Research window: 2026-11-10 04:30 ~ 2026-11-11 04:30 (Beijing time, past 24 hours; directly contiguous with the cutoff of the previous 11-10 report).

Sources: GitHub key repository and organization push verification (RuyiAI-Stack, ruyisdk, openRuyi-Project, buddy-compiler, with per-repository checks of commits, PRs, Releases and tags within the window, branch-level atom as fallback), ruyi package manager official Release and changelog comparison, packages-index, openRuyi packaging repository SPECS, ruyiai-daily daily intelligence archive, RuyiSDK developer community forum, three-page snapshot of the RuyiAI official homepage, openRuyi distribution news page, Google News RSS targeted queries in Chinese and English (via proxy), arXiv search.


This Issue’s Index

  • Today’s Highlights: ruyi 0.54.0 officially released, host ABI compatibility check enters user commands (11-10)
  • Today’s Highlights: triton-riscv enables the AMD GPU backend by default on RISC-V hosts (11-10)
  • I. RuyiAI Stack Progress (PyTorch / Triton / TileLang RISC-V adaptation)
    • 1.1 triton-riscv: plugin rename PR under review, environment variable values will change accordingly (11-10)
    • 1.2 pytorch, tilelang-riscv, llvm-project: zero commits in the window (11-11 status)
  • II. RuyiSDK Toolchain and Distributions
    • 2.1 ruyi-imager: WiX upgraded to 7.0.1 and its EULA accepted in CI (11-10)
    • 2.2 openRuyi: eight commits, four security fixes plus LLVM snapshot and Wayland switch (11-10)
    • 2.3 openRuyi: release spec repo aligns requirement drafting and stack naming (11-10)
    • 2.4 openRuyi monthly cadence: tags and news page still stuck at 2026.09 (11-11 status)
  • III. Vector Extensions and Operator Optimization
    • 3.1 Zero commits across the three vector repos, vector-side increments concentrated in VIR reductions and size constraints (11-11 status)
  • IV. Buddy Compiler News
    • 4.1 buddy-mlir: YOLOv8 detection and pose run on the K3 matrix engine, two fixes for weights and TCM copies (11-10)
    • 4.2 buddy-mlir: vir.set_vl regions tightened to a single block, multi-block regions no longer silently drop operators (11-10)
    • 4.3 buddy-mlir: printing, parsing, and validation completed for dynamic vector types and scaling factors (11-10)
    • 4.4 buddy-mlir: reassoc reductions merged by lane first, fixing bit inconsistencies on VLEN 128 hosts (11-10)
    • 4.5 buddy-mlir: fifty Python files renamed, A100 llc options consolidated into a unified entry point (11-10)
    • 4.6 buddy-mlir nightly stuck at 0.1.4.dev20261105, six merged commits not yet packaged (11-11 status)
  • V. Community, Events, and Official Homepage
    • 5.1 Official homepage unchanged across three pages, two blog posts, twenty-three documentation chapters (11-11 status)
    • 5.2 No new forum posts in the window, latest biweekly progress still issue 080 (11-11 status)
    • 5.3 ruyiai-daily archive stuck at the 11-08 issue (11-11 status)
    • 5.4 Zero external news hits, latest arXiv entry stuck at 11-03 (11-11 status)
  • VI. Trend Observations
    • 6.1 The 0.54 release interval doubled, at the cost of ABI collection across three architectures
    • 6.2 “Can it be installed” becomes a computable fact for the first time
    • 6.3 RISC-V host plus AMD card enters the mainline
    • 6.4 VIR fixes shift from “computing wrong” to “failing silently”

Today’s Highlight: ruyi 0.54.0 Officially Released, Host ABI Compatibility Checks Enter User Commands (11-10)

Date: 2026-11-10 Source: ruyi 0.54.0 release, 0.53.0 to 0.54.0 comparison, PR #507

0.54.0 was released on 11-10 at 21:15 (Beijing time) as a stable release, not a pre-release. The previous version, 0.53.0, was released on 09-24, a gap of forty-seven days; in between there was only 0.54.0-beta.20261102 on 11-02. Relative to that pre-release, the stable release adds only a version-number commit, so its feature set is exactly the full content of the pre-release; counting from 0.53.0, this version comprises seventy-one commits across seventy-four files.

