julian/amd-strix-halo-toolboxes
1
0
Fork 0
Code Issues Pull Requests 2 Actions Packages Projects Releases Wiki Activity

chore: update llama.cpp patch to PR 21566 for gemma-4 inference fix

Browse Source
This commit is contained in:
Donato Capitella
2026-04-07 17:49:16 +01:00
parent d0281bb526
commit a58d133c5e
1 changed files with 2 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
toolboxes/Dockerfile.rocm7-nightlies
+2 -1
View File
@@ -61,8 +61,9 @@ RUN git clone -b ${BRANCH} --single-branch --recursive ${REPO} . \
COPY llama-grammar.patch /tmp/llama-grammar.patch COPY llama-grammar.patch /tmp/llama-grammar.patch
# PR 21566: test a fix for gemma-4 inference bug (CUDA: check for buffer overlap before fusing)
RUN patch -p1 < /tmp/llama-grammar.patch \ RUN patch -p1 < /tmp/llama-grammar.patch \
&& curl -sSL https://github.com/ggml-org/llama.cpp/pull/21506.patch | patch -p1 \ && curl -sSL https://github.com/ggml-org/llama.cpp/pull/21566.patch | patch -p1 \
&& cmake -S . -B build \ && cmake -S . -B build \
-DGGML_HIP=ON \ -DGGML_HIP=ON \
-DAMDGPU_TARGETS=gfx1151 \ -DAMDGPU_TARGETS=gfx1151 \