本期总览
本期跟踪 9 个仓库,其中 9 个仓库有有效动态,筛出 22 项重点变化。整体重点是:版本发布 9 项、硬件/架构支持 5 项、安全修复 3 项、工具链/开发体验 2 项。
本期跟踪 9 个仓库,其中 9 个仓库有有效动态,筛出 22 项重点变化。整体重点是:版本发布 9 项、硬件/架构支持 5 项、安全修复 3 项、工具链/开发体验 2 项。
AI 与开发工具 本周期有 4 个活跃仓库,筛出 10 项重点变化,主要集中在版本发布、安全修复、硬件/架构支持、性能优化。代表仓库包括 llama.cpp、vLLM、Transformers、Ollama。这类变化更适合从本地推理、开发者工作流和模型应用集成角度跟进。
Linux 与基础软件 本周期有 2 个活跃仓库,筛出 6 项重点变化,主要集中在版本发布、工具链/开发体验、功能变化、硬件/架构支持。代表仓库包括 systemd、LLVM。这类变化更适合从发行版同步、系统稳定性、网络/安全边界和长期维护成本角度评估。
RTOS 与嵌入式 本周期有 2 个活跃仓库,筛出 5 项重点变化,主要集中在版本发布、硬件/架构支持、安全修复。代表仓库包括 Zephyr、FreeRTOS。这类变化更适合从板级适配、驱动成熟度、实时系统可维护性和量产风险角度评估。
工具链与调试 本周期有 1 个活跃仓库,筛出 1 项重点变化,主要集中在硬件/架构支持。代表仓库包括 pyOCD。这类变化会直接影响编译、烧录、调试和硬件 bring-up 效率。
版本发布出现 9 项。适合检查新版本说明、兼容性变化和是否值得进入下一轮评估。
硬件/架构支持出现 5 项。对嵌入式开发的意义主要在于芯片、板级、驱动和架构覆盖继续补齐。
安全修复出现 3 项。优先关注是否影响现有系统的升级窗口、补丁回归和下游发行版同步节奏。
工具链/开发体验出现 2 项。对日常工程效率、调试链路、构建系统和 CI 稳定性更有参考价值。
Issues fixed These GitHub issues were addressed since the previous 4.3.0 tagged release: 55186 posix: fnmatch: fix known bugs 61464 USB device stack new and old assertion on STM32 95359 spi loopback fails on frdm rw612 with DMA and/or cs loopback 96699 drivers: spi: nrf spim: Unused function warning when driver deinit Kconfig is disabled 96762 drivers: serial: uart nrfx uarte: Unused function warning when driver deinit Kconfig is disabled 98491 riscv: userspace: undefined symbol: z stack space get 98501 pm: device runtime: Issues with set/clear PD CLAIMED flag in ISR SAFE context 98523 ring buffer: ring buf init may trigger assertion depending on Kconfig 98588 drivers: i2c: dw i2c: i2c read time out on certain type of DW I2C 98768 STM32F303 bxCAN: Last 2 bytes corrupted on TX pin, but registers are correct 98782 esp32c6: esp32h2: Openthread issue 98797 boards: nxp: mimxrt1180 evk: Jlink
代表性合并:#109310 tests: drivers/comparator/gpio_loopback: Add FRDM-iMXRT1186;#111104 [Backport v3.7-branch] frdm_mcxc242 lpuart support async api;#111185 drivers: ethernet: Add missing iface arg Cyclone V Ethernet
代表性合并:#110951 tests: with_mcumgr: Rework to use prebuilt signed images;#110797 boards: st: stm32h573i_dk/b_u585i_iot02a: Use EC-P256 as default signature type;#111198 release: Zephyr 4.3.1-rc1
代表性合并:#109547 Backport CVE 2026-5590 fix for 4.3
代表性合并:#1401 Fix: Replace size_t with uint32_t in TaskNotifyArray.c for 16-bit dev…
代表性合并:#42700 [Bugfix] Replace deprecated Qwen2VLImageProcessorFast with Qwen2VLImageProcessor;#45376 [Bugfix] Set type/role explicitly in streaming message_start event
Release v5.12.0 New Model additions MiniMax M3 VL <img width="886" height="583" alt="image" src="https://github.com/user attachments/assets/ae9dd96f 6877 4531 a06b a756686f24e5" / MiniMax M3 VL is the vision language member of the MiniMax M3 family that pairs a CLIP style vision tower with 3D rotary position embeddings with the MiniMax M3 text backbone. It uses a mixed dense/sparse Mixture of Experts decoder with SwiGLU OAI gated experts and a lightning indexer for block sparse attention. The model processes images through a Conv3d patch embedding system and includes specialized components for efficient multimodal understanding and generation. Links: Documentation https://huggingface.co/docs/transformers/main/en/model doc/minimax m3 vl Add minimax m3vl 46600 by @ArthurZucker in 46600 https://github.com/huggingface/transformers/pull/46600 PP OCRv6: update documentation and slow tests 4657
