing match statements (#1862) 2024-03-17 22:19:12 +01:00
e316cb6997 add support for casting between all datatypes (#1860) 2024-03-17 20:55:11 +01:00
db8b24ae92 Add support for index u8/i64 and input f16/bf16 scatter-add on metal (#1849)
* add support and tests for scatter add on metal

* add support for all datatypes
2024-03-17 08:09:43 +01:00
b60064780d feat: add silu activation function (#1706)
* feat: add silu activation function

* use silu/arg in grad

* update candle-nn

* use node
2024-02-14 10:27:22 +01:00
db923517b3 Merge branch 'main' into ivarflakstad/metal-prng 2024-01-17 18:03:57 +01:00
403680f17d Quantized GGUF style (#1523)
* 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>
2024-01-17 10:27:58 +01:00
79478ff5a1 Seed should be updated by random kernel result. 2024-01-15 11:58:25 +01:00
e06e8d0dbe fmt 2024-01-12 07:26:42 +01:00
e63bb8661b Merge branch 'main' into ivarflakstad/metal-prng 2024-01-12 07:19:58 +01:00
402349d120 feat(bf16): add cast support + tests for cast + bin ops (#1524) 2024-01-11 15:49:13 +01:00
6bf52b9fdf Gaussian normal distribution of PRNG via Box-Muller transform 2024-01-07 11:39:46 +01:00
955e63c803 Implement hybrid Tausworthe + LCG psuedo random number generator in metal 2024-01-05 13:27:59 +01:00
10d94659c3 Adding the convolutions (1d + 2d) to candle on metal. 2023-12-21 10:39:24 +01:00
03641293ee Clippy pass. 2023-12-18 15:22:43 +01:00
972903021c Finish reduce kernels. 2023-12-17 19:07:00 +01:00
6bc92e63cb Addressing a lot of comments. 2023-12-15 13:06:04 +01:00
4eeaf205d6 Fix softmax for long sequences (missing barrier). 2023-12-14 19:37:03 +01:00
931432ed55 Fixing tests + matmul from MFA 2023-12-13 16:58:36 +01:00
6e25822d4f Fix gelu for large x 2023-12-06 09:59:44 -05:00
2ca086939f Put back affine strided tests 2023-11-30 11:40:39 +01:00
4349ff1fc2 Starting to fix some tests.
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.
2023-11-30 11:30:31 +01:00
60f624a902 Moving tests around. 2023-11-20 16:17:19 +01:00