Skip to content

Commit f15faf1

Browse files
nSircombenSircombe
authored
Merge pull request #392 from nSircombe/feature/r25.10_updates
Updates CHANGELOGS for 25.10 release
2 parents 4e5d2ac + ff22ffa commit f15faf1

File tree

2 files changed

+77
-52
lines changed

2 files changed

+77
-52
lines changed

‎ML-Frameworks/pytorch-aarch64/CHANGELOG.md‎

Lines changed: 64 additions & 50 deletions
Original file line numberDiff line numberDiff line change
@@ -12,80 +12,94 @@ where `YY` is the year, and `MM` the month of the increment.
1212
### Changed
1313

1414
### Removed
15-
- Delete unused submodules of PyTorch's third-party modules.
16-
- Removed all references to tcmalloc as the default build now uses mimalloc.
15+
16+
### Fixed
17+
18+
## [r25.10] 2025-10-15
19+
https://github.com/ARM-software/Tool-Solutions/tree/r25.10
20+
21+
### Added
22+
- Adds PyTorch [PR](https://github.com/pytorch/pytorch/pull/164741) to enable mimalloc by default on AArch64 build:
23+
- Yields ~28% and ~20% mean improvement for fp32 and bf16 respectively.
24+
25+
### Changed
26+
27+
### Removed
28+
- Completely removes unnecessary python packages, including setuptools, from PyTorch Docker image.
29+
- Delete unused submodules of PyTorch's third-party modules.
30+
- Removed all references to tcmalloc as the default build now uses mimalloc.
1731

1832
### Fixed
1933

2034
## [r25.09] 2025-09-29
2135
https://github.com/ARM-software/Tool-Solutions/tree/r25.09
2236

2337
### Added
24-
- Adds `SECURITY.md` to built image.
38+
- Adds `SECURITY.md` to built image.
2539

2640
### Changed
27-
- Updates ACL URL from [ML Platform](https://review.mlplatform.org/ml/ComputeLibrary) to [GitHub](https://github.com/ARM-software/ComputeLibrary.git).
28-
- Updates hashes for:
29-
- PYTORCH_HASH to 45d9dcccc5bfb46a9eaca037270278bc3d7c95ea, 2.10.0.dev20250923 from viable/strict, September 23rd.
30-
- IDEEP_HASH to fd11055f4800ac89291e30b5387a79a1e6496aa6, from ideep_pytorch, September 10th.
31-
- ONEDNN_HASH to 9e8f619477469ed75d323d4915bf7a2513f01713, from main, September 23rd.
32-
- ACL_HASH to 531a4968cecb7b4fc0a3b65482e2c524289e087e, from main, September 23rd.
33-
- TORCH_AO_HASH to 8e2ca35ea603349e71c2467e10fd371e34bf52bc, from main, September 23rd.
34-
- KLEIDIAI_HASH to bd2e6ae060014035e25bf4986be682762c446c2d, v1.14 from main.
35-
- Update torchvision from 0.23.0 to a nightly build, 0.25.0.dev20250923.
36-
- Change of flag name in `./build.sh` from `--force` to `--fresh`
37-
- Add `intx_packing_format="opaque_aten_kleidiai"` to `Int8DynamicActivationIntxWeightConfig` due to torchao API change
38-
- Updates `dockerize.sh` build to use BuildKit (to add extra build contexts).
41+
- Updates ACL URL from [ML Platform](https://review.mlplatform.org/ml/ComputeLibrary) to [GitHub](https://github.com/ARM-software/ComputeLibrary.git).
42+
- Updates hashes for:
43+
- PYTORCH_HASH to 45d9dcccc5bfb46a9eaca037270278bc3d7c95ea, 2.10.0.dev20250923 from viable/strict, September 23rd.
44+
- IDEEP_HASH to fd11055f4800ac89291e30b5387a79a1e6496aa6, from ideep_pytorch, September 10th.
45+
- ONEDNN_HASH to 9e8f619477469ed75d323d4915bf7a2513f01713, from main, September 23rd.
46+
- ACL_HASH to 531a4968cecb7b4fc0a3b65482e2c524289e087e, from main, September 23rd.
47+
- TORCH_AO_HASH to 8e2ca35ea603349e71c2467e10fd371e34bf52bc, from main, September 23rd.
48+
- KLEIDIAI_HASH to bd2e6ae060014035e25bf4986be682762c446c2d, v1.14 from main.
49+
- Update torchvision from 0.23.0 to a nightly build, 0.25.0.dev20250923.
50+
- Change of flag name in `./build.sh` from `--force` to `--fresh`
51+
- Add `intx_packing_format="opaque_aten_kleidiai"` to `Int8DynamicActivationIntxWeightConfig` due to torchao API change
52+
- Updates `dockerize.sh` build to use BuildKit (to add extra build contexts).
3953

