commit efa6a3eca0144acac701909d63b2afd0d32c40d5
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Tue Oct 27 01:21:34 2026 +0100

sci-ml/fastflowlm: drop 1.0.7-r1

Superseded; 1.0.8, 1.0.9 and 1.0.10 stay as the last three releases,
and nothing pins an older one.

commit ae0f9cd81d7003ffb7aedbcaae1bec7bb729aef0
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun Oct 25 07:35:19 2026 +0100

sci-ml/fastflowlm: add 1.0.10

1.0.10 adds xgrammar as a submodule for constrained decoding; its GitHub
archive omits the dlpack submodule whose headers it compiles against, so
both are fetched at their pinned commits.

FFmpeg, FFTW and GNU readline are gone: audio decoding uses vendored
dr_libs plus opus/vorbis, FFTs use a vendored pocketfft, and line
editing uses libedit, since upstream rules out linking GPL-3 readline.
Upstream also downloads and statically builds libjpeg-turbo, opus, ogg
and vorbis through ExternalProject at configure time; a patch links the
system libraries through pkg-config instead.

FLM_ENABLE_RAI is renamed to FLM_ENABLE_CORE; the old name is now
ignored, so the corelib backend is disabled under the new one.

The codec patch calls find_package(PkgConfig REQUIRED), so pkg-config
is a build dependency.

commit 5ff448cb7d6fc6b78e0efa4d3128cb5c9f4652ae
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri Oct 16 12:45:42 2026 +0200

sci-ml/fastflowlm: drop 1.0.6-r1

Superseded; the last three releases stay. It was the last ebuild
installing the plain-named init and conf.d files; the rest use the
1.0.7 ones.

commit 02183cd0d92d4442fbe2076e61db91d4c8f19efb
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri Oct 16 12:40:17 2026 +0200

sci-ml/fastflowlm: move 9999 to the NPU binary licence

Upstream main ships the NPU kernels under LICENSE_BINARY.txt since
1.0.9; TERMS.md is gone. Match the 1.0.9 ebuild: FastFlowLM-NPU-Binary
and no bindist restriction.

commit d02638c8205d864d7c9f9171df2705a6c72bea64
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri Oct 16 10:12:41 2026 +0200

sci-ml/fastflowlm: add 1.0.9

Upstream relicensed the NPU kernels and engine libraries: TERMS.md,
with its revenue cap and no grant to redistribute, gave way to
LICENSE_BINARY.txt, which allows any use and unmodified redistribution.
Ship that text as FastFlowLM-NPU-Binary and lift bindist and mirror.

The HRX runtime pin moved to the 2026.10.13 release.

commit 77b7033af123c79ce5febae78852fc145a2cf72f
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Tue Oct 13 13:20:28 2026 +0200

sci-ml/fastflowlm: add USE=hrx to 1.0.8 and 9999

Upstream's HRX backend drives the NPU through libhrx and the amdxdna
kernel driver, without XRT. The flag swaps the XRT dependencies for
the HRX release this version pins and links the HRX-built engines.

Upstream's HRX install copies flm's whole shared-library closure,
system libraries included, into the prefix; the ebuild removes those
copies and symlinks the packaged libhrx where the RUNPATHs look.

commit 3b47dd5f6c293bce207ba2f9c3fbdb832aa5c877
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Tue Oct 13 11:54:42 2026 +0200

sci-ml/fastflowlm: follow 1.0.8's build options in 9999

main is 1.0.8 plus documentation commits, so the live ebuild gains
USE=gemma4e-source with its OpenMP check, the <climits> fix and
-fno-strict-aliasing, and pins the HRX, ryzenai-corelib and
native-arch options explicitly.

ryzenai-corelib stays off for more than packaging reasons: the
library is not public, and a build that links it switches the model
catalogue to an unreleased NPU generation and offers none of the
current models.

commit 59cf4fa832a8ffaf89157e124029ce383816ca42
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Tue Oct 13 11:33:26 2026 +0200

sci-ml/fastflowlm: drop 1.0.5-r1

Superseded by three newer releases.