The main feature is the collection and matching of ELF ABI compatibility information. Whether a prebuilt binary can run on a target device depends on a set of facts scattered across ELF file headers: which dynamic libraries it needs, which symbol versions it requests from those libraries, and architecture-specific capability requirements. Previously nothing in the software source recorded these; ruyi now scans this information out, stores it, and compares it before installation.

Collection is split by architecture: x86-64 records ISA feature level, AArch64 records BTI and PAC, and RISC-V records the ISA string. The RISC-V path is the most labor-intensive, since its extensions are assembled into a string with version numbers; this cycle’s commits include parsing versioned single-letter extensions, filtering ld.so.cache entries by host floating-point ABI, and factoring in stack alignment requirements. On the host side, probing for x86-64 and AArch64 is also implemented; host libraries are resolved by the requested soname, and an unresolvable soname is reported as “undetermined” rather than a hard failure.

The user-visible changes land in three commands: ruyi list shows host compatibility in its notes (non-binary packages hide this item), ruyi install warns when installing an incompatible package, and ruyi admin check-abi is available for troubleshooting. The data plane goes through repository-side ABI sidecars, resolved by distfile name; rescan-package-abi and plugin artifacts gain strip-components to handle archives with leading directories. Sidecars have a size limit, and a bad sidecar is discarded as “no ABI information” rather than failing the entire installation.

Another category is fixes to the fetch pipeline: #509 makes non-2xx HTTP responses report errors faithfully, discards a failed first download, and adds a cap to stalled transfers. Such changes are inconspicuous, but for a package manager that pulls hundreds of megabytes of toolchains from mirror sites, they affect the experience more than new features do.

The design document docs/elf-abi-compat-design.md is committed to the repository with the release; the installation paths given in the release notes remain PyPI and the mirror site operated by the service, and the single-file binary must be renamed to ruyi before use.


Today’s Highlight: triton-riscv Enables the AMD GPU Backend by Default on RISC-V Hosts (11-10)

Date: 2026-11-10 Source: PR #110, README

PR #110 was merged on 11-10 at 17:51 (Beijing time), titled “enable GPU backend by default”. The change does three things: builds and packages the riscv64 AMD codegen library using the existing Buddy LLVM SDK; fixes the ambiguity on riscv64 when choosing between the CPU and GPU drivers, with the rule being to prefer an available GPU; and changes the check in the driver layer that made the choice ambiguous (that file has a net deletion of four lines).

Validation was done on real hardware: an SG2044 paired with an RX 7900 XTX, with torch and ROCm from the openRuyi software repository; it also tested whether the system automatically falls back to CPU when GPU devices are hidden. This bar is much harder than “it compiles”—it shows that a RISC-V host actually dispatched kernels to an AMD card rather than falling back to the CPU.

The README adds a section on AMD GPU usage: when ROCm is present, the backend activates itself via torch.version.hip, in the same style as the upstream tutorial; the wheel does not bundle the ROCm runtime, so torch and ROCm must be installed separately; to force CPU, use TRITON_DEFAULT_BACKEND=triton_shared. The change spans six files, concentrated in README, backend/driver.py, and four scripts under scripts: build, env, release, and setup.

One timing gap is worth noting: the nightly build nightly/2026-11-10-209274f published that day at 10:22 (Beijing time) was built on commit 209274f, earlier than this merge, so that package does not yet include the GPU backend. Getting it requires waiting for the next nightly.

For the stack as a whole, this is a roadmap-level change: triton-riscv previously only answered “how to run Triton on the vector units of a RISC-V CPU”; now it must also answer “how to drive heterogeneous accelerators from a RISC-V host”. At a time when no mainstream AI accelerator card ships a RISC-V host driver, supporting AMD cards first is the most realistic step—the RISC-V side of ROCm work is already maintained in the openRuyi repository.


I. RuyiAI Stack Progress (PyTorch / Triton / TileLang RISC-V Porting)

1.1 triton-riscv: Plugin Rename PR Under Review, Environment Variable Values Will Change Accordingly (11-10)

Date: 2026-11-10 Source: PR #109, commit list

The problem #109 addresses is a misnomer: name.conf determines the installed package name and entry point, so the plugin is currently called triton.backends.triton_shared, yet the target it reports is cpu, and the option prefix is already triton_riscv. The fix is to unify the name as riscv, on the grounds of aligning with upstream naming conventions—plugins are named by target family (nvidia, amd), while device names (cuda, hip) are left to the target.

