dev-python/nvidia-cutlass-dsl
NVIDIA CUTLASS Python DSL — metapackage for libs-base + libs-cu13
Screenshots
View
Download
BrowseChangeLog
commit 3d968d5ed9f1520849964d1bd35dd97ee0ced10c
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Mon May 25 19:11:39 2026 +0200
dev-util/rtk: new package, 0.42.0
Rust Token Killer — CLI proxy that reduces LLM token consumption by
filtering common dev-command output before it reaches the LLM
session. Single Rust binary, MIT/Apache-2.0 dual-licensed (LICENSE
file is Apache-2.0; Cargo.toml claims MIT — go with the LICENSE file
per Gentoo policy).
Source build via cargo eclass with all 202 transitive crates inlined
in CRATES from the upstream Cargo.lock. rusqlite uses its bundled
SQLite, flate2 uses miniz_oxide, so no system C deps. Build-verified
against ~amd64; RUST_MIN_VER bumped to 1.87.0 to satisfy the strictest
dep in the transitive set (caught at cargo eclass QA).
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 4eefdc980cfb2816d5b654c1a94fd65451cf2e55
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Mon May 25 14:11:40 2026 +0200
dev-python/nvidia-cutlass-dsl: add 4.5.2
Metapackage patch bump from 4.5.1 — SRC_URI hash path refreshed for
the py3-none-any metawheel via PyPI JSON API. RDEPEND tracks the
4.5.2 sibling libs-base and libs-cu13 ebuilds via the existing ~PV
pin. Trivial install, build-verified.
commit 638c64841a6b0d206ae9715f4d2bcb6095046003
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun May 17 00:54:12 2026 +0200
dev-python/nvidia-cutlass-dsl: drop 4.5.0
Retention: keep 4.4.2 (upstream vllm 0.21.0 cuda.txt exact pin)
and 4.5.1 (current). 4.5.0 is intermediate within the 4.5.x series.
commit ac81bd18b6c79dbc83fece6ae232121000f81a18
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun May 17 00:52:23 2026 +0200
dev-python/nvidia-cutlass-dsl: add 4.4.2
Upstream vllm 0.21.0 pins this version exactly; without it vllm's
cuda? branch can't solve. Keep 4.5.0/4.5.1 alongside for newer
consumers.
commit d946180127a1c407bdc67daa9491dd9e912cb36d
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sat May 16 10:23:00 2026 +0200
dev-python/nvidia-cutlass-dsl: add 4.5.1
PyPI metapackage hash refresh for the umbrella wheel. Pulls libs-base and (under
cu13 extra) libs-cu13, both bumped.
commit 846ead5b7003409b582023630303d6b47a74a3c6
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed May 13 14:35:00 2026 +0200
dev-python/nvidia-cutlass-dsl: disable py3.11
commit 30530de06924aa0a16f1bdfe9c6f5e4621c49841
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun May 10 00:15:22 2026 +0200
dev-python/nvidia-cutlass-dsl: fix dead docs URL in HOMEPAGE
NVIDIA reorganized the CUTLASS docs site: media/docs/python/ moved to
media/docs/pythonDSL/ under /latest/. Point HOMEPAGE at the stable
/cutlass/ root which auto-redirects to /latest/, so future URL shuffles
inside the docs hierarchy don't recur as DeadUrl findings.
commit 4404d36e95038d697674616d15deabfe5a8bb2bc
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu May 7 17:41:34 2026 +0200
dev-python/nvidia-cutlass-dsl: new package, 4.5.0
Tier 3 of the vllm CUDA target packaging cycle. Trivial metapackage
that pulls libs-base + libs-cu13 as a unit. Upstream's PyPI release
is a single empty py3-none-any wheel with `requires_dist` listing
the two libs sub-packages. No source release.
Always pulls the cu13 path on this overlay (amd64 + CUDA 13.2 at
/opt/cuda); a cu12 path would need a libs-cu12 sibling we haven't
packaged. Adds a USE flag only if a cu12 user emerges.
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Mon May 25 19:11:39 2026 +0200
dev-util/rtk: new package, 0.42.0
Rust Token Killer — CLI proxy that reduces LLM token consumption by
filtering common dev-command output before it reaches the LLM
session. Single Rust binary, MIT/Apache-2.0 dual-licensed (LICENSE
file is Apache-2.0; Cargo.toml claims MIT — go with the LICENSE file
per Gentoo policy).
Source build via cargo eclass with all 202 transitive crates inlined
in CRATES from the upstream Cargo.lock. rusqlite uses its bundled
SQLite, flate2 uses miniz_oxide, so no system C deps. Build-verified
against ~amd64; RUST_MIN_VER bumped to 1.87.0 to satisfy the strictest
dep in the transitive set (caught at cargo eclass QA).
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 4eefdc980cfb2816d5b654c1a94fd65451cf2e55
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Mon May 25 14:11:40 2026 +0200
dev-python/nvidia-cutlass-dsl: add 4.5.2
Metapackage patch bump from 4.5.1 — SRC_URI hash path refreshed for
the py3-none-any metawheel via PyPI JSON API. RDEPEND tracks the
4.5.2 sibling libs-base and libs-cu13 ebuilds via the existing ~PV
pin. Trivial install, build-verified.
commit 638c64841a6b0d206ae9715f4d2bcb6095046003
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun May 17 00:54:12 2026 +0200
dev-python/nvidia-cutlass-dsl: drop 4.5.0
Retention: keep 4.4.2 (upstream vllm 0.21.0 cuda.txt exact pin)
and 4.5.1 (current). 4.5.0 is intermediate within the 4.5.x series.
commit ac81bd18b6c79dbc83fece6ae232121000f81a18
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun May 17 00:52:23 2026 +0200
dev-python/nvidia-cutlass-dsl: add 4.4.2
Upstream vllm 0.21.0 pins this version exactly; without it vllm's
cuda? branch can't solve. Keep 4.5.0/4.5.1 alongside for newer
consumers.
commit d946180127a1c407bdc67daa9491dd9e912cb36d
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sat May 16 10:23:00 2026 +0200
dev-python/nvidia-cutlass-dsl: add 4.5.1
PyPI metapackage hash refresh for the umbrella wheel. Pulls libs-base and (under
cu13 extra) libs-cu13, both bumped.
commit 846ead5b7003409b582023630303d6b47a74a3c6
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed May 13 14:35:00 2026 +0200
dev-python/nvidia-cutlass-dsl: disable py3.11
commit 30530de06924aa0a16f1bdfe9c6f5e4621c49841
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun May 10 00:15:22 2026 +0200
dev-python/nvidia-cutlass-dsl: fix dead docs URL in HOMEPAGE
NVIDIA reorganized the CUTLASS docs site: media/docs/python/ moved to
media/docs/pythonDSL/ under /latest/. Point HOMEPAGE at the stable
/cutlass/ root which auto-redirects to /latest/, so future URL shuffles
inside the docs hierarchy don't recur as DeadUrl findings.
commit 4404d36e95038d697674616d15deabfe5a8bb2bc
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu May 7 17:41:34 2026 +0200
dev-python/nvidia-cutlass-dsl: new package, 4.5.0
Tier 3 of the vllm CUDA target packaging cycle. Trivial metapackage
that pulls libs-base + libs-cu13 as a unit. Upstream's PyPI release
is a single empty py3-none-any wheel with `requires_dist` listing
the two libs sub-packages. No source release.
Always pulls the cu13 path on this overlay (amd64 + CUDA 13.2 at
/opt/cuda); a cu12 path would need a libs-cu12 sibling we haven't
packaged. Adds a USE flag only if a cu12 user emerges.