Merge remote-tracking branch 'origin/master' into sl/micro-batching

.devops/main-intel.Dockerfile

@@ -0,0 +1,26 @@
+ARG ONEAPI_VERSION=2024.0.1-devel-ubuntu22.04
+ARG UBUNTU_VERSION=22.04
+
+FROM intel/hpckit:$ONEAPI_VERSION as build
+
+RUN apt-get update && \
+    apt-get install -y git
+
+WORKDIR /app
+
+COPY . .
+
+# for some reasons, "-DLLAMA_BLAS=ON -DLLAMA_BLAS_VENDOR=Intel10_64lp -DLLAMA_NATIVE=ON" give worse performance
+RUN mkdir build && \
+    cd build && \
+    cmake .. -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx && \
+    cmake --build . --config Release --target main server
+
+FROM ubuntu:$UBUNTU_VERSION as runtime
+
+COPY --from=build /app/build/bin/main /main
+COPY --from=build /app/build/bin/server /server
+
+ENV LC_ALL=C.utf8
+
+ENTRYPOINT [ "/main" ]

.devops/nix/nixpkgs-instances.nix

@@ -7,6 +7,18 @@
     { system, ... }:
     {
       _module.args = {
+        # Note: bringing up https://zimbatm.com/notes/1000-instances-of-nixpkgs
+        # again, the below creates several nixpkgs instances which the
+        # flake-centric CLI will be forced to evaluate e.g. on `nix flake show`.
+        #
+        # This is currently "slow" and "expensive", on a certain scale.
+        # This also isn't "right" in that this hinders dependency injection at
+        # the level of flake inputs. This might get removed in the foreseeable
+        # future.
+        #
+        # Note that you can use these expressions without Nix
+        # (`pkgs.callPackage ./devops/nix/scope.nix { }` is the entry point).
+
         pkgsCuda = import inputs.nixpkgs {
           inherit system;
           # Ensure dependencies use CUDA consistently (e.g. that openmpi, ucc,

.devops/nix/package.nix

@@ -73,6 +73,7 @@ let
     ps: [
       ps.numpy
       ps.sentencepiece
+      ps.tiktoken
       ps.torchWithoutCuda
       ps.transformers
     ]
@@ -114,14 +115,22 @@ effectiveStdenv.mkDerivation (
     pname = "llama-cpp${pnameSuffix}";
     version = llamaVersion;
 
+    # Note: none of the files discarded here are visible in the sandbox or
+    # affect the output hash. This also means they can be modified without
+    # triggering a rebuild.
     src = lib.cleanSourceWith {
       filter =
         name: type:
-        !(builtins.any (_: _) [
+        let
+          noneOf = builtins.all (x: !x);
+          baseName = baseNameOf name;
+        in
+        noneOf [
           (lib.hasSuffix ".nix" name) # Ignore *.nix files when computing outPaths
-          (name == "README.md") # Ignore *.md changes whe computing outPaths
-          (lib.hasPrefix "." name) # Skip hidden files and directories
-        ]);
+          (lib.hasSuffix ".md" name) # Ignore *.md changes whe computing outPaths
+          (lib.hasPrefix "." baseName) # Skip hidden files and directories
+          (baseName == "flake.lock")
+        ];
       src = lib.cleanSource ../../.;
     };
 
@@ -159,7 +168,7 @@ effectiveStdenv.mkDerivation (
 
     cmakeFlags =
       [
-        (cmakeBool "LLAMA_NATIVE" true)
+        (cmakeBool "LLAMA_NATIVE" false)
         (cmakeBool "LLAMA_BUILD_SERVER" true)
         (cmakeBool "BUILD_SHARED_LIBS" true)
         (cmakeBool "CMAKE_SKIP_BUILD_RPATH" true)
@@ -216,6 +225,9 @@ effectiveStdenv.mkDerivation (
         description = "contains numpy and sentencepiece";
         buildInputs = [ llama-python ];
         inputsFrom = [ finalAttrs.finalPackage ];
+        shellHook = ''
+          addToSearchPath "LD_LIBRARY_PATH" "${lib.getLib effectiveStdenv.cc.cc}/lib"
+        '';
       };
 
       shell-extra = mkShell {

.devops/nix/scope.nix

@@ -4,6 +4,10 @@
   llamaVersion ? "0.0.0",
 }:
 