commit 8a173aa3e18edd5034ca05ba708ea26a2dc94e39
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Tue Oct 13 11:33:25 2026 +0200

sci-ml/fastflowlm: add 1.0.8

Adds the minicpm-v-4.7:1b model. Upstream also opened the source of
the Gemma 4 E2B/E4B engine; the new USE=gemma4e-source builds it in
place of the prebuilt libgemma4e_npu.so, and stays off by default as
upstream's option does. That build needs OpenMP, a <climits> include
missing from npu_cmd.hpp, and -fno-strict-aliasing: the engine
source type-puns through pointers.

The new ryzenai-corelib backend is pinned off: the library is not
public, and a build that links it switches the model catalogue to an
unreleased NPU generation and offers none of the current models.

The sed that removed a /usr/local symlink block is gone, as
CMakeLists.txt no longer has such a block.

commit 275900081902b4df25b8322481d76ae1f6d2d14d
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun Oct 11 09:45:21 2026 +0200

sci-ml/fastflowlm: add flm-custom-sync, revbump

A fine-tune converted to .q4nx runs on the NPU kernels of the model it
is derived from, but FastFlowLM only serves tags that its registry
lists, and the packaged registry and kernel folders belong to the
package manager.

flm-custom-sync keeps a per-user registry under ~/.config/flm: the
packaged model list plus the entries named in custom_models.json, and
kernel links for each custom model. FastFlowLM reads it through
FLM_CONFIG_PATH and FLM_XCLBIN_PATH. The generated list is a copy, so
it has to be regenerated after each upgrade; pkg_postinst says so.

The helper refuses what would let FastFlowLM damage or misread a
model. A custom tag may not reuse a packaged family or folder name:
FastFlowLM checks packaged tags against stored hashes and removes
files that differ. A custom model whose weights are older than its
base model now requires is left out of the list, because FastFlowLM
1.0.7 finds nothing to download for such a tag, treats it as complete
and loads the old weights on kernels built for the new layout. A
model with files missing from its folder is left out as well.

The OpenRC init script exports none of its settings, so
FLM_CONFIG_PATH and FLM_XCLBIN_PATH set in conf.d would never reach
the daemon. This revision installs an init script and conf.d file
that pass them, and FLM_MODEL_PATH, through; they carry a version in
their names because the older ebuilds keep the files they were
published with. The systemd unit already takes all three from its
environment file.

The script uses the standard library only; Python becomes a runtime
dependency for it.

commit c65ea7829e8126ad1c0a33b649c638b7d7e90208
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Oct 8 00:39:17 2026 +0200

sci-ml/fastflowlm: pin the tokenizers-cpp crates with a Cargo.lock

tokenizers-cpp ships no Cargo.lock, so each build resolved its semver
ranges afresh and the crate set could change between two installs of
the same ebuild. Ship a lockfile for the pinned tokenizers-cpp commit
(80 crates, tokenizers 0.21.4) and build with cargo --locked.

commit b4ba7fa2e9b689e947c618edc9cac83f5de64f97
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Tue Oct 6 13:13:08 2026 +0200

sci-ml/fastflowlm: fix the category in the 9999 rebuild hint

The pkg_postinst message has told users to emerge
=dev-ml/fastflowlm-9999 since the package was added; the package
lives in sci-ml. Message text only, so no revbump.

commit abaf7168043652ab5fa85997d7aa20c2d12654ee
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Mon Oct 5 12:51:10 2026 +0200

sci-ml/fastflowlm: record upstream's current terms for the NPU kernels

Every version packaged here ships the November 2025 binary licence
under assets/superseded/, and TERMS.md now governs the kernels: free
to use, commercially only up to USD 10 million of annual company
revenue, with no grant to redistribute, so bindist and the mirror
restriction stay. Add those terms as FastFlowLM-Terms and drop the
superseded FastFlowLM-Binary, which nothing else uses; anyone who
accepted the old licence has to accept the new terms.