4054
### Removed
41-
- Removes WIP ComputeLibrary patch https://review.mlplatform.org/c/ml/ComputeLibrary/+/12818/1.
42-
- Removed unused dependencies:
43-
- Delete unused submodules to avoid fetching.
44-
- Replace google-perftool bundle with specific libtcmalloc-minimal4 and add symlink so that you can still use it with `LD_PRELOAD=/usr/lib/aarch64-linux-gnu/libtcmalloc.so.4`.
45-
- Remove unused git apt package.
46-
- Remove outdated versions of pip/setuptools/wheel from image after pip updates itself.
47-
- Remove manual install of transformers in `transformers_llm_text_gen.py`, we now use a single version everywhere.
48-
- Remove unused pip packages: datasets, omegaconf, psutil, tiktoken, torchaudio, torchdata and torchtune.
55+
- Removes WIP ComputeLibrary patch https://review.mlplatform.org/c/ml/ComputeLibrary/+/12818/1.
56+
- Removed unused dependencies:
57+
- Delete unused submodules to avoid fetching.
58+
- Replace google-perftool bundle with specific libtcmalloc-minimal4 and add symlink so that you can still use it with `LD_PRELOAD=/usr/lib/aarch64-linux-gnu/libtcmalloc.so.4`.
59+
- Remove unused git apt package.
60+
- Remove outdated versions of pip/setuptools/wheel from image after pip updates itself.
61+
- Remove manual install of transformers in `transformers_llm_text_gen.py`, we now use a single version everywhere.
62+
- Remove unused pip packages: datasets, omegaconf, psutil, tiktoken, torchaudio, torchdata and torchtune.
4963

5064
### Fixed
5165

5266
## [r25.08] 2025-08-28
5367
https://github.com/ARM-software/Tool-Solutions/tree/r25.08
5468

5569
### Added
56-
- Adds https://github.com/pytorch/pytorch/pull/159859, a WIP LUT implmentation of bf16 GELU
57-
this gives an ~8x speedup on GELU and an ~1.8x speedup for attention for llama3.2 11B Vision (both on 16 threads).
58-
- Adds https://github.com/pytorch/pytorch/pull/158250, to integrate INT4->BF16 via KleidiAI, with fallback.
59-
- Adds https://github.com/pytorch/pytorch/pull/160080, a VLA PoC for PyTorch, and
60-
https://github.com/pytorch/pytorch/pull/161049, an optimised SVE exp_u20 implementation,
61-
Note: there may be some regressions on Neoverse-V1 with this WIP patch.
62-
- Adds a new example script llama_vision_instruct.py to run and benchmark
63-
Llama-3.2-11B-Vision-Instruct using text + image input and text output.
70+
- Adds https://github.com/pytorch/pytorch/pull/159859, a WIP LUT implmentation of bf16 GELU
71+
this gives an ~8x speedup on GELU and an ~1.8x speedup for attention for llama3.2 11B Vision (both on 16 threads).
72+
- Adds https://github.com/pytorch/pytorch/pull/158250, to integrate INT4->BF16 via KleidiAI, with fallback.
73+
- Adds https://github.com/pytorch/pytorch/pull/160080, a VLA PoC for PyTorch, and
74+
https://github.com/pytorch/pytorch/pull/161049, an optimised SVE exp_u20 implementation,
75+
Note: there may be some regressions on Neoverse-V1 with this WIP patch.
76+
- Adds a new example script llama_vision_instruct.py to run and benchmark
77+
Llama-3.2-11B-Vision-Instruct using text + image input and text output.
6478