A related change is the environment variable: the value of TRITON_DEFAULT_BACKEND changes from triton_shared to riscv, meaning the “force back to CPU” approach described in today’s highlight will stop working once this is merged. The tool name triton-shared-opt and the C++ side remain unchanged. A second commit adds ruff to the sixty files touched by the rename; these files had never been linted before and carry nineteen existing violations.

The PR is open, with an update at 18:56 on 11-10 (Beijing time), and has not yet been merged. Other open PRs updated the same day include #103 (adding tutorials) and #104 (BOSCAME FPGA SPM example and AOT contract frontend), all three being foundational work that is “hands-on and reproducible.”

1.2 pytorch, tilelang-riscv, llvm-project: Zero Commits in Window (11-11 status)

Date: 2026-11-11 (status) Source: pytorch riscv branch, tilelang-riscv, llvm-project

None of the three repos had any commits in the window: the head commit of pytorch’s riscv branch is still the 11-04 batch of syncs, tilelang-riscv is stopped at 11-03 06:13 UTC, and llvm-project’s main is stopped at 11-02 14:48 UTC. This issue’s RuyiAI stack increments all come from one merged commit and one PR under review in triton-riscv.


II. RuyiSDK Toolchain and Distributions

2.1 ruyi-imager: WiX upgraded to 7.0.1 with its EULA accepted in CI (11-10)

Date: 2026-11-10 Source: ruyi-imager commit list

Two commits landed in the flashing tool’s window, both on the dependency side: WiX was upgraded from an older version to 7.0.1, and its EULA is now explicitly accepted in CI because the new build requires it; zstd-jni was bumped to 1.5.7-26. The former affects the Windows installer generation pipeline, the latter affects image decompression. For a cross-platform flashing tool, neither changes functionality, but both determine whether CI can still produce packages.

2.2 openRuyi: eight commits — four security fixes plus an LLVM snapshot and a Wayland switch (11-10)

Date: 2026-11-10 Source: policycoreutils commit, lxc commit, cups commit, llvm-snapshot commit

Eight commits landed in the packaging repo’s window, falling into three categories.

Four security commits: policycoreutils was upgraded to 3.11 with a backported fixfiles fix, while lxc, libmtp, and cups each received backported upstream security patches. All four carry the security-fixes tag — routine upstream CVE tracking, but a part the distribution must follow up on the same day.

Two toolchain commits: llvm-snapshot and clang-wrap were both bumped to 20261109 on the same day. These two packages are the foundation of the self-hosting compile chain on RISC-V, and also the source of the LLVM archives that triton-riscv depends on at build time; a daily snapshot refresh means downstream gets a compiler that changes every week.

The remaining two: libssh was upgraded to 0.12.2, and SDL2 enabled its Wayland backend (previously only the X11 path existed). The SDL2 commit matters more to the desktop and gaming stack than it appears — many graphics programs create windows through SDL, and with the Wayland switch on, the desktop finally has a genuinely native path.

2.3 openRuyi: release spec repo aligns requirement writing and stack naming (11-10)

Date: 2026-11-10 Source: release spec repo commits

One commit landed in the spec repo’s window, unifying how requirement entries are written and how stacks are named. Documentation changes like this produce no artifacts, but they determine what the same thing is called in issues and release notes — for a distribution project with many people across institutions, naming drift is harder to clean up than code errors.

2.4 openRuyi monthly cadence: tags and news page still at 2026.09 (11-11 status)

Date: 2026-11-11 (status) Source: openRuyi distribution news page, release tag list

The two latest posts on the news page are both from 10-19 (the openRuyi-Project positioning piece and the Ruyi Firmware release), and the monthly version entry is still 2026.09 from 09-30; the repo’s tags also stop at 2026.09-experimental. Following the one-release-per-month cadence from 05 through 09, the October monthly release had not appeared as of this window, while the packaging repo has kept up a pace of several to a dozen-plus commits per day. The two are not contradictory: rolling packaging and monthly releases are two separate tracks, but the externally visible “version” has not been updated for a month.


III. Vector Extensions and Operator Optimization

3.1 Zero commits across the three vector repos; vector-side progress concentrates in VIR reductions and size constraints (11-11 status)

Date: 2026-11-11 (status) Source: FBGEMM, VecIntrinBench, IntrinTrans

The three vector-related repos saw zero commits in the window, with head commits stalled at 07-13, 02-13, and 06-25 respectively. This does not mean there is no progress on the vector side: this issue’s vector changes all occur at buddy-mlir’s VIR layer — the ordering of cross-block reduction merging (4.4) and the region constraints of vir.set_vl (4.2) are both preconditions that directly determine which vector instructions get generated. The benchmark and translation repos have been silent for months, indicating that investment on this track remains concentrated in the compiler rather than in measurement tooling.