commit 740e79d5c7263e1eda5a0219b1330dc95a633917
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Mon Oct 5 10:13:12 2026 +0200

sci-ml/fastflowlm: link XRT's aiebu instead of the bundled copy

Upstream ships a prebuilt libaiebu.a and its headers in src/lib/xrt
and src/include/aiebu, and its link path puts them ahead of the
system's. flm calls only aiebu_assembler_get_elf(), which the aiebu of
every packaged XRT declares identically, so remove the bundled copy
and build against XRT's, requiring the first xrt that installs the
headers. Only flm's statically linked aiebu code changes, through the
same API, so existing installs need not rebuild.

commit 5bfb2199e43a92b7aac8c85447241820b4411557
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun Oct 4 13:52:41 2026 +0200

sci-ml/fastflowlm: restrict binary packages

The FastFlowLM binary licence forbids redistributing the binaries to
any third party without written permission, and a binary package
built from these ebuilds would do that. Add bindist next to the
existing mirror restriction.

commit 4d5cbbdb1b7df01ebef77132320ea5ef61d45860
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Oct 1 08:58:54 2026 +0200

sci-ml/fastflowlm: drop 1.0.4-r1

Retention keeps the last three releases.

commit 568ce3ac45b8b2b78c54dcb4ffde7844d151461d
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Oct 1 08:58:54 2026 +0200

sci-ml/fastflowlm: add 1.0.7

Upstream removed the stray libq4_npu_eXpress.so.bak-20260826 that its
*.so* install glob used to catch, so the rm that excluded it goes. The
release adds a qwen3_8mtp NPU model; the tokenizers-cpp pin, and with
it the nested sentencepiece and msgpack-c commits, and NPU_VERSION are
unchanged.

commit b1b13c701d3cfdba01e9e2f0551d3c018aeec5b4
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Mon Sep 21 11:22:10 2026 +0200

sci-ml/fastflowlm: drop 0.9.46-r1 1.0.3-r1

This removes the last 0.9.x release, so no 0.9 rollback anchor remains.
Nothing in the tree pins either version: lemonade depends on the bare
atom.

commit 84999eb275cd91215504bbf0e559ba7b58c8bce4
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Mon Sep 21 11:18:01 2026 +0200

sci-ml/fastflowlm: drop 0.9.46 1.0.3 1.0.4 1.0.5 1.0.6

Superseded by their -r1 revisions. The helper script goes with them, as
nothing references it any more.

commit 95ab1c7876b4351a6bbe0b9f69cc2aec8c176a22
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Mon Sep 21 11:17:36 2026 +0200

sci-ml/fastflowlm: drop the Whisper config helper

flm-patch-whisper worked around FastFlowLM#545, where LM_Config asserted
on decoder-only fields that a HuggingFace Whisper config.json does not
carry. No version this overlay ships has that assert any more, and
Whisper_Config maps d_model and the ffn width to the canonical keys
itself. On 1.0.6 a fresh "flm pull whisper-v3:turbo", then list and
"serve --asr 1", all run on a config the helper never touched, and the
same audio clip transcribes byte-identically either way.

The revision bump lets portage remove the installed helper when the new
revision replaces the old one. The live ebuild is edited in place.

commit 8187dbc43aca47fa2f3a3de62a512ed1a485fdfb
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sat Sep 19 15:53:01 2026 +0200

sci-ml/fastflowlm: add 1.0.6

commit acdbd5a123371d5fee9c0ca0eb8996ce4e281fe3
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri Sep 11 10:46:10 2026 +0200

sci-ml/fastflowlm: trim 1.0.4 comments

commit cbd87aeecdf2d1be57ec49069b2acd682d0fb0c6
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri Sep 11 10:46:10 2026 +0200

sci-ml/fastflowlm: trim 1.0.3 comments

commit 87c41471af93ec15db1a75a13d28fc83924ce64c
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri Sep 11 10:46:10 2026 +0200

sci-ml/fastflowlm: trim 0.9.46 comments

commit 4505b0f35af9e174519e8b5a2d34b36994111adb
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri Sep 11 00:50:28 2026 +0200

sci-ml/fastflowlm: trim 1.0.5 comments

commit f84736a04767a1fa90d0b8172879e0fb4b493906
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Sep 10 05:41:59 2026 +0200

sci-ml/fastflowlm: trim live comment prose

commit 9ba721f5bc65202d439c54d482f680b9783eeed1
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Sep 10 23:59:37 2026 +0200

sci-ml/fastflowlm: fix libdir in live ebuild

The previous ebuild behavior no longer matches the supported upstream
build or runtime interface.

