0
Fork 0
mirror of https://github.com/zml/zml.git synced 2026-11-16 19:11:09 +00:00

Compare commits

..

31 commits

Author SHA1 Message Date
Corentin Godeau
82a23317ad zml/module.zig: increase SerializedExecutable size limit
On Neuron, some artifacts can reach more than 100MB.
2024-11-13 15:01:15 +00:00
Corentin Godeau
26637d52ce zml/nn: hack for sdpaMemEfficient 2024-11-13 11:58:14 +00:00
Corentin Godeau
fef181e0e8 zml/platform.zig: increase max device limit 2024-11-13 11:58:14 +00:00
Guillaume Wenzek
5b13dc8f42 add more scatterSlices testcases 2024-11-13 11:57:36 +00:00
Guillaume Wenzek
80d1c4bad1 zml: remove helpers.mapTensors in favor of zml.meta.map 2024-11-13 11:57:36 +00:00
Guillaume Wenzek
68e9c0c8e3 fix shape when for loop is fully unrolled 2024-11-13 11:57:36 +00:00
Guillaume Wenzek
d116292fa8 zml.nn: properly handle overflows in top_p
fixup! zml.nn: properly handle overflows in top_p
2024-11-13 11:57:36 +00:00
Guillaume Wenzek
64eedebc8b zml.Tensor.pad should accept i64 to be compatible with dims 2024-11-13 11:57:36 +00:00
Guillaume Wenzek
93276ff728 allow to have array of tensor inside model struct 2024-11-13 11:57:36 +00:00
Guillaume Wenzek
87f772c942 implement advanced sampling logic
fixup! implement advanced sampling logic
2024-11-13 11:57:36 +00:00
Guillaume Wenzek
74d8bee54f iota should only take one dtype
currently iota accept a dtype and a shape, making one wonder which dtype is used.
the `withDtype` allow to elegantly change dtype on the fly if needed
2024-11-13 11:57:36 +00:00
Guillaume Wenzek
cdee3fa0bd add DotAlgorim support to dot_general
for now it's not exposed to users because the API is a bit limited.
in particular only a subset of algorithm are supported by openxla,
and only a subset of that is supported by chips.

I was able to successfully change the accumulation type for dot on Cuda,
but want to experiment more before deciding what to expose.
2024-11-13 11:57:36 +00:00
Guillaume Wenzek
465a7a276e skip gather/scatter with batches tests 2024-11-13 11:57:36 +00:00
Guillaume Wenzek
1a474379ea pjrtx: error instead of panic 2024-11-13 11:57:36 +00:00
Steeve Morin
cfe4581f85 examples,runtimes: add layers for each runtime
Some runtimes require additional layers to be built. This change adds
a generic `layers` filegroup to each runtime. It should be embedded into
OCI images.
2024-11-08 17:49:55 +01:00
Steeve Morin
c748ee155e runtimes/neuron: rework neuronx-cc sandboxing
- use codegen for python environment setup
- rework use of runfiles to do that properly
- switch to the script version of the py_binary rule
2024-11-08 14:35:34 +01:00
Steeve Morin
7e88f7cee1 runtimes/neuron: add compiler env and flags
Notably:
- `--enable-fast-loading-neuron-binaries
- `--optlevel=1`: 30% speed up in compilation, no performance penalty on llama
- NEURON_RT_STOCHASTIC_ROUNDING_EN=1 for more precision
2024-11-08 14:35:34 +01:00
Steeve Morin
9901933eb3 runtimes/neuron: disable pyc precompiling
Given the time it takes to compile, I'm not sure it's worth it.
2024-11-08 14:35:34 +01:00
Corentin Godeau
4469db42c3 zml/module.zig: increase loaded module size 2024-11-08 10:59:35 +00:00
Steeve Morin
30716dd802 zml: rework device assignment when compiling the modul
First, do not copy `DeviceAssignmentProto` but use `@TypeOf` to fetch it.
Then, use a named block to set it inline.
2024-11-05 20:00:50 +01:00
Corentin Godeau
acedc63bb8 zml/module: disable donation for Neuron 2024-11-05 19:28:38 +01:00
Corentin Godeau
b311070dc5 runtimes/neuron: Set required env before Python initialization 2024-11-05 19:28:38 +01:00
Steeve Morin
bef2d3f009 runtimes/neuron: bump requirements 2024-11-05 19:28:38 +01:00
Steeve Morin
1b036259f2 +neuron 2024-11-05 19:28:38 +01:00
Steeve Morin
2110847c9a +neuron 2024-11-05 19:28:38 +01:00
Steeve Morin
c13a8fe278 runtimes/neuron: wchar handling 2024-11-05 19:28:38 +01:00
Steeve Morin
54873b8476 neurontmp 2024-11-05 19:28:35 +01:00
Steeve Morin
e60ebe064f NEURON 2024-11-05 19:26:39 +01:00
Steeve Morin
38e3bd2a04 neuron 2024-11-05 19:26:39 +01:00
Steeve Morin
3efc78d273 neuron 2024-11-05 19:26:39 +01:00
Steeve Morin
27b7d31b40 neuron 2024-11-05 19:26:39 +01:00
975 changed files with 46569 additions and 234774 deletions

1
.bazelignore Symbolic link

1
.bazelrc Symbolic link

1
.bazelversion Symbolic link

14
.gitignore vendored

1
.nvim.lua Symbolic link

1
.vscode Symbolic link

3212
MODULE.bazel.lock generated

File diff suppressed because one or more lines are too long

Some files were not shown because too many files have changed in this diff Show more