ci : add model tests + script wrapper (#4586 )

* scripts : add lib.sh and lib_test.sh * scripts : stub out new ci-run.sh script * scripts : switch to PascalCase for functions This looks a little odd at first, but I find it very useful as a convention to know if a command is part of our code vs a builtin. * scripts : add some fancy conversion from snake_case to PascalCase * Add venv to ci/run.sh * Revert scripts work * scripts : add wrapper script for local use of ci/run.sh * Simplify .gitignore for tests, clang-tidy fixes * Label all ctest tests * ci : ctest uses -L main * Attempt at writing ctest_with_model * Update test-model-load-cancel * ci : add ctest_with_model for debug and release ggml-ci * Fix gg_get_model function ggml-ci * got stuck on CMake * Add get_model.cpp to tests/CMakeLists.txt ggml-ci * Fix README.md output for ctest_with_model ggml-ci * workflows : use `-L main` for all ctest ggml-ci * Fixes * GG_RUN_CTEST_MODELFILE => LLAMACPP_TESTMODELFILE * Always show warning rather than failing if model file variable is not set * scripts : update usage text for ci-run.sh
metal : remove unused n_buffers and buffers (#5129 )
2026-07-14 08:25:55 +02:00 · 2024-01-26 14:18:00 +02:00 · 2024-01-26 14:16:07 +02:00 · 2024-01-26 11:10:28 +02:00 · 2024-01-26 10:52:33 +02:00 · 2024-01-26 09:14:39 +02:00
15 changed files with 268 additions and 110 deletions
@@ -72,7 +72,7 @@ jobs:
        id: cmake_test
        run: |
          cd build
-          ctest --verbose --timeout 900
+          ctest -L main --verbose --timeout 900

  ubuntu-latest-cmake-sanitizer:
    runs-on: ubuntu-latest
@@ -107,7 +107,7 @@ jobs:
        id: cmake_test
        run: |
          cd build
-          ctest --verbose --timeout 900
+          ctest -L main --verbose --timeout 900

  ubuntu-latest-cmake-mpi:
    runs-on: ubuntu-latest
@@ -141,7 +141,7 @@ jobs:
        id: cmake_test
        run: |
          cd build
-          ctest --verbose
+          ctest -L main --verbose

  # TODO: build with LLAMA_NO_METAL because test-backend-ops fail on "Apple Paravirtual device" and I don't know
  #       how to debug it.
@@ -202,7 +202,7 @@ jobs:
        id: cmake_test
        run: |
          cd build
-          ctest --verbose --timeout 900
+          ctest -L main --verbose --timeout 900

  macOS-latest-cmake-ios:
    runs-on: macos-latest
@@ -394,7 +394,7 @@ jobs:
        if: ${{ matrix.build != 'clblast' && (matrix.build != 'avx512' || env.HAS_AVX512F == '1') }} # not all machines have native AVX-512
        run: |
          cd build
-          ctest -C Release --verbose --timeout 900
+          ctest -L main -C Release --verbose --timeout 900

      - name: Test (Intel SDE)
        id: cmake_test_sde
@@ -406,7 +406,7 @@ jobs:
          7z x "-o${env:RUNNER_TEMP}" $env:RUNNER_TEMP/sde.tar
          $sde = $(join-path $env:RUNNER_TEMP sde-external-${env:SDE_VERSION}-win/sde.exe)
          cd build
-          & $sde -future -- ctest -C Release --verbose --timeout 900
+          & $sde -future -- ctest -L main -C Release --verbose --timeout 900

      - name: Determine tag name
        id: tag
@@ -27,7 +27,7 @@
 lcov-report/
 gcovr-report/