commit 2bbd352b1f7bb32373ac5bfed6d1d1fea189d990
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Sep 10 23:59:20 2026 +0200

sci-ml/fastflowlm: backport libdir fix to 0.9.46

Retained releases must locate the installed backend through Gentoo
multilib paths instead of assuming usr/lib.

commit 78f2f44fd82636de6bac3fc4ef436a6ea337eacc
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Sep 10 23:58:59 2026 +0200

sci-ml/fastflowlm: backport libdir fix to 1.0.3

Retained releases must locate the installed backend through Gentoo
multilib paths instead of assuming usr/lib.

commit 053b0cd35ecb9c540dc25ff3f7c86078f0593133
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Sep 10 23:58:35 2026 +0200

sci-ml/fastflowlm: backport libdir fixes to 1.0.4

Retained releases must locate the installed backend through Gentoo
multilib paths instead of assuming usr/lib.

commit ff9d7318fb0828a9f8a9f7cebd33c94b539e7f68
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Sep 10 23:11:34 2026 +0200

sci-ml/fastflowlm: add 1.0.5

commit b104e83c659a81589af157648e691d833324205e
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Sep 10 23:02:55 2026 +0200

sci-ml/fastflowlm: restrict mirroring 1.0.4

The upstream artifact is not suitable for redistribution through
Gentoo mirrors, so retained releases need the same restriction as
the current ebuild.

commit fdaa5867c2e52f397c9f9179dff25a1772e1a37c
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Sep 10 23:01:54 2026 +0200

sci-ml/fastflowlm: restrict mirroring 1.0.3

The upstream artifact is not suitable for redistribution through
Gentoo mirrors, so retained releases need the same restriction as
the current ebuild.

commit cb3fdbb2b762a0c2969f5b9277d159b42bd20907
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Sep 10 23:01:44 2026 +0200

sci-ml/fastflowlm: restrict mirroring 0.9.46

The upstream artifact is not suitable for redistribution through
Gentoo mirrors, so retained releases need the same restriction as
the current ebuild.

commit cb37d4764605f915afc3853e32be004edbac85f1
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Sep 10 23:00:40 2026 +0200

sci-ml/fastflowlm: fix live property in 1.0.4

Versioned releases must remain ordinary fetchable packages; only the
live ebuild may carry live-package semantics.

commit ebd5e4e719c512680532db9f3b6640dfa98c1fe2
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Sep 10 23:00:30 2026 +0200

sci-ml/fastflowlm: fix live property in 1.0.3

Versioned releases must remain ordinary fetchable packages; only the
live ebuild may carry live-package semantics.

commit ab971c0bfebbf62cad71feb08e7b1c268ad73f9a
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Sep 10 23:00:23 2026 +0200

sci-ml/fastflowlm: fix live property in 0.9.46

Versioned releases must remain ordinary fetchable packages; only the
live ebuild may carry live-package semantics.

commit 4c3fcc8e966729edd139862fa42cd234da69f5de
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed Sep 2 09:25:39 2026 +0200

sci-ml/fastflowlm: drop 1.0.1, 1.0.2

Retention: 0.9.46 stays as the last of the 0.x line, 1.0.3 and 1.0.4
are the newest two. Both dropped ebuilds differ from what remains only
in dated verification comments and in predating the 1.0.3
requantisation ewarn -- no unique build or runtime capability goes with
them.

commit abacde61ded00c155e7543cf8aa4de61a5aedf4c
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed Sep 2 09:25:38 2026 +0200

sci-ml/fastflowlm: add 1.0.4

Deps unchanged. Re-checked the two things this ebuild's own convention
says to re-check on every bump, rather than carrying them forward
blind:

The tokenizers-cpp gitlink is still acbdc5a at v1.0.4, identical to
v1.0.3, so TOKENIZERS_CPP_COMMIT stands and the nested sentencepiece
and msgpack pins inside it are unmoved as well.

