feat(strata): bump 0.2.13 -> 0.2.15

Strata 0.2.14 runs the prompt on its own kernels and vendors llama.cpp
(third_party/strata-llama), so the separate llama.cpp fetch for the
vision encoder goes. The build keeps upstream's halo flags: MMQ stays
out (STRATA_MMQ defaults to OFF), 47 of 48 UD-Q5_K_XL layers run on the
fused kernels, and the ROCm 10.1 hipBLASLt table still ships.

UD-Q5_K_XL on ms-s1-max, 16K cold prompts: 715 -> ~810 tok/s on the
defaults, ~1180 tok/s with the gfx1151 prompt set (unchanged).

Bead: nix-overlay-ydg
This commit is contained in:
2026-11-16 20:51:14 +03:00
parent d14f6d5e37
commit b3287c926f
3 changed files with 11 additions and 13 deletions
+4 -10
View File
@@ -10,13 +10,6 @@
therockRocm,
}:
let
# The vision encoder needs the full llama.cpp tree at Strata's pinned ggml revision.
llama = fetchFromGitHub {
owner = "ggml-org";
repo = "llama.cpp";
rev = "3cf03257f219afbe7334045ff7c6a06ac68c627d";
hash = "sha256-SRGoXa+4ACBCB3eaG9XFYhMN1i0FyPEy9Rrer+dFGYI=";
};
python = python3.withPackages (
ps: with ps; [
jinja2
@@ -40,13 +33,13 @@ let
in
stdenv.mkDerivation (finalAttrs: {
pname = "strata";
version = "0.2.13";
version = "0.2.15";
src = fetchFromGitHub {
owner = "Niko1221";
repo = "Strata";
tag = "v${finalAttrs.version}";
hash = "sha256-ygLtDYyrwfRKez46xKnPkO7BtY0Q4uMwxlvh3JyUHbE=";
hash = "sha256-ehbQHpljUeclywp0ubGcrGG+xQ35SppRoosVWZLJiDY=";
};
nativeBuildInputs = [
@@ -88,9 +81,10 @@ stdenv.mkDerivation (finalAttrs: {
cp ${./gfx1151-hipblaslt-100401.txt} tools/hip/gfx1151-hipblaslt-100401.txt
'';
# Vision builds from the vendored third_party/strata-llama (LLAMA_DIR default).
postBuild = ''
cmake -S ../tools/vision -B ../build-vision -G Ninja \
-DCMAKE_BUILD_TYPE=Release -DLLAMA_DIR=${llama} \
-DCMAKE_BUILD_TYPE=Release \
-DSTRATA_VISION_CUDA=OFF -DSTRATA_VISION_HIP=OFF -DSTRATA_PORTABLE=ON
cmake --build ../build-vision --target strata-vision -j $NIX_BUILD_CORES
'';