-build*/
+build*
 out/
 tmp/

@@ -89,20 +89,3 @@ examples/jeopardy/results.txt

 poetry.lock
 poetry.toml
-
-# Test binaries
-/tests/test-grammar-parser
-/tests/test-llama-grammar
-/tests/test-double-float
-/tests/test-grad0
-/tests/test-opt
-/tests/test-quantize-fns
-/tests/test-quantize-perf
-/tests/test-sampling
-/tests/test-tokenizer-0-llama
-/tests/test-tokenizer-0-falcon
-/tests/test-tokenizer-1-llama
-/tests/test-tokenizer-1-bpe
-/tests/test-rope
-/tests/test-backend-ops
-/tests/test-autorelease
@@ -9,7 +9,7 @@ TEST_TARGETS = \
 	tests/test-llama-grammar tests/test-grammar-parser tests/test-double-float tests/test-grad0 tests/test-opt \
 	tests/test-quantize-fns tests/test-quantize-perf tests/test-sampling tests/test-tokenizer-0-llama          \
 	tests/test-tokenizer-0-falcon tests/test-tokenizer-1-llama tests/test-tokenizer-1-bpe tests/test-rope      \
-	tests/test-backend-ops tests/test-autorelease
+	tests/test-backend-ops tests/test-model-load-cancel tests/test-autorelease

 # Code coverage output files
 COV_TARGETS = *.gcno tests/*.gcno *.gcda tests/*.gcda *.gcov tests/*.gcov lcov-report gcovr-report
@@ -748,5 +748,8 @@ tests/test-c.o: tests/test-c.c llama.h
 tests/test-backend-ops: tests/test-backend-ops.cpp ggml.o $(OBJS)
 	$(CXX) $(CXXFLAGS) $(filter-out %.h,$^) -o $@ $(LDFLAGS)

-tests/test-autorelease: tests/test-autorelease.cpp ggml.o llama.o $(COMMON_DEPS) $(OBJS)
+tests/test-model-load-cancel: tests/test-model-load-cancel.cpp ggml.o llama.o tests/get-model.cpp $(COMMON_DEPS) $(OBJS)
+	$(CXX) $(CXXFLAGS) $(filter-out %.h,$^) -o $@ $(LDFLAGS)
+
+tests/test-autorelease: tests/test-autorelease.cpp ggml.o llama.o tests/get-model.cpp $(COMMON_DEPS) $(OBJS)
 	$(CXX) $(CXXFLAGS) $(filter-out %.h,$^) -o $@ $(LDFLAGS)
@@ -10,11 +10,11 @@ Inference of [LLaMA](https://arxiv.org/abs/2302.13971) model in pure C/C++

 ### Hot topics

+- ⚠️ Incoming backends: https://github.com/ggerganov/llama.cpp/discussions/5138
 - New SOTA quantized models, including pure 2-bits: https://huggingface.co/ikawrakow
 - Collecting Apple Silicon performance stats:
  - M-series: https://github.com/ggerganov/llama.cpp/discussions/4167
  - A-series: https://github.com/ggerganov/llama.cpp/discussions/4508
- Added Mixtral support: https://github.com/ggerganov/llama.cpp/pull/4406
 - Looking for contributions to improve and maintain the `server` example: https://github.com/ggerganov/llama.cpp/issues/4216

 ----
@@ -112,6 +112,7 @@ as the main playground for developing new features for the [ggml](https://github
 - [x] [Bakllava](https://huggingface.co/models?search=SkunkworksAI/Bakllava)
 - [x] [Obsidian](https://huggingface.co/NousResearch/Obsidian-3B-V0.5)
 - [x] [ShareGPT4V](https://huggingface.co/models?search=Lin-Chen/ShareGPT4V)
+- [x] [MobileVLM 1.7B/3B models](https://huggingface.co/models?search=mobileVLM)