The CMake build interface is unchanged apart from one added target:
src/CMakeLists.txt gains gemma4_12b_npu and nothing else, with the
option() and find_package() surfaces byte-identical to 1.0.3. So no
dependency moves.

Build-verified on haarmek, and checked the delta actually landed rather
than just that the build was green: libgemma4_12b_npu.so is present in
the image alongside the other 22 model libraries, and bin/flm installs.
The 1.0.3 Qwen3.5/Qwen3.6-MoE requantisation ewarn carries forward
unchanged. verified 2026-09-02

commit 0607ece91488c8857fcbc7e9975bc98cfa66ae37
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Aug 27 09:31:40 2026 +0200

sci-ml/fastflowlm: add 1.0.3

commit cd953202ae773c82c8ca4d561c6489c3bd5d8505
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Aug 20 10:20:46 2026 +0200

sci-ml/fastflowlm: drop 1.0.0

commit df9b762d1efdde03614b2c5b3698a72e506d19b1
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Aug 20 10:20:40 2026 +0200

sci-ml/fastflowlm: add 1.0.2

commit aee606d3a536b9d1be5692df5e52a9115fdbcfff
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun Aug 16 16:08:01 2026 +0200

sci-ml/fastflowlm: add ncurses dependency

The non-portable Linux build requires the ncursesw pkg-config module and links it
into the runtime. Declare ncurses explicitly rather than relying on an incidental
installation.

commit b12547123095b3931491eb572ee8cddf14a99c1c
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri Aug 14 21:12:09 2026 +0200

sci-ml/fastflowlm: follow ROCm repository move

The upstream repository moved to the ROCm organization. Update package metadata and
all user-facing issue references to the canonical location.

commit 3a6762b582a565d4b956f98937736bf406df7058
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri Aug 14 09:41:47 2026 +0200

sci-ml/fastflowlm: normalize copyright header

Use the overlay-wide required copyright range consistently across retained and live
ebuilds.

commit 1c3576145edb355c281bcd3f41a6a9e4dcbe923c
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed Aug 12 09:19:44 2026 +0200

sci-ml/fastflowlm: add 1.0.1

commit 63dc95edace90a9c75eb23f818608bb1e80015e0
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Tue Aug 11 23:50:50 2026 +0200

sci-ml/fastflowlm: drop 0.9.45

commit 5895ed7da8ef2d0ec1edc42406face6d4db95117
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Tue Aug 11 23:50:50 2026 +0200

sci-ml/fastflowlm: add 1.0.0

commit 27299d71b68bc415d0fff013d4013791f4df8628
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed Jul 29 10:40:19 2026 +0200

sci-ml/fastflowlm: drop 0.9.44

commit 856b624dd7247ad35e194233d6680c4474d9dae8
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed Jul 29 10:40:18 2026 +0200

sci-ml/fastflowlm: add 0.9.46

commit 30692f441f189400c4c7f45b1d200d4ea5c4fffc
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed Jul 29 10:37:43 2026 +0200

sci-ml/fastflowlm: switch upstream to the ROCm org

AMD adopted FastFlowLM into the ROCm organisation, and ROCm/FastFlowLM is
now the canonical home where newer releases land. The vX.Y.Z tags mirror
the original repo commit-for-commit, so the release tarballs are
byte-identical and the Manifest is unchanged.

Retarget SRC_URI and HOMEPAGE on the release ebuild and EGIT_REPO_URI on
the live ebuild. The #545 issue references stay on FastFlowLM/FastFlowLM,
which is still live and carries the issue history.