代表性合并:#24550 server: clean up static assets handling;#24260 Add arch support for cohere2-MoE;#18039 [Speculative decoding] feat: add EAGLE3 speculative decoding support
代表性合并:#45412 [Doc] Fix uv dependency resolution failure for setuptools during CPU source builds (x86 & ARM);#45252 [Security] Fix DoS via prompt_embeds on M-RoPE models;#45118 [Security] Add timeout guard for regex compilation in structured outp…
vLLM v0.23.0 Release Notes Please note that Minimax M3 is not yet supported in this version. Please follow vLLM recipe https://recipes.vllm.ai/MiniMaxAI/MiniMax M3 for usage guides for M3. Highlights This release features 408 commits from 200 contributors 63 new ! DeepSeek V4 matures across backends : Following its introduction in v0.22.0, DeepSeek V4 received another large hardening and optimization pass. Its sparse MLA metadata is now decoupled from DeepSeek V3.2 44699 , it gained a TRTLLM gen attention kernel 43827 , EPLB support for the Mega MoE 43339 , selective prefix cache retention for sliding window KV cache 43447 , and an index share feature for DSA MTP 44420 . The model was also detached from torch.compile 43746, 43891 , its attention and RoPE paths were refactored 44569, 44262, 43926 , and an XPU attention decode path was added 42953 . Model Runner V2 expands to more dense mo
代表性合并:#46596 Fix docstring formatting issues causing Sphinx autodoc warnings;#46605 Fix `peft` lower bound;#46588 Lfm2: thread `seq_idx` through ShortConv for packed/varlen inputs
评分:105
代表性合并:#45460 [Bugfix] Return the tokenizer from maybe_make_thread_pool so it survives pickling;#45003 [Frontend] Support strict mode for tool calling;#45468 [Bugfix] Reject structured outputs for diffusion decoders with a clear error
代表性合并:#46591 [CI] Enable PR CI for all fork PRs via security gate;#46483 Require `trust_remote_code` to run a local-directory `custom_generate`
What's Changed Fixed ollama launch selecting the wrong provider in some cases Improved prompt caching by decoupling it from context shift for better KV cache reuse More stable MLX inference with hardened linear and embedding layers MLX runner now creates snapshots during prompt processing and speculative decoding for improved reliability Improved recurrent model support with per boundary states from the gated delta kernels Full Changelog : https://github.com/ollama/ollama/compare/v0.30.7...v0.30.8
代表性合并:#46533 [CB] [Minor] Add parameter to tune default compile level;#46593 [CB] Fix seqlens and use TypedDict
代表性合并:#42539 vsock-util: allow VMADDR_CID_ANY as local CID if enabled in hwdb;#42569 journal: fix byte-order conversion in journal_file_append_data()
代表性合并:#202490 [clang-tidy] Avoid non-const-parameter fix-it conflicts with overloads;#203689 Reapply "[lit] Deprecate execute_external=True in ShTest" (#203316);#203674 [orc-rt] Add ORC_RT_C_NOTHROW and apply it to the Error C API.