 **Bindings:**
@@ -22,9 +22,9 @@ mkdir -p "$2"
 OUT=$(realpath "$1")
 MNT=$(realpath "$2")

-rm -v $OUT/*.log
-rm -v $OUT/*.exit
-rm -v $OUT/*.md
+rm -f "$OUT/*.log"
+rm -f "$OUT/*.exit"
+rm -f "$OUT/*.md"

 sd=`dirname $0`
 cd $sd/../
@@ -94,7 +94,7 @@ function gg_run_ctest_debug {
    (time cmake -DCMAKE_BUILD_TYPE=Debug ${CMAKE_EXTRA} .. ) 2>&1 | tee -a $OUT/${ci}-cmake.log
    (time make -j                                          ) 2>&1 | tee -a $OUT/${ci}-make.log

-    (time ctest --output-on-failure -E test-opt ) 2>&1 | tee -a $OUT/${ci}-ctest.log
+    (time ctest --output-on-failure -L main -E test-opt ) 2>&1 | tee -a $OUT/${ci}-ctest.log

    set +e
 }
@@ -123,9 +123,9 @@ function gg_run_ctest_release {
    (time make -j                                            ) 2>&1 | tee -a $OUT/${ci}-make.log

    if [ -z ${GG_BUILD_LOW_PERF} ]; then
-        (time ctest --output-on-failure ) 2>&1 | tee -a $OUT/${ci}-ctest.log
+        (time ctest --output-on-failure -L main ) 2>&1 | tee -a $OUT/${ci}-ctest.log
    else
-        (time ctest --output-on-failure -E test-opt ) 2>&1 | tee -a $OUT/${ci}-ctest.log
+        (time ctest --output-on-failure -L main -E test-opt ) 2>&1 | tee -a $OUT/${ci}-ctest.log
    fi

    set +e
@@ -141,6 +141,61 @@ function gg_sum_ctest_release {
    gg_printf '```\n'
 }

+function gg_get_model {
+    local gguf_3b="$MNT/models/open-llama/3B-v2/ggml-model-f16.gguf"
+    local gguf_7b="$MNT/models/open-llama/7B-v2/ggml-model-f16.gguf"
+    if [[ -s $gguf_3b ]]; then
+        echo -n "$gguf_3b"
+    elif [[ -s $gguf_7b ]]; then
+        echo -n "$gguf_7b"
+    else
+        echo >&2 "No model found. Can't run gg_run_ctest_with_model."
+        exit 1
+    fi
+}
+
+function gg_run_ctest_with_model_debug {
+    cd ${SRC}
+
+    local model; model=$(gg_get_model)
+    cd build-ci-debug
+    set -e
+    (LLAMACPP_TEST_MODELFILE="$model" time ctest --output-on-failure -L model) 2>&1 | tee -a $OUT/${ci}-ctest.log
+    set +e
+    cd ..
+}
+
+function gg_run_ctest_with_model_release {
+    cd ${SRC}
+
+    local model; model=$(gg_get_model)
+    cd build-ci-release
+    set -e
+    (LLAMACPP_TEST_MODELFILE="$model" time ctest --output-on-failure -L model) 2>&1 | tee -a $OUT/${ci}-ctest.log
+    set +e
+    cd ..
+}
+
+function gg_sum_ctest_with_model_debug {
+    gg_printf '### %s\n\n' "${ci}"
+
+    gg_printf 'Runs ctest with model files in debug mode\n'
+    gg_printf '- status: %s\n' "$(cat $OUT/${ci}.exit)"
+    gg_printf '```\n'
+    gg_printf '%s\n' "$(cat $OUT/${ci}-ctest.log)"
+    gg_printf '```\n'
+}
+
+function gg_sum_ctest_with_model_release {
+    gg_printf '### %s\n\n' "${ci}"
+
+    gg_printf 'Runs ctest with model files in release mode\n'
+    gg_printf '- status: %s\n' "$(cat $OUT/${ci}.exit)"
+    gg_printf '```\n'
+    gg_printf '%s\n' "$(cat $OUT/${ci}-ctest.log)"
+    gg_printf '```\n'
+}
+
 # open_llama_3b_v2

 function gg_run_open_llama_3b_v2 {
@@ -183,8 +238,6 @@ function gg_run_open_llama_3b_v2 {

    wiki_test_60="${path_wiki}/wiki.test-60.raw"