ROCm/FastFlowLM keeps FastFlowLM's own vX.Y.Z tags rather than the
org-wide rocm-X.Y.Z convention, so nvchecker's blanket ROCm-org filter
would match zero tags and silently drop tracking; a per-package override
pins the v-scheme, and the regenerated config tracks the new repo.

commit 669c42f25c1c649a406d2d2d951a64c5cf844341
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Mon Jul 27 21:02:36 2026 +0200

sci-ml/fastflowlm: restore the copyright header to line 1

A stray blank line preceded the header in the live ebuild, so the file
did not open with the copyright notice the header format requires.

The header checks are commit-scoped: they compare a file against the
commit that touches it, and never re-examine it afterwards. Once the
blank line had landed, no amount of full-tree scanning could surface
it again.

commit 263ac9de0f868422f19a12f154679b17914ef7f0
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sat Jul 11 09:13:36 2026 +0200

sci-ml/fastflowlm: drop 0.9.43-r1

commit c551e454e9dfa837e74384b741f053acb09bb7de
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sat Jul 11 09:13:35 2026 +0200

sci-ml/fastflowlm: add 0.9.45

commit 695e6cf6714b7059991443fc2d3f94e9982667cf
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Mon Jul 6 01:30:09 2026 +0200

sci-ml/fastflowlm: drop 0.9.42-r1

Retain the last two releases (0.9.43-r1, 0.9.44).

commit 0822fb1a08951be44589366b91189b0ac5178165
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Mon Jul 6 01:30:09 2026 +0200

sci-ml/fastflowlm: add 0.9.44

New upstream release. The tokenizers-cpp/sentencepiece/msgpack submodule
pins are unchanged from 0.9.43, and the CMakeLists install rules the
ebuild depends on are intact (re-verified against v0.9.44). #545 (Whisper
config crash) is not fixed upstream, so the workaround helper stays.

commit ec166869bde79f5db3c7f274f906572a56b28324
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sat Jun 6 20:55:49 2026 +0200

sci-ml/fastflowlm: drop 0.9.40-r1, 0.9.41-r1

commit 7c384c8cac70105c06770e0ccb7af0fc295f2471
Author: Raukaan Cogbrother <cogbrother@raukaan.local>
Date: Sun May 31 03:29:35 2026 +0200

metadata: normalize tab → 2-space indent in 42 imported files

42 metadata.xml files inherited tab indentation from their origin
overlays (mostly the ROCm cluster, plus a handful of dev-python and
AMD-tooling imports). The overlay convention is 2-space; this brings
them in line with the other 498 files. Maintainer attributions
preserved verbatim.

commit dc43643ab26b51b7d29843f8879f257ff10b76e3
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed May 27 18:20:35 2026 +0200

sci-ml/fastflowlm: set CMAKE_XCLBIN_PREFIX so the runtime finds xclbins

Upstream's linux-default cmake preset sets it to match the install
rule's share/flm DESTINATION; src_configure doesn't drive the build
via the preset, so without an explicit -D the macro was baked as ""
and the runtime fell back to dirname(/proc/self/exe), looking under
bin/xclbins/ instead of share/flm/xclbins/. #269.

commit 97b222aeced2669a0c84267c1de0b8a6574292e2
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed May 27 13:14:09 2026 +0200

sci-misc/llama-swap, sci-ml/fastflowlm: add openrc + systemd services

Both get supervise-daemon-driven openrc init.d/conf.d and a systemd
per-instance template (pkg@<user>.service). Runners are USE-gated
and default off; the existing manual-start flow stays the unchanged
baseline.

llama-swap openrc refuses to start until LLAMA_SWAP_USER is set;
auto-derives LLAMA_SWAP_CONFIG from that user's $HOME via getent if
unset. Listener defaults to 127.0.0.1:8080 so a fresh install doesn't
expose the LLM API to the LAN. systemd unit has no default
LLAMA_SWAP_CONFIG -- systemd's %h resolves to /root for system-
manager units and /home/%i would bake in a passwd layout we can't
promise -- so EnvironmentFile=/etc/default/llama-swap@%i is required.

fastflowlm openrc refuses to start until FLM_USER is set; rc_ulimit
'-l unlimited' (and LimitMEMLOCK=infinity in systemd) is needed
because flm mlocks NPU buffers. systemd ExecStart goes through
/bin/sh -c with $$ escapes so ${FLM_PORT:+--port "$FLM_PORT"}
parameter expansion runs in the shell -- systemd's variable parser
has no :+ semantics.