代表性合并:#42562 core: clear pending_reload_message_vl on manager varlink teardown;#42537 pcrextend: skip measurement gracefully when no usable PCR bank is available;#42561 ask-password: do not reuse flags across Varlink calls
代表性合并:#203289 [mlir][sparse_tensor] Fix out-of-bounds read in SparseAssembleOpConverter;#203113 [SPIR-V] Preserve pointer-to-pointer element type for T*& parameters;#201749 [libc++] LWG3767: `codecvt<charN_t, char8_t, mbstate_t>` incorrectly added to locale
代表性合并:#202475 Update remaining tests for usage of CLANG_RESOURCE_DIR;#193069 [VPlan] Run final removeBranchOnConst after late simplifyRecipes.;#200130 [Xtensa] Implement support of the ESP32S2 target.
代表性合并:#40987 Assorted fixes;#42573 po: Translated using Weblate (Chinese (Simplified) (zh_CN));#42572 Translations update from Fedora Weblate
评分:88
代表性合并:#1982 cbuild-run: add new node processing
工具链/开发体验:4 项,core: clear pending_reload_message_vl on manager varlink teardown;pcrextend: skip measurement gracefully when no usable PCR bank is available
功能变化:3 项,Assorted fixes;po: Translated using Weblate (Chinese (Simplified) (zh_CN))
版本发布:2 项,vsock-util: allow VMADDR_CID_ANY as local CID if enabled in hwdb;journal: fix byte-order conversion in journal_file_append_data()
工具链/开发体验:15 项,[clang-tidy] Avoid non-const-parameter fix-it conflicts with overloads;Reapply "[lit] Deprecate execute_external=True in ShTest" (#203316)
功能变化:8 项,[mlir][sparse_tensor] Fix out-of-bounds read in SparseAssembleOpConverter;[SPIR-V] Preserve pointer-to-pointer element type for T*& parameters
硬件/架构支持:3 项,Update remaining tests for usage of CLANG_RESOURCE_DIR;[VPlan] Run final removeBranchOnConst after late simplifyRecipes.
硬件/架构支持:17 项,tests: drivers/comparator/gpio_loopback: Add FRDM-iMXRT1186;[Backport v3.7-branch] frdm_mcxc242 lpuart support async api
功能变化:8 项,llext: misc fixes and error checking;Bluetooth: BAP: SD: Fix log of wrong err value
版本发布:3 项,tests: with_mcumgr: Rework to use prebuilt signed images;boards: st: stm32h573i_dk/b_u585i_iot02a: Use EC-P256 as default signature type
安全修复:1 项,Backport CVE 2026-5590 fix for 4.3
硬件/架构支持:1 项,Fix: Replace size_t with uint32_t in TaskNotifyArray.c for 16-bit dev…
硬件/架构支持:1 项,cbuild-run: add new node processing
版本发布:6 项,server: clean up static assets handling;Add arch support for cohere2-MoE
工具链/开发体验:1 项,ui : fix llama-ui-embed crash when no asset dir is given
性能优化:1 项,fit : wrap llama_device_memory_data
硬件/架构支持:4 项,[Bugfix] Return the tokenizer from maybe_make_thread_pool so it survives pickling;[Frontend] Support strict mode for tool calling
工具链/开发体验:4 项,Treat null completion max_tokens like the default;[Model Runner V2] Fix `openai.InternalServerError: Error code: 500 - 'list index out of range'`
安全修复:3 项,[Doc] Fix uv dependency resolution failure for setuptools during CPU source builds (x86 & ARM);[Security] Fix DoS via prompt_embeds on M-RoPE models
版本发布:2 项,[Bugfix] Replace deprecated Qwen2VLImageProcessorFast with Qwen2VLImageProcessor;[Bugfix] Set type/role explicitly in streaming message_start event
版本发布:6 项,Fix docstring formatting issues causing Sphinx autodoc warnings;Fix `peft` lower bound
工具链/开发体验:4 项,[CI] capture checkers output in OTEL;Fix stop string matching for byte-fragment tokens
功能变化:3 项,a11 for checkers;Add minimax m3vl
安全修复:2 项,[CI] Enable PR CI for all fork PRs via security gate;Require `trust_remote_code` to run a local-directory `custom_generate`
工具链/开发体验:1 项,launch: Fix launch provider drift