Skip to content

[Infra:Chore] Merge latest upstream MNN - #2

Merged
pruthvikar merged 219 commits into
cleanup/remove-unused-apps-and-projectsfrom
codex/update-mnn-latest
Jul 20, 2026
Merged

[Infra:Chore] Merge latest upstream MNN#2
pruthvikar merged 219 commits into
cleanup/remove-unused-apps-and-projectsfrom
codex/update-mnn-latest

Conversation

@pruthvikar

Copy link
Copy Markdown
Member

What

  • merge Alibaba MNN master at df6f64903de86384cf36e1001389148fe9746bb2
  • retain the cleaned fork's removal of the unused apps/ and project/ trees
  • preserve the upstream commit ancestry for subsequent updates

Why

The cleaned fork was 217 upstream commits behind Alibaba master. mnn-sys should pin a commit reachable from the maintained fork branch instead of depending on an unmerged feature branch.

Validation

Validated through getcarv/mnn-sys:

  • cargo test
  • cargo clippy --features metal,coreml -- -D warnings
  • cargo test --features metal,coreml

cargo fmt --check reports existing formatting differences in the unchanged mnn-sys/build.rs.

Juude and others added 30 commits March 6, 2026 11:08
* 解决benchmark crash、add tests

* feat(mnnchat): commit android app and downloader updates

* chore(mnnchat): update jsonviewer dependency artifact

* refactor(mnnchat): remove manual executor scope from LlmSession::Load

* feat(mnnllmchat): merge MNN API compat and add API regression tests

* fix(mnnllmchat): back arrow button in SelectTextActivity now works (alibaba#4185)

The back arrow button was enabled but had no click handler.
Added onOptionsItemSelected() override to handle android.R.id.home
and call finish() to close the activity.

Co-Authored-By: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com>

* [MNN:Version] Update to 3.4.1.

Discussed-in: Merge-Request 26193851 , URL: https://code.alibaba-inc.com/AliNN/AliNNPrivate/codereview/26193851
GitOrigin-RevId: 9da7011c8d4ba35e3c710dbf085bf76a06939a0f

* fix(api/settings): unify 401 body path and dropdown init order

* fix(api-service): support runtime model switch and foreground start ordering

* chore(mnnllmchat): release 0.8.1 notes and version bump

* feat(mnnllmchat): update test and diffusion integration flows

* chore: restore root .gitignore to match origin/master

---------

Co-authored-by: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com>
Co-authored-by: MNNSyncBot <hi@zhaode.wang>
Discussed-in: Merge-Request 26183427 , URL: https://code.alibaba-inc.com/AliNN/AliNNPrivate/codereview/26183427
GitOrigin-RevId: 30739177bf5d7c89cacdcc1b5435b19911d47465
Discussed-in: Merge-Request 26209838 , URL: https://code.alibaba-inc.com/AliNN/AliNNPrivate/codereview/26209838
GitOrigin-RevId: 97839fd7d72ae022fa302b983d427419f1b237c5
…e no-bias path; Remove defer encoder (alibaba#4219)

* Metal:Speed: Optimize gemv of int4/int8 by enlarge localsize from 64->128

* Metal:Speed: When scale's bias is empty, reduce memory and compute

* Metal:Refractor: Remove defer encoder

* Metal:Speed: Reduce extra action compute for MetalBinary

* Metal:Speed: Add plane opt for MetalBinary
…a#4195)