IV. Buddy Compiler Updates

4.1 buddy-mlir: YOLOv8 detection and pose run on the K3 matrix engine; two fixes for weights and TCM copies (11-10)

Date: 2026-11-10 Source: PR #1042

#1042 brings Ultralytics YOLOv8 into buddy-cli’s rax format: yolov8n detection (eighty COCO classes) and yolov8n-pose (seventeen keypoints), both with 320x320 input. The f32 model goes through VIR lowering; if the spec sets ime: true, it instead runs on the SpacemiT K3 matrix engine.

The two fixes added during review are more informative than the model itself. First, weight validation: the manifest carries params_size, and the runner requires the weight file to match it — previously a short weight file would let the model read past the buffer; the runner is also now required to accept only rax packages, because when the library and weights were supplied separately, the pose model would be run with the detection model’s parameters. Second, TCM copies: the TCM copy of IME weights was cached by weight address, so when new weights were read into a freed buffer they would hit the stale copy; now the kernel-side invalidation function is called after every weight load.

The spec keeps both params_size and ime_params_size, so the f32 version can still be built when ime: false. This design follows the same line as the previous day’s end-to-end VIR run: the same image, two backend paths, one comparable baseline.

4.2 buddy-mlir: vir.set_vl region narrowed to a single block; multi-block regions no longer silently drop ops (11-10)

Date: 2026-11-10 Source: PR #1045

The region type of vir.set_vl was originally AnyRegion, with a “region restriction TBD” comment left in the source. The verifier and every lowering in -lower-vir-to-vector only take region.front(), so a multi-block region could pass verification while ops in the remaining blocks were silently dropped. The repro is clear: a load and store inside a set_vl land in the second block (after cf.br ^bb1), and the lowering result is two empty affine.for loops; an empty region passes verification and then fails in the pass with “anchor type not found”.

The fix changes the type to SizedRegion<1>, deletes the comment, and stops the verifier from guarding against empty regions itself (ODS checks the block count first), keeping only the “block has no arguments” check. Tests add two illegal inputs: a two-block region and a zero-block region. This kind of change adds no new functionality, but it turns a class of “compiles fine, computes wrong” paths into compile-time errors — its value is in diagnosability.

4.3 buddy-mlir: printing, parsing, and verification of dynamic vector types and scale factors completed (11-10)

Date: 2026-11-10 Source: PR #1046

Scale factors were previously printed in attribute form as #vir.sf<m2>, while the type syntax and examples give m2; the parser side still had a “error message TBD” comment, and behavior was inconsistent in several places: #vir.sf<x2> and mq would crash (parsing returned a null attribute without reporting an error, and the type parser treated it as valid), m3, m0, and f128 were accepted, !vir.vec<0x?xf32> was accepted, while !vir.vec<?xvector<4xf32>> was rejected with the location reported as unknown.

The change has three parts: printing now uses sf.str(), while #vir.sf<m2> can still be read in; the attribute side gains genVerifyDecl, requiring the ratio to be m or f with a value from 2 to 64 that is a power of two (matching the check already in the Python bindings); the type side requires all dimensions before the dynamic one to be static and positive. A second commit fixes error locations to point at the real position. For an intermediate dialect still taking shape, parser strictness directly determines whether frontends can report errors reliably.

4.4 buddy-mlir: reassoc reductions combine blocks by lane first, fixing bit inconsistency on VLEN 128 hosts (11-10)

Date: 2026-11-10 Source: PR #1047

On a VLEN 128 RISC-V host (the C920 core in SG2044), qk_norm in test_k3_w4_kernels.py showed bit inconsistency. The cause: q/k normalization in K3 decode attention puts the sixty-four f32 squares of one head into a set_vl region without shuffles, and lowering sizes by host (#1037), which under VLEN 128 becomes two blocks of vector<32xf32> (LMUL 8); but vir.reduce folds each block into the previous result in turn, which differs from the single vector<64xf32> reduction order in the vector kernel — LLVM splits the wide vector in half, adds lane by lane, then reduces once. CI’s riscv64 runs on VLEN 256, where there is only one block, so it always passed.