-    ./bin/test-autorelease ${model_f16}
-
    ./bin/quantize ${model_f16} ${model_q8_0} q8_0
    ./bin/quantize ${model_f16} ${model_q4_0} q4_0
    ./bin/quantize ${model_f16} ${model_q4_1} q4_1
@@ -507,14 +560,18 @@ function gg_sum_open_llama_7b_v2 {
 ## main

 if [ -z ${GG_BUILD_LOW_PERF} ]; then
+    # Create symlink: ./llama.cpp/models-mnt -> $MNT/models/models-mnt
    rm -rf ${SRC}/models-mnt
-
    mnt_models=${MNT}/models
    mkdir -p ${mnt_models}
    ln -sfn ${mnt_models} ${SRC}/models-mnt

-    python3 -m pip install -r ${SRC}/requirements.txt
-    python3 -m pip install --editable gguf-py
+    # Create a fresh python3 venv and enter it
+    python3 -m venv "$MNT/venv"
+    source "$MNT/venv/bin/activate"
+
+    pip install -r ${SRC}/requirements.txt --disable-pip-version-check
+    pip install --editable gguf-py --disable-pip-version-check
 fi

 ret=0
@@ -529,6 +586,8 @@ if [ -z ${GG_BUILD_LOW_PERF} ]; then
        else
            test $ret -eq 0 && gg_run open_llama_7b_v2
        fi
+        test $ret -eq 0 && gg_run ctest_with_model_debug
+        test $ret -eq 0 && gg_run ctest_with_model_release
    fi
 fi

@@ -26,15 +26,6 @@

 #define GGML_METAL_MAX_KERNELS 256

-struct ggml_metal_buffer {
-    const char * name;
-
-    void   * data;
-    size_t   size;
-
-    id<MTLBuffer> metal;
-};
-
 struct ggml_metal_kernel {
    id<MTLFunction>             function;
    id<MTLComputePipelineState> pipeline;
@@ -172,9 +163,6 @@ struct ggml_metal_context {

    dispatch_queue_t d_queue;

-    int n_buffers;
-    struct ggml_metal_buffer buffers[GGML_METAL_MAX_BUFFERS];
-
    struct ggml_metal_kernel kernels[GGML_METAL_MAX_KERNELS];

    bool support_simdgroup_reduction;
@@ -242,24 +230,20 @@ static struct ggml_metal_context * ggml_metal_init(int n_cb) {
    // Show all the Metal device instances in the system
    NSArray * devices = MTLCopyAllDevices();
    for (id<MTLDevice> device in devices) {
-        NSString * s = [device name];
-        GGML_METAL_LOG_INFO("%s: found device: %s\n", __func__, [s UTF8String]);
+        GGML_METAL_LOG_INFO("%s: found device: %s\n", __func__, [[device name] UTF8String]);
    }
    [devices release]; // since it was created by a *Copy* C method
 #endif

    // Pick and show default Metal device
    id<MTLDevice> device = MTLCreateSystemDefaultDevice();
-    NSString * s = [device name];
-    GGML_METAL_LOG_INFO("%s: picking default device: %s\n", __func__, [s UTF8String]);
+    GGML_METAL_LOG_INFO("%s: picking default device: %s\n", __func__, [[device name] UTF8String]);

    // Configure context
    struct ggml_metal_context * ctx = malloc(sizeof(struct ggml_metal_context));
    ctx->device = device;
    ctx->n_cb   = MIN(n_cb, GGML_METAL_MAX_BUFFERS);
    ctx->queue  = [ctx->device newCommandQueue];
-    ctx->n_buffers = 0;
-
    ctx->d_queue = dispatch_queue_create("ggml-metal", DISPATCH_QUEUE_CONCURRENT);