+# We're using `makeScope` instead of just writing out an attrset
+# because it allows users to apply overlays later using `overrideScope'`.
+# Cf. https://noogle.dev/f/lib/makeScope
+
 lib.makeScope newScope (
   self: {
     inherit llamaVersion;

.github/workflows/build.yml

@@ -295,7 +295,7 @@ jobs:
       OPENBLAS_VERSION: 0.3.23
       OPENCL_VERSION: 2023.04.17
       CLBLAST_VERSION: 1.6.0
-      SDE_VERSION: 9.21.1-2023-04-24
+      SDE_VERSION: 9.33.0-2024-01-07
 
     strategy:
       matrix:
@@ -400,7 +400,7 @@ jobs:
         id: cmake_test_sde
         if: ${{ matrix.build == 'avx512' && env.HAS_AVX512F == '0' }} # use Intel SDE for AVX-512 emulation
         run: |
-          curl.exe -o $env:RUNNER_TEMP/sde.tar.xz -L "https://downloadmirror.intel.com/777395/sde-external-${env:SDE_VERSION}-win.tar.xz"
+          curl.exe -o $env:RUNNER_TEMP/sde.tar.xz -L "https://downloadmirror.intel.com/813591/sde-external-${env:SDE_VERSION}-win.tar.xz"
           # for some weird reason windows tar doesn't like sde tar.xz
           7z x "-o${env:RUNNER_TEMP}" $env:RUNNER_TEMP/sde.tar.xz
           7z x "-o${env:RUNNER_TEMP}" $env:RUNNER_TEMP/sde.tar

.github/workflows/docker.yml

@@ -35,6 +35,7 @@ jobs:
           - { tag: "full-cuda", dockerfile: ".devops/full-cuda.Dockerfile", platforms: "linux/amd64" }
           - { tag: "light-rocm", dockerfile: ".devops/main-rocm.Dockerfile", platforms: "linux/amd64,linux/arm64" }
           - { tag: "full-rocm", dockerfile: ".devops/full-rocm.Dockerfile", platforms: "linux/amd64,linux/arm64" }
+          - { tag: "light-intel", dockerfile: ".devops/main-intel.Dockerfile", platforms: "linux/amd64" }
     steps:
       - name: Check out the repo
         uses: actions/checkout@v3

.github/workflows/nix-ci-aarch64.yml

@@ -2,13 +2,20 @@ name: Nix aarch64 builds
 
 on:
   workflow_dispatch: # allows manual triggering
+  schedule:
+    # Rebuild daily rather than on every push because QEMU is expensive (e.g.
+    # 1.5h instead of minutes with the cold cache).
+    #
+    # randint(0, 59), randint(0, 23)
+    - cron: '26 12 * * *'
+  # But also rebuild if we touched any of the Nix expressions:
   push:
     branches:
       - master
-    paths: ['.github/workflows/**', '**/CMakeLists.txt', '**/Makefile', '**/*.h', '**/*.hpp', '**/*.c', '**/*.cpp', '**/*.cu', '**/*.swift', '**/*.m', '**/*.sh', '**/*.py', '**/*.nix']
+    paths: ['**/*.nix', 'flake.lock']
   pull_request:
     types: [opened, synchronize, reopened]
-    paths: ['**/CMakeLists.txt', '**/Makefile', '**/*.h', '**/*.hpp', '**/*.c', '**/*.cpp', '**/*.cu', '**/*.swift', '**/*.m', '**/*.sh', '**/*.py', '**/*.nix']
+    paths: ['**/*.nix', 'flake.lock']
 
 jobs:
   nix-build-aarch64:

.github/workflows/nix-ci.yml

@@ -5,10 +5,8 @@ on:
   push:
     branches:
       - master
-    paths: ['.github/workflows/**', '**/CMakeLists.txt', '**/Makefile', '**/*.h', '**/*.hpp', '**/*.c', '**/*.cpp', '**/*.cu', '**/*.swift', '**/*.m', '**/*.sh', '**/*.py', '**/*.nix']
   pull_request:
     types: [opened, synchronize, reopened]
-    paths: ['**/CMakeLists.txt', '**/Makefile', '**/*.h', '**/*.hpp', '**/*.c', '**/*.cpp', '**/*.cu', '**/*.swift', '**/*.m', '**/*.sh', '**/*.py', '**/*.nix']
 
 jobs:
   nix-eval:

CMakeLists.txt

@@ -108,6 +108,13 @@ option(LLAMA_BUILD_TESTS                     "llama: build tests"    ${LLAMA_STA
 option(LLAMA_BUILD_EXAMPLES                  "llama: build examples" ${LLAMA_STANDALONE})
 option(LLAMA_BUILD_SERVER                    "llama: build server example"                      ON)
 
+
+# add perf arguments
+option(LLAMA_PERF                            "llama: enable perf"                               OFF)
+if (LLAMA_PERF)
+    add_definitions(-DGGML_PERF)
+endif()
+
 # Required for relocatable CMake package
 include(${CMAKE_CURRENT_SOURCE_DIR}/scripts/build-info.cmake)
 
@@ -471,6 +478,11 @@ function(get_flags CCID CCVER)
         if (CCVER VERSION_GREATER_EQUAL 8.1.0)
             set(CXX_FLAGS ${CXX_FLAGS} -Wextra-semi)
         endif()
+    elseif (CCID MATCHES "Intel")
+        # enable max optimization level when using Intel compiler
+        set(C_FLAGS   -ipo -O3 -static -fp-model=fast -flto -fno-stack-protector)
+        set(CXX_FLAGS -ipo -O3 -static -fp-model=fast -flto -fno-stack-protector)
+        add_link_options(-fuse-ld=lld -static-intel)
     endif()
 
     set(GF_C_FLAGS   ${C_FLAGS}   PARENT_SCOPE)