Discussed-in: Merge-Request 26218300 , URL: https://code.alibaba-inc.com/AliNN/AliNNPrivate/codereview/26218300
GitOrigin-RevId: c48cf69d99d9c2fbd62a5eb3c87564741357c674
将子图名称的生成从使用 `char` 数组和 `sprintf` 改为使用 `std::string` 和字符串拼接,以提高代码的安全性和可读性。
Link: https://code.alibaba-inc.com/AliNN/AliNNPrivate/codereview/26239698
* [MNNConvert:Bugfix] fix overflow in MNNDump2Json (alibaba#4225)
GitOrigin-RevId: 7b3278084f40d4870bdf787bade3a68ef1591956
Discussed-in: Merge-Request 26242650 , URL: https://code.alibaba-inc.com/AliNN/AliNNPrivate/codereview/26242650
GitOrigin-RevId: fe5583cf6ea930d0d613d6133fdea09ec4c842a3
…rruption (alibaba#4171)

  - Add hardware AEC (VOICE_COMMUNICATION audio source) and NoiseSuppressor for clean audio capture
  - Add software Auto-Mute fallback mode for devices with poor hardware AEC                                                                                      
  - Add manual mute button with strict silence injection
  - Implement generation ID mechanism to invalidate stale tasks on interruption
  - Keep mic recording during LLM generation to enable real-time speech detection
  - Add echo cancellation mode toggle UI and smooth scroll for transcripts
  - Add supertonic-tts-mnn model to model market
Discussed-in: Merge-Request 26257966 , URL: https://code.alibaba-inc.com/AliNN/AliNNPrivate/codereview/26257966
GitOrigin-RevId: b2dfdc6e1fa7e9eedade45ee70be5fefb39ddd92
1. Fix CloneContext::getOrClone removed but still used in ConvBNReluFusedModule
   (from alibaba#4102: Don't copy kvcache when meta is not the same)
2. Fix null pointer crash in PTQ calibration when debug mode is off
   (from alibaba#4151: Fix cosine distance calculation error in offline quantization)
3. Fix Shape op defaultDimentionFormat forced to NCHW breaking TF models
   (from alibaba#4230: Fix shape op compute onnx's NC4HW4 input error)
4. Fix ShapeOnnx converter always setting OpParameter_ShapeParam, causing new
   models without start/end to be incompatible with old engines
   (from alibaba#4217: Support ONNX Shape start/end)

Discussed-in: Merge-Request 26261909 , URL: https://code.alibaba-inc.com/AliNN/AliNNPrivate/codereview/26261909
GitOrigin-RevId: 853d4f4d6b794e24023b8ac758e3377b49db580f
Discussed-in: Merge-Request 26271429 , URL: https://code.alibaba-inc.com/AliNN/AliNNPrivate/codereview/26271429
GitOrigin-RevId: 3355da4651aca92859cb8107cc503c7a1621dbf9
…ibaba#4249)

* feat(mnncli): add Anthropic compatibility and verification coverage

* feat(api-service): add local Anthropic API compatibility and regression tooling

* feat(android): improve model, storage, swipe, and diffusion workflows

* feat(chat): improve LaTeX rendering and streaming regression coverage

* test(smoke): add voice UI coverage and harden regressions

* feat(voice): add dumpapp coverage and tts demo diagnostics
…alibaba#4239)

* Converter:Bugfix: When onnx prelu's slope size is 1, set to relu

* QNN:Feature: Add Transpose/UnPack/Pack in QNN

* QNN:Bugfix: Fix bug for some name may cause compile error
Qxinyu and others added 26 commits July 1, 2026 15:38
[QNN:Bugfix] Raise stack limit for context generation
在 `CPUConvolution.cpp` 和 `CPUBackend.cpp` 中添加了检查,确保在执行 `ConvInt8` 操作时具有量化权重数据,以防止因缺少量化权重而导致的空指针解引用崩溃。
Link: https://code.alibaba-inc.com/AliNN/AliNNPrivate/codereview/28339484
* [CPU:Bugfix] Fix ConvInt8 SIGSEGV when Conv lacks quantized weight
GitOrigin-RevId: 3ab854da7248261bcf77ff0d89fa9589df23cb0b
GitOrigin-RevId: 7c398bf40f2d120ce4b59323be1167d903276b59
GitOrigin-RevId: b10594c9a301d5eec89b7ee35a379b21006ae419
…g traversal order, optimize gemm int4 performance, split binary_layernorm_c4 kernel (alibaba#4606)