The fix makes cross-block reassoc reductions combine the way LLVM does: n blocks are first padded to a power of two h, block p is added to block p+h/2, until one block remains and is reduced once. This is already the fourth fix on the same theme (preceded by load-time VLEN validation, host sizing, and FMA tail rounding), all pointing at the same acceptance criterion: changing the implementation is fine, changing the numbers is not.

4.5 buddy-mlir: fifty Python files renamed; A100 llc options consolidated into a single entry point (11-10)

Date: 2026-11-10 Source: PR #1043, PR #1044

#1043 renames fifty hyphenated Python file names to underscores: with underscore-named files already in the same directories, hyphenated ones cannot be imported, so their functions cannot be tested or reused. The change covers models, examples, .agents/skills, and tests, with references in CMake, shell scripts, README, and docs updated accordingly. Of the two hundred forty-one Python files upstream in MLIR, two hundred thirty-seven use underscores; this aligns with upstream convention.

#1044 fixes the YOLOv8 IME path: lower_to_obj.sh assembled its own set of A100 llc options (including +zvl1024b and -riscv-v-vector-bits-max=1024), yet only used the latter to check the built VLEN. Since #1032, VLEN is specified uniformly via BUDDY_RISCV_VLEN, other models take it from a100_llc_args, and the check is against the exact 1024 that llc actually reads. The old script could be fooled by +zvl256b or a build with no VLEN: VIR lowered for an arbitrary VLEN while the object was compiled for 1024. Now the script appends the options from the unified entry point directly, stops if VLEN 1024 is not satisfied, and adds a test run with stub tools.

4.6 buddy-mlir nightly stuck at 0.1.4.dev20261105; six merges not packaged (11-11 status)

Date: 2026-11-11 (status) Source: Release list, commit list

The latest nightly is still 0.1.4.dev20261105 from 11-05. Since then, the 11-06 SPM merge, the two VLEN changes on 11-08, the 11-09 VIR convolution merge, plus the six in this window, have all gone unpackaged — five consecutive days without a nightly. Nightlies are a dependency of triton-riscv, and renovate tracks versions automatically, so a stalled release means downstream cannot get these dozen-plus changes, including the bit-consistency fix in 4.4.


V. Community, Events, and Official Homepage

5.1 Official homepage’s three pages unchanged; two blog posts, twenty-three documentation chapters (11-11 status)

Date: 2026-11-11 (status) Source: RuyiAI Official Homepage

The hashes of the index, blog, and docs pages match the previous inspection. The blog still has two posts (Triton’s RISC-V adaptation, PyTorch RISC-V CI practices), and the documentation index has twenty-three chapters. This issue’s triton-riscv GPU backend is the content most worth adding an article about on the site, and there is no corresponding article yet.

5.2 No new posts in the forum window; latest biweekly progress still issue 080 (11-11 status)

Date: 2026-11-11 (status) Source: Latest topics in the RuyiSDK community, Biweekly Progress Issue 080

There are no new topics in the forum window; the latest entry is still the 11-06 repost (Infineon RISC-V automotive MCU). The latest RuyiSDK biweekly progress is issue 080 from 11-05; issue 081 has not yet been published. Issue 080 stated that 0.54 would be released around mid-November; that version landed on schedule today, and the next biweekly progress issue will cover the official release as well.

5.3 ruyiai-daily archive stops at the 11-08 issue (11-11 status)

Date: 2026-11-11 (status) Source: ruyiai-daily commit list

The last commit in the daily intelligence repository was at 11-08 06:21 (Beijing time), corresponding to the 11-08 issue; the 11-09, 11-10, and 11-11 issues were not generated. Previously the repository produced one issue per day; this time it has been interrupted for three days.

5.4 Zero hits in external news; latest arXiv entry stops at 11-03 (11-11 status)

Date: 2026-11-11 (status) Source: Google News, arXiv RISC-V search

Within the general RISC-V crawl window, the results returned were all personal-learning CPU repositories, teaching-oriented pipeline implementations, and toolchain articles related to the Infineon AURIX RC1, unrelated to the Ruyi community; the five targeted queries for RuyiAI, RuyiSDK, Ruyi, openRuyi, and Buddy Compiler returned zero hits. The latest entry in the arXiv RISC-V search is from 11-03, with no new papers in the window.