Hardening:
- llama-swap: ProtectSystem=full rather than =strict so backends
spawned by it (llama.cpp et al.) can still write to ~/.cache/.
- fastflowlm: deliberately omits ProtectKernelTunables (NPU power-
mode may touch /sys/) and MemoryDenyWriteExecute (XDNA path may
use JIT); revisit once empirically verified safe.

Service files live at files/<pkg>.service (no @) because pkgcheck
BannedCharacter rejects @ in files/* filenames; systemd_newunit's
target arg adds the @ at install.

commit 73ec11612adb4a09558b153fa200ad19050f992b
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed May 27 11:57:40 2026 +0200

sci-ml/fastflowlm: add 0.9.43

Only ebuild-level change: TOKENIZERS_CPP_COMMIT reverted from 0.9.42's
34885cf back to acbdc5a (the 0.9.41-and-earlier pin). nested
sentencepiece + msgpack pins at acbdc5a still match the existing
values, so MSGPACK_COMMIT and SENTENCEPIECE_COMMIT stay put.

src/CMakeLists.txt is byte-identical to 0.9.42, so the sed pattern
stripping the symlink-into-/usr/local/bin block and the
NPU_VERSION/FLM_VERSION cmake args still apply unchanged. Verification
stamp re-dated accordingly.

Upstream substance is mostly in the model code: a new
common/whisper/ subsystem (which links a libwhisper_npu .so into the
binary) and a Whisper_Config : LM_Config subclass land in 0.9.43.
src/include/utils/json_check.hpp -- the validator that triggered
the #545 Whisper crash -- was deleted, which may mean the workaround
in flm-patch-whisper / the ewarn is no longer necessary. Conservative
call: keep both for one release cycle; if no user reports needing
the patch we can drop the ewarn at the next bump.

Build-checked locally: 91/91 ninja targets clean against
=fastflowlm-0.9.43 (default USE; no IUSE flags).

commit b17a9c755ac32614184a0cf85eb1eb0f8c6a13d3
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Mon May 25 16:07:27 2026 +0200

sci-ml/fastflowlm: ship flm-patch-whisper helper (upstream #545)

flm crashes at startup on any invocation that loads whisper-v3:turbo
because LM_Config::from_pretrained asserts on decoder-only LM-shape
fields (hidden_size, intermediate_size, num_attention_heads,
num_hidden_layers, num_key_value_heads) that HuggingFace's Whisper
config.json doesn't carry — Whisper uses the encoder-decoder
equivalents (d_model, encoder_ffn_dim, encoder_attention_heads,
encoder_layers) instead.

Ship a small Python helper at /usr/bin/flm-patch-whisper that walks
~/.config/flm/models/* (honoring $FLM_MODEL_PATH), finds Whisper
configs missing the decoder-only fields, and adds them with values
mapped from the encoder-* keys; idempotent on repeat runs. A
pkg_postinst ewarn points users at it as the required follow-up
after `flm pull whisper-v3:turbo`. Drop both once upstream merges
a real fix for issue #545.

commit d7891d49aca438b2da5ea2a5aceaf7f7dcfe40fd
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu May 14 11:49:41 2026 +0200

sci-ml/fastflowlm: add 0.9.42

commit c79366016e1891d1f4f488e0357c8b134de7b160
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed May 13 14:15:10 2026 +0200

sci-ml/fastflowlm: drop 0.9.38, 0.9.39

commit 27ae6103d77b7d627bf17bd5517801eada91dcd7
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Mon May 11 21:11:45 2026 +0200

sci-ml/fastflowlm: add 0.9.38

Older release retained alongside the newer ones. Third-party submodule
pins (tokenizers-cpp, sentencepiece, msgpack) are identical to 0.9.39's,
so the corresponding distfile entries already exist in the Manifest;
only fastflowlm-0.9.38.tar.gz is newly tracked.