6579
### Changed
66-
- Updates hashes for:
67-
- PYTORCH_HASH to 4e2ddb5db67617f9f5309c8bba0c17adc84cadbc, 2.9.0.dev20250808 from viable/strict, August 8th.
68-
- IDEEP_HASH to 3527b0bf2127aa2de93810feb6906d173c24037f, from ideep_pytorch, August 1st.
69-
- ONEDNN_HASH to 7e85b94b5f6be27b83c5435603ab67888b99da32, from main, August 1st.
70-
- ACL_HASH to 3c32d706d0245dcb55181c8ced526eab05e2ff8d, from main, August 1st.
71-
- TORCH_AO_HASH to 8d4a5d83d7be4d7807feabe38d37704c92d40900, from main, August 1st.
72-
- KLEIDIAI_HASH to 8ca226712975f24f13f71d04cda039a0ee9f9e2f, v1.12 from main.
73-
- Update the examples/transformers_llm_text_gen.py to use the new quantizer api Int8DynamicActivationIntxWeightConfig.
74-
- Deleted torchchat_llm_text_gen.py.
75-
- Removed Dockerfile lines cloning TorchChat repo and setting safe.directory.
76-
- Updates huggingface_hub to 0.34.0.
80+
- Updates hashes for:
81+
- PYTORCH_HASH to 4e2ddb5db67617f9f5309c8bba0c17adc84cadbc, 2.9.0.dev20250808 from viable/strict, August 8th.
82+
- IDEEP_HASH to 3527b0bf2127aa2de93810feb6906d173c24037f, from ideep_pytorch, August 1st.
83+
- ONEDNN_HASH to 7e85b94b5f6be27b83c5435603ab67888b99da32, from main, August 1st.
84+
- ACL_HASH to 3c32d706d0245dcb55181c8ced526eab05e2ff8d, from main, August 1st.
85+
- TORCH_AO_HASH to 8d4a5d83d7be4d7807feabe38d37704c92d40900, from main, August 1st.
86+
- KLEIDIAI_HASH to 8ca226712975f24f13f71d04cda039a0ee9f9e2f, v1.12 from main.
87+
- Update the examples/transformers_llm_text_gen.py to use the new quantizer api Int8DynamicActivationIntxWeightConfig.
88+
- Deleted torchchat_llm_text_gen.py.
89+
- Removed Dockerfile lines cloning TorchChat repo and setting safe.directory.
90+
- Updates huggingface_hub to 0.34.0.
7791

7892
### Removed
79-
- https://github.com/pytorch/pytorch/pull/151547, to update OpenBLAS commit as this has been merged upstream.
80-
- Temporarily removed --compile flag from transformers_llm_text_gen.py example as there is an unresolved issue with torch compile support in this case.
93+
- https://github.com/pytorch/pytorch/pull/151547, to update OpenBLAS commit as this has been merged upstream.
94+
- Temporarily removed --compile flag from transformers_llm_text_gen.py example as there is an unresolved issue with torch compile support in this case.
8195

8296
### Fixed
83-
- Updates various Python packages to address known vulnerabilies with a high CVSS score:
84-
- Updates Transformers to 4.55.2, this also provides a mitigation for CVE-2025-2099.
85-
- Updates Wheel version to 0.38.0 as a mitigation for CVE-2022-40898.
86-
- Updates setup-tools to 78.1.1 as a mitigation for CVE-2025-47273 and CVE-2024-6345.
87-
- Updates Torchvison to 0.23 to avoid the need to use `--extra-index-url`
88-
this is the recomended mitgation against CVE-2018-20225, affecting all versions of pip.
97+
- Updates various Python packages to address known vulnerabilies with a high CVSS score:
98+
- Updates Transformers to 4.55.2, this also provides a mitigation for CVE-2025-2099.
99+
- Updates Wheel version to 0.38.0 as a mitigation for CVE-2022-40898.
100+
- Updates setup-tools to 78.1.1 as a mitigation for CVE-2025-47273 and CVE-2024-6345.
101+
- Updates Torchvison to 0.23 to avoid the need to use `--extra-index-url`
102+
this is the recomended mitgation against CVE-2018-20225, affecting all versions of pip.
89103

90104
## [r25.07] 2025-07-11
91105
https://github.com/ARM-software/Tool-Solutions/tree/r25.07

‎ML-Frameworks/tensorflow-aarch64/CHANGELOG.md‎

Lines changed: 13 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -12,8 +12,19 @@ where `YY` is the year, and `MM` the month of the increment.
1212
### Changed
1313

1414
### Removed
15-
- Removes protobuf packages from Docker image that are not required for running tensorflow. Mitigation for CVE-2025-4565.
16-
- Removes system level python packages that aren't used.
15+
16+
### Fixed
17+
18+
## [r25.10] 2025-10-15
19+
https://github.com/ARM-software/Tool-Solutions/tree/r25.10
20+
21+
### Added
22+
23+
### Changed
24+
25+
### Removed
26+
- Removes protobuf packages from Docker image that are not required for running tensorflow. Mitigation for CVE-2025-4565.
27+
- Removes system level python packages that aren't used.
1728

1829
### Fixed
1930
- Updates Transformers to 4.53.0 as a mitigation for: CVE-2025-6051, CVE-2025-3933, CVE-2025-6638, CVE-2025-3262, CVE-2025-3263, CVE-2025-3264, CVE-2025-5197, CVE-2025-6921, CVE-2025-3777.

0 commit comments

Comments
 (0)