GitOrigin-RevId: 96071f89fc65f5a4df0eb67feee84cc1dbf70f9b
GitOrigin-RevId: 47af53ebaad0eff97de2afac4670b5626d430d69
…une & missing NC4HW4 output layout (alibaba#4609)

GitOrigin-RevId: c8dc18ebcb23d15c8ce0c4f7a1d9b2a2dbe7c782
…aba#4617)

GitOrigin-RevId: e1d27551e71f9adf6c82a67479d91a3c4d65186d
…#4619)

Discussed-in: Merge-Request 28555540 , URL: https://code.alibaba-inc.com/AliNN/AliNNPrivate/codereview/28555540
GitOrigin-RevId: de82056cbbfed0f160a93823874bda2d7ede248a
GitOrigin-RevId: 12deb95cd39e3926a1a7280cca81ba15d9e9feb4
Discussed-in: Merge-Request 28514756 , URL: https://code.alibaba-inc.com/AliNN/AliNNPrivate/codereview/28514756
GitOrigin-RevId: 61a575a5b05d34ce06106624478ae02cd0db5676
…nds (alibaba#4630)

When mmap-backed allocation fails (disk full, mmap error, file too short),
backends silently stored null device pointers into tensors, causing
EXC_BAD_ACCESS / SIGBUS crashes on iOS.

Fixes across all backends:
- LLM: grow mmap cache file before mmap to avoid SIGBUS past EOF;
  release module before runtime in load() to fix use-after-free
- Metal: check nil Metal buffer from failed mmap, fall back to
  non-mmap GPU allocator (mStaticAllocatorRaw)
- OpenCL: null-check all alloc() paths in onAcquire, fall back to
  non-mmap BufferPool; propagate MmapPool failure instead of wrapping null
- CPU: fall back to mStaticAllocatorRaw when mmap static allocation fails
- Vulkan: fix CALL_VK no-op in release builds; null-check allocations
  in onAcquire, VulkanBuffer/VulkanImage constructors
- CUDA: add null checks after pool->alloc() in all 26 execution kernels,
  return OUT_OF_MEMORY or set mValid=false

Discussed-in: Merge-Request 28543428 , URL: https://code.alibaba-inc.com/AliNN/AliNNPrivate/codereview/28543428
GitOrigin-RevId: b59217536d041370f22167289464ad7225a14178

Co-authored-by: jingbang.yjb <jingbang.yjb@alibaba-inc.com>
Discussed-in: Merge-Request 28608916 , URL: https://code.alibaba-inc.com/AliNN/AliNNPrivate/codereview/28608916
GitOrigin-RevId: c5236270381e64c1a74570c3c26667009e45d6bb
…ba#4634)

Discussed-in: Merge-Request 28635716 , URL: https://code.alibaba-inc.com/AliNN/AliNNPrivate/codereview/28635716
GitOrigin-RevId: 0ddd7811d82c82d7dd0c881bfecf6c1139a7d4f4
Discussed-in: Merge-Request 28636477 , URL: https://code.alibaba-inc.com/AliNN/AliNNPrivate/codereview/28636477
GitOrigin-RevId: 98c97dc87771fcce8ce1dc4f56a18ede69f8dd6d
…ba#4644)

GitOrigin-RevId: 65961f320a0a67d698e12b829d6f39ad2b5d4e18
@pruthvikar pruthvikar self-assigned this Jul 20, 2026
@github-actions

Copy link
Copy Markdown

Seems you are using me but didn't get OPENAI_API_KEY seted in Variables/Secrets for this repo. you could follow readme for more information

@pruthvikar
pruthvikar merged commit 3fa5d18 into cleanup/remove-unused-apps-and-projects Jul 20, 2026
1 check passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.