    // load library
@@ -534,10 +518,6 @@ static struct ggml_metal_context * ggml_metal_init(int n_cb) {
 static void ggml_metal_free(struct ggml_metal_context * ctx) {
    GGML_METAL_LOG_INFO("%s: deallocating\n", __func__);

-    for (int i = 0; i < ctx->n_buffers; ++i) {
-        [ctx->buffers[i].metal release];
-    }
-
    for (int i = 0; i < GGML_METAL_MAX_KERNELS; ++i) {
        if (ctx->kernels[i].pipeline) {
            [ctx->kernels[i].pipeline release];
@@ -580,51 +560,30 @@ struct ggml_backend_metal_buffer_context {
 // the assumption is that there is 1-to-1 mapping between the host and device memory buffers, so we can find the
 // Metal buffer based on the host memory pointer
 //
-static id<MTLBuffer> ggml_metal_get_buffer(struct ggml_metal_context * ctx, struct ggml_tensor * t, size_t * offs) {
+static id<MTLBuffer> ggml_metal_get_buffer(struct ggml_tensor * t, size_t * offs) {
    //GGML_METAL_LOG_INFO("%s: data tensor '%16s', offs_data = %8ld, offs_eval = %8ld, offs_cach = %8ld\n", __func__, t->name, offs_data, offs_eval, offs_cach);

    const int64_t tsize = ggml_nbytes(t);

    ggml_backend_buffer_t buffer = t->view_src ? t->view_src->buffer : t->buffer;

-    // compatibility with ggml-backend
-    if (buffer && buffer->buft == ggml_backend_metal_buffer_type()) {
-        struct ggml_backend_metal_buffer_context * buf_ctx = (struct ggml_backend_metal_buffer_context *) buffer->context;
-
-        // find the view that contains the tensor fully
-        for (int i = 0; i < buf_ctx->n_buffers; ++i) {
-            const int64_t ioffs = (int64_t) t->data - (int64_t) buf_ctx->buffers[i].data;
-
-            //GGML_METAL_LOG_INFO("ioffs = %10ld, tsize = %10ld, sum = %10ld, buf_ctx->buffers[%d].size = %10ld\n", ioffs, tsize, ioffs + tsize, i, buf_ctx->buffers[i].size);
-            if (ioffs >= 0 && ioffs + tsize <= (int64_t) buf_ctx->buffers[i].size) {
-                *offs = (size_t) ioffs;
-
-                //GGML_METAL_LOG_INFO("%s: tensor '%16s', offs = %8ld\n", __func__, t->name, *offs);
-
-                return buf_ctx->buffers[i].metal;
-            }
-        }
-
-        GGML_METAL_LOG_ERROR("%s: error: tensor '%s' buffer is nil\n", __func__, t->name);
-
-        return nil;
-    }
+    struct ggml_backend_metal_buffer_context * buf_ctx = (struct ggml_backend_metal_buffer_context *) buffer->context;

    // find the view that contains the tensor fully
-    for (int i = 0; i < ctx->n_buffers; ++i) {
-        const int64_t ioffs = (int64_t) t->data - (int64_t) ctx->buffers[i].data;
+    for (int i = 0; i < buf_ctx->n_buffers; ++i) {
+        const int64_t ioffs = (int64_t) t->data - (int64_t) buf_ctx->buffers[i].data;

-        //GGML_METAL_LOG_INFO("ioffs = %10ld, tsize = %10ld, sum = %10ld, ctx->buffers[%d].size = %10ld, name = %s\n", ioffs, tsize, ioffs + tsize, i, ctx->buffers[i].size, ctx->buffers[i].name);
-        if (ioffs >= 0 && ioffs + tsize <= (int64_t) ctx->buffers[i].size) {
+        //GGML_METAL_LOG_INFO("ioffs = %10ld, tsize = %10ld, sum = %10ld, buf_ctx->buffers[%d].size = %10ld\n", ioffs, tsize, ioffs + tsize, i, buf_ctx->buffers[i].size);
+        if (ioffs >= 0 && ioffs + tsize <= (int64_t) buf_ctx->buffers[i].size) {
            *offs = (size_t) ioffs;

