Merge pull request #475 from janhq/473-chore-pump-llama.cpp

chore: Pump `llama.cpp` version

Author: tikikun

.github/workflows/build.yml

@@ -231,7 +231,7 @@ jobs:
         run: |
           ./install_deps.sh
           mkdir build && cd build
-          cmake  -DLLAMA_NATIVE=OFF -DLLAMA_CUBLAS=ON -DLLAMA_CUBLAS=ON -DWHISPER_CUBLAS=ON -DNITRO_VERSION=${{ needs.set-nitro-version.outputs.version }} ..
+          cmake  -DLLAMA_NATIVE=OFF -DLLAMA_CUDA=ON -DWHISPER_CUDA=ON -DNITRO_VERSION=${{ needs.set-nitro-version.outputs.version }} ..
           make -j $(nproc)
           ls -la
 
@@ -285,7 +285,7 @@ jobs:
         run: |
           ./install_deps.sh
           mkdir build && cd build
-          cmake -DWHISPER_COREML=1 -DNITRO_VERSION=${{ needs.set-nitro-version.outputs.version }} .. 
+          cmake -DLLAMA_METAL_EMBED_LIBRARY=ON -DWHISPER_COREML=1 -DNITRO_VERSION=${{ needs.set-nitro-version.outputs.version }} .. 
           CC=gcc-8 make -j $(sysctl -n hw.ncpu)
           ls -la
 

CMakeLists.txt

@@ -26,13 +26,13 @@ set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
 set(CMAKE_PREFIX_PATH ${CMAKE_CURRENT_SOURCE_DIR}/build_deps/_install)
 # This is the critical line for installing another package
 
-if(LLAMA_CUBLAS)
+if(LLAMA_CUDA)
   cmake_minimum_required(VERSION 3.17)
 
   find_package(CUDAToolkit)
   if(CUDAToolkit_FOUND)
     message(STATUS "cuBLAS found")
-    add_compile_definitions(GGML_USE_CUBLAS)
+    add_compile_definitions(GGML_USE_CUDA)
   endif()
 endif()
 

audio.md

@@ -6,7 +6,7 @@
 
 ```bash
 mkdir build && cd build
-cmake -DLLAMA_CUBLAS=ON -DWHISPER_CUBLAS=ON ..
+cmake -DLLAMA_CUDA=ON -DWHISPER_CUDA=ON ..
 make -j$(nproc)
 ```
 
@@ -31,7 +31,7 @@ cmake --build build -j --config Release
 ```
 mkdir -p build
 cd build
-cmake .. -DLLAMA_CUBLAS=ON -DBUILD_SHARED_LIBS=ON -DWHISPER_CUBLAS=ON -DWHISPER_SDL2=ON
+cmake .. -DLLAMA_CUDA=ON -DBUILD_SHARED_LIBS=ON -DWHISPER_CUDA=ON -DWHISPER_SDL2=ON
 cmake --build . --config Release
 ```
 

context/llama_server_context.h

@@ -1525,7 +1525,7 @@ struct llama_server_context {
             slot.id, slot.params.n_keep, n_left, n_discard);
         llama_kv_cache_seq_rm(ctx, slot.id, slot.params.n_keep + 1,
                               slot.params.n_keep + n_discard + 1);
-        llama_kv_cache_seq_shift(ctx, slot.id,
+        llama_kv_cache_seq_add(ctx, slot.id,
                                  slot.params.n_keep + 1 + n_discard,
                                  slot.n_past, -n_discard);
 

controllers/llamaCPP.cc

@@ -630,10 +630,6 @@ bool llamaCPP::LoadModelImpl(std::shared_ptr<Json::Value> jsonBody) {
       log_set_target(llama_log_folder + "llama.log");
     }  // Set folder for llama log
   }
-#ifdef GGML_USE_CUBLAS
-  LOG_INFO << "Setting up GGML CUBLAS PARAMS";
-  params.mul_mat_q = false;
-#endif  // GGML_USE_CUBLAS
   if (params.model_alias == "unknown") {
     params.model_alias = params.model;
   }

docs/docs/new/build-source.md

@@ -69,7 +69,7 @@ Now, let's generate the build files.
 
   ```bash
   mkdir build && cd build
-  cmake -DLLAMA_CUBLAS=ON ..
+  cmake -DLLAMA_CUDA=ON ..
   ```
 
 ## Build the Application

examples/example-docker/cuda.Dockerfile

@@ -12,7 +12,7 @@ RUN apt-get update && \
     ./install_deps.sh && \
     mkdir build && \
     cd build && \
-    cmake .. -DDEBUG=ON -DLLAMA_CUBLAS=ON -DLLAMA_CUDA_F16=ON -DLLAMA_CUDA_DMMV_X=64 -DLLAMA_CUDA_MMV_Y=32 && \
+    cmake .. -DDEBUG=ON -DLLAMA_CUDA=ON -DLLAMA_CUDA_F16=ON -DLLAMA_CUDA_DMMV_X=64 -DLLAMA_CUDA_MMV_Y=32 && \
     cmake --build . --config Release -j $(nproc) && \
     apt-get remove --purge -y git cmake && \
     apt-get autoremove -y && \