patch qwen3.5 fix

Dockerfile

@@ -33,7 +33,7 @@ WORKDIR /home/user
 RUN mkdir -p /home/user/code/models && \
     mkdir -p /home/user/code/app/wwwroot && \
     cd /home/user/code/models && \
-    wget -q https://huggingface.co/Mungert/Qwen3-4B-Thinking-2507-GGUF/resolve/main/Qwen3-4B-Thinking-2507-q4_k_m.gguf
+    wget -4 --tries=12 --waitretry=2 --retry-connrefused --timeout=30 --no-verbose https://huggingface.co/Mungert/Qwen3-4B-Thinking-2507-GGUF/resolve/main/Qwen3-4B-Thinking-2507-q4_k_m.gguf
 
 
 # Clone and build OpenBLAS as the non-root user
@@ -50,13 +50,8 @@ RUN cd /home/user/code/models/OpenBLAS && \
 # Switch back to the non-root user
 USER user
 
-# Clone and build llama.cpp with OpenBLAS support as the non-root user
-RUN git clone https://github.com/ggerganov/llama.cpp /home/user/code/models/llama.cpp && \
-    cd /home/user/code/models/llama.cpp && \
-    export PKG_CONFIG_PATH=/opt/OpenBLAS/lib/pkgconfig:$PKG_CONFIG_PATH && \
-    cmake -B build -DGGML_BLAS=ON -DGGML_BLAS_VENDOR=OpenBLAS -DBLAS_INCLUDE_DIRS=/home/user/code/models/OpenBLAS -DLLAMA_CURL=OFF && \
-    cmake --build build --config Release -j2 && \
-    cp /home/user/code/models/llama.cpp/build/bin/* /home/user/code/models/llama.cpp/
+# Clone llama.cpp as the non-root user
+RUN git clone https://github.com/ggerganov/llama.cpp /home/user/code/models/llama.cpp
 
 
 
@@ -78,6 +73,17 @@ RUN --mount=type=secret,id=GITHUB_TOKEN,mode=0444,required=true \
     git clone https://x-access-token:$(cat /run/secrets/GITHUB_TOKEN)@github.com/Mungert69/NetworkMonitorLLM.git /home/user/code/NetworkMonitorLLM && \
     git clone https://x-access-token:$(cat /run/secrets/GITHUB_TOKEN)@github.com/Mungert69/NetworkMonitorData.git /home/user/code/NetworkMonitorData
 
+# Apply llama.cpp patches from NetworkMonitorLLM
+RUN git -C /home/user/code/models/llama.cpp apply \
+        /home/user/code/NetworkMonitorLLM/patches/qwen35_session_tokens_insert_fix.diff \
+        /home/user/code/NetworkMonitorLLM/patches/qwen35_mrope_kv_restore_fix.diff
+
+# Build llama.cpp with OpenBLAS support after patches
+RUN cd /home/user/code/models/llama.cpp && \
+    export PKG_CONFIG_PATH=/opt/OpenBLAS/lib/pkgconfig:$PKG_CONFIG_PATH && \
+    cmake -B build -DGGML_BLAS=ON -DGGML_BLAS_VENDOR=OpenBLAS -DBLAS_INCLUDE_DIRS=/home/user/code/models/OpenBLAS -DLLAMA_CURL=OFF && \
+    cmake --build build --config Release -j2 && \
+    cp /home/user/code/models/llama.cpp/build/bin/* /home/user/code/models/llama.cpp/
 
 # Copy files into the container as the non-root user
 COPY --chown=user:user appsettings.json /home/user/code/app/appsettings.json
@@ -104,4 +110,3 @@ RUN cp -r /home/user/code/NetworkMonitorLLM/bin/Release/net10.0/* /home/user/cod
 WORKDIR /home/user/code/app
 
 CMD ["dotnet", "NetworkMonitorLLM.dll", "--urls", "http://0.0.0.0:7860"]
-