common/common.cpp

@@ -203,6 +203,23 @@ bool gpt_params_parse_ex(int argc, char ** argv, gpt_params & params) {
             params.prompt_cache_all = true;
         } else if (arg == "--prompt-cache-ro") {
             params.prompt_cache_ro = true;
+        } else if (arg == "-bf" || arg == "--binary-file") {
+            if (++i >= argc) {
+                invalid_param = true;
+                break;
+            }
+            std::ifstream file(argv[i], std::ios::binary);
+            if (!file) {
+                fprintf(stderr, "error: failed to open file '%s'\n", argv[i]);
+                invalid_param = true;
+                break;
+            }
+            // store the external file name in params
+            params.prompt_file = argv[i];
+            std::ostringstream ss;
+            ss << file.rdbuf();
+            params.prompt = ss.str();
+            fprintf(stderr, "Read %zu bytes from binary file %s\n", params.prompt.size(), argv[i]);
         } else if (arg == "-f" || arg == "--file") {
             if (++i >= argc) {
                 invalid_param = true;
@@ -659,6 +676,12 @@ bool gpt_params_parse_ex(int argc, char ** argv, gpt_params & params) {
             if (params.logdir.back() != DIRECTORY_SEPARATOR) {
                 params.logdir += DIRECTORY_SEPARATOR;
             }
+        } else if (arg == "--save-all-logits" || arg == "--kl-divergence-base") {
+            if (++i >= argc) {
+                invalid_param = true;
+                break;
+            }
+            params.logits_file = argv[i];
         } else if (arg == "--perplexity" || arg == "--all-logits") {
             params.logits_all = true;
         } else if (arg == "--ppl-stride") {
@@ -695,6 +718,16 @@ bool gpt_params_parse_ex(int argc, char ** argv, gpt_params & params) {
                 break;
             }
             params.winogrande_tasks = std::stoi(argv[i]);
+        } else if (arg == "--multiple-choice") {
+            params.multiple_choice = true;
+        } else if (arg == "--multiple-choice-tasks") {
+            if (++i >= argc) {
+                invalid_param = true;
+                break;
+            }
+            params.multiple_choice_tasks = std::stoi(argv[i]);
+        } else if (arg == "--kl-divergence") {
+            params.kl_divergence = true;
         } else if (arg == "--ignore-eos") {
             params.ignore_eos = true;
         } else if (arg == "--no-penalize-nl") {
@@ -894,6 +927,8 @@ void gpt_print_usage(int /*argc*/, char ** argv, const gpt_params & params) {
     printf("  --in-suffix STRING    string to suffix after user inputs with (default: empty)\n");
     printf("  -f FNAME, --file FNAME\n");
     printf("                        prompt file to start generation.\n");
+    printf("  -bf FNAME, --binary-file FNAME\n");
+    printf("                        binary file containing multiple choice tasks.\n");
     printf("  -n N, --n-predict N   number of tokens to predict (default: %d, -1 = infinity, -2 = until context filled)\n", params.n_predict);
     printf("  -c N, --ctx-size N    size of the prompt context (default: %d, 0 = loaded from model)\n", params.n_ctx);
     printf("  -b N, --batch-size N  batch size for prompt processing (default: %d)\n", params.n_batch);
@@ -944,6 +979,9 @@ void gpt_print_usage(int /*argc*/, char ** argv, const gpt_params & params) {
     printf("  --hellaswag-tasks N   number of tasks to use when computing the HellaSwag score (default: %zu)\n", params.hellaswag_tasks);
     printf("  --winogrande          compute Winogrande score over random tasks from datafile supplied with -f\n");
     printf("  --winogrande-tasks N  number of tasks to use when computing the Winogrande score (default: %zu)\n", params.winogrande_tasks);
+    printf("  --multiple-choice     compute multiple choice score over random tasks from datafile supplied with -f\n");
+    printf("  --multiple-choice-tasks N number of tasks to use when computing the multiple choice score (default: %zu)\n", params.winogrande_tasks);
+    printf("  --kl-divergence       computes KL-divergence to logits provided via --kl-divergence-base");
     printf("  --keep N              number of tokens to keep from the initial prompt (default: %d, -1 = all)\n", params.n_keep);
     printf("  --draft N             number of tokens to draft for speculative decoding (default: %d)\n", params.n_draft);
     printf("  --chunks N            max number of chunks to process (default: %d, -1 = all)\n", params.n_chunks);

common/common.h

@@ -92,6 +92,7 @@ struct gpt_params {
     std::string input_suffix      = "";  // string to suffix user inputs with
     std::vector<std::string> antiprompt; // string upon seeing which more user input is prompted
     std::string logdir            = "";  // directory in which to save YAML log files
+    std::string logits_file       = "";  // file for saving *all* logits
 
     std::vector<llama_model_kv_override> kv_overrides;
 
@@ -109,6 +110,11 @@ struct gpt_params {
     bool   winogrande      = false; // compute Winogrande score over random tasks from datafile supplied in prompt
     size_t winogrande_tasks= 0;     // number of tasks to use when computing the Winogrande score. If 0, all tasks will be computed
 
+    bool   multiple_choice = false; // compute TruthfulQA score over random tasks from datafile supplied in prompt
+    size_t multiple_choice_tasks = 0;     // number of tasks to use when computing the TruthfulQA score. If 0, all tasks will be computed
+
+    bool   kl_divergence   = false; // compute KL-divergence
+
     bool mul_mat_q         = true;  // if true, use mul_mat_q kernels instead of cuBLAS
     bool random_prompt     = false; // do not randomize prompt if none provided
     bool use_color         = false; // use color to distinguish generations and inputs