commit e2f3b66d00008de38b637d8d7da84335eb1061db
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Mon May 11 17:34:26 2026 +0200

sci-ml/fastflowlm: add missing direct RDEPENDs

The flm binary links directly against libav{format,codec,util},
libsw{resample,scale}, libcurl, libboost_program_options, libfftw3f,
and libreadline (per readelf -d on /opt/fastflowlm/bin/flm), none of
which were declared. Build and merge succeed on systems where these
libraries are commonly installed; a minimal box could break after
depclean with no dependency trail.

Adds ffmpeg, curl, boost, fftw:3.0, and readline (subslot operators
for ABI tracking) to all four ebuild versions. Matches the Arch
official package's depends list and the build-deps documented in
upstream docs/linux-getting-started.md. Metadata-only — same binary
contents, same versions, just honest dependency declaration.

commit 95cd21b4cb4c970cd0bd2f8361bcf4f55aef2c51
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun May 10 00:29:00 2026 +0200

sci-ml/fastflowlm: drop www. prefix from HOMEPAGE

fastflowlm.com 308-redirects www.fastflowlm.com to the bare apex. Strip
the www. prefix across all four ebuilds (0.9.39/40/41 + 9999) so
pkgcheck stops reporting RedirectedUrl.

commit 262574fdcdf7dc7ee8c0257a476a8003bdd42e27
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed May 6 19:28:35 2026 +0200

sci-ml/fastflowlm: fix BDEPEND, sync -9999 with tagged ebuilds

- Replace nonexistent virtual/rust with || ( dev-lang/rust
dev-lang/rust-bin ) on 0.9.39, 0.9.40, 0.9.41.
- Add the same rust BDEPEND to -9999 (cargo via tokenizers-cpp/rust
is fetched and built on the live ebuild too).
- Add PROPERTIES=live + RESTRICT=network-sandbox to -9999 — same
cargo-fetch-online behavior that motivates them on the tagged
ebuilds; without RESTRICT, the build fails under sandbox.
- Carry the memlock/limits.d hint into -9999's pkg_postinst so the
NPU stack guidance is consistent across release and live.

commit 78b5135ded7e76a3cb7cf92b257b3334dc618063
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed May 6 19:15:28 2026 +0200

sci-ml/fastflowlm: add 0.9.40

Same submodule pin set as 0.9.39 (tokenizers-cpp acbdc5a, plus
unchanged sentencepiece + msgpack); ebuild byte-identical except
for filename.

commit d518de1486ae9354e400597aba4f888bf9c8ef6b
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed May 6 19:13:36 2026 +0200

sci-ml/fastflowlm: new package, add 9999 + 0.9.39 + 0.9.41

NPU-first LLM runtime for AMD Ryzen AI XDNA2 processors.

Three ebuilds: live -9999 tracks upstream main (EGIT_SUBMODULES='*'),
0.9.39 + 0.9.41 are tagged releases with submodule commit pins
(tokenizers-cpp + nested sentencepiece + msgpack-c) baked into SRC_URI.
0.9.41 bumps tokenizers-cpp from acbdc5a to 34885cf; the nested pins
are unchanged.

Build keeps PROPERTIES=live + RESTRICT=network-sandbox because cargo
inside tokenizers-cpp/rust fetches crates online. src_prepare strips
upstream's symlink-into-/usr/local/bin block; we wrap flm via newbin
and env.d to set LD_LIBRARY_PATH for XRT + FLM_CONFIG_PATH defaults.

LICENSE='MIT FastFlowLM-Binary': orchestration MIT (LICENSE_RUNTIME),
NPU compute kernels proprietary (LICENSE_BINARY, free for commercial
use under USD 10M annual revenue).

RDEPEND on dev-util/xrt + dev-libs/{xdna-driver,xrt-xdna} not yet
present in this overlay or ::gentoo; ride along ::guru's xrt-2.21.75
stack for now (cascade in a follow-up commit).