VI. Trend Observations

6.1 The release interval doubled from 0.54, at the cost of ABI collection across three architectures

0.52 to 0.53 took twenty-two days; 0.53 to 0.54 took forty-seven. During that doubled stretch, commits were almost all concentrated on one line of work: taking ELF ABI compatibility from “no record at all” to “can collect, can store, can compare, can warn before installation.” This path requires handling x86-64 ISA levels, AArch64 BTI and PAC, and RISC-V versioned extension strings simultaneously—knowledge across three architectures rather than three similar coding tasks.

Notably, it landed first in a package manager rather than in some distribution’s quality process. The package manager is the only place where a judgment can be made at the moment the user clicks install; putting compatibility judgment here effectively moves distribution-level validation cost forward to a single local comparison.

6.2 “Can it install” becomes a computable fact for the first time

RISC-V fragmentation is not in the instruction set itself but in the baseline: the same riscv64 binary may fail to run on an RVA20 board, or may hit an illegal instruction directly on a core without a vector unit. Previously such problems relied on documentation, forum replies, and trial and error. 0.54 writes each package’s dynamic library requirements, symbol versions, and architecture capability requirements into a sidecar, probes the corresponding facts on the host side, and then compares them—so “this package won’t install” turns from a piece of experience into a reproducible conclusion.

The side effect is data maintenance: sidecars must be updated with packages, archive differences such as strip-components must be handled separately, and bad sidecars need an explicit degradation strategy. 0.54 wrote all these edge cases into the code, indicating the maintenance cost was already accounted for during design.

6.3 RISC-V host plus AMD card enters the mainline

triton-riscv expanded from “running Triton on RISC-V vector units” to “driving an AMD GPU from a RISC-V host,” with the default path selecting the GPU directly. This combination previously appeared only in vendor demos; now it has entered the default branch and has real-hardware validation (SG2044 paired with RX 7900 XTX).

The practical significance: the host driver ecosystems of mainstream AI accelerator cards are almost all prepared for x86 and Arm, and RISC-V machines have long been able to run inference only on the CPU. Letting a RISC-V host dispatch kernels to a consumer-grade AMD card is the least-effort way around this ecosystem gap—ROCm is already maintained on RISC-V; only the compiler backend link is missing.

6.4 VIR’s fixes shift from “computing wrong” to “failing silently”

Of buddy-mlir’s six commits in this window, at least three target “no error but wrong result”: operators silently dropped across multiple set_vl regions, scale factor parsing crashing or letting illegal values through, and cross-block reduction order changing numerical values. The first two categories previously let the compiler quietly produce a wrong result; now they become compile-time errors. The third can only be caught by byte-by-byte comparison.

What these problems share is low fix cost and high discovery cost. That they can be cleared in batches at this stage shows VIR already has a test surface that can run on real hardware and compare against baselines—without the comparisons on K3 and SG2044, differences like the VLEN 128 versus 256 in 4.4 would never surface.


Source Verification Table

Source Verification Result
RuyiAI-Stack key repos triton-riscv: one merge (#110 GPU backend enabled by default), one under review (#109 plugin rename), nightly 2026-11-10-209274f predates the merge; pytorch, tilelang-riscv, llvm-project: zero commits; remaining repos have no commits in the window
ruyisdk org ruyi: one formal release 0.54.0 (11-10 21:15), zero commits to the main repo, three fix PRs still stalled at 11-04; ruyi-imager two dependency upgrades; packages-index zero commits; remaining repos no updates
openRuyi-Project packaging repo eight commits (four security fixes, llvm-snapshot and clang-wrap bumped to 20261109, libssh and SDL2 Wayland); release spec repo one commit; firmware repo zero commits; tags and news page stalled at 2026.09
buddy-compiler buddy-mlir six merges (#1042 to #1047); nightly stalled at 0.1.4.dev20261105; ten open PRs, most recently updated #1016 and #1019
RuyiSDK Forum no new topics in the window; latest is still the 11-06 repost; biweekly progress latest is issue 080 (11-05)
RuyiAI official site index, blog, and docs pages all unchanged; 2 blog posts, 23 documentation chapters
ruyiai-daily archive stalled at the 11-08 issue; the 11-09 to 11-11 issues were not generated
Site repo one contributor index commit (11-09 23:57 UTC)
Google News five targeted queries returned zero hits; general RISC-V crawls were all irrelevant items
arXiv no new papers in the window; the latest is stalled at 11-03

Complete Source List