-            //GGML_METAL_LOG_INFO("%s: '%s' tensor '%16s', offs = %8ld\n", __func__, ctx->buffers[i].name, t->name, *offs);
+            //GGML_METAL_LOG_INFO("%s: tensor '%16s', offs = %8ld\n", __func__, t->name, *offs);

-            return ctx->buffers[i].metal;
+            return buf_ctx->buffers[i].metal;
        }
    }

-    GGML_METAL_LOG_ERROR("%s: error: buffer is nil\n", __func__);
+    GGML_METAL_LOG_ERROR("%s: error: tensor '%s' buffer is nil\n", __func__, t->name);

    return nil;
 }
@@ -817,9 +776,9 @@ static bool ggml_metal_graph_compute(
            const enum ggml_type src1t = src1 ? src1->type : GGML_TYPE_COUNT;
            const enum ggml_type dstt  = dst  ? dst->type  : GGML_TYPE_COUNT;

-            id<MTLBuffer> id_src0 = src0 ? ggml_metal_get_buffer(ctx, src0, &offs_src0) : nil;
-            id<MTLBuffer> id_src1 = src1 ? ggml_metal_get_buffer(ctx, src1, &offs_src1) : nil;
-            id<MTLBuffer> id_dst  = dst  ? ggml_metal_get_buffer(ctx, dst,  &offs_dst)  : nil;
+            id<MTLBuffer> id_src0 = src0 ? ggml_metal_get_buffer(src0, &offs_src0) : nil;
+            id<MTLBuffer> id_src1 = src1 ? ggml_metal_get_buffer(src1, &offs_src1) : nil;
+            id<MTLBuffer> id_dst  = dst  ? ggml_metal_get_buffer(dst,  &offs_dst)  : nil;

            //GGML_METAL_LOG_INFO("%s: op - %s\n", __func__, ggml_op_name(dst->op));
            //if (src0) {
@@ -1601,7 +1560,7 @@ static bool ggml_metal_graph_compute(
                                struct ggml_tensor * src_cur = dst->src[2 + (j % n_as)];

                                size_t offs_src_cur = 0;
-                                id<MTLBuffer> id_src_cur = ggml_metal_get_buffer(ctx, src_cur, &offs_src_cur);
+                                id<MTLBuffer> id_src_cur = ggml_metal_get_buffer(src_cur, &offs_src_cur);

                                [encoder setBuffer:id_src_cur offset:offs_src_cur atIndex:19 + j];
                            }
@@ -1746,7 +1705,7 @@ static bool ggml_metal_graph_compute(
                                struct ggml_tensor * src_cur = dst->src[2 + (j % n_as)];

                                size_t offs_src_cur = 0;
-                                id<MTLBuffer> id_src_cur = ggml_metal_get_buffer(ctx, src_cur, &offs_src_cur);
+                                id<MTLBuffer> id_src_cur = ggml_metal_get_buffer(src_cur, &offs_src_cur);

                                [encoder setBuffer:id_src_cur offset:offs_src_cur atIndex:23 + j];
                            }
@@ -107,7 +107,7 @@ class GGUFReader:
        offs, tensors_fields = self._build_tensors_fields(offs, tensor_count)
        new_align = self.fields.get('general.alignment')
        if new_align is not None:
-            if new_align.types != [GGUFValueType.UINT64]:
+            if new_align.types != [GGUFValueType.UINT32]:
                raise ValueError('Bad type for general.alignment field')
            self.alignment = new_align.parts[-1][0]
        padding = offs % self.alignment
@@ -7956,10 +7956,57 @@ void llama_sample_top_k(struct llama_context * ctx, llama_token_data_array * can
        auto comp = [](const llama_token_data & a, const llama_token_data & b) {
            return a.logit > b.logit;
        };
-        if (k == (int) candidates->size) {
-            std::sort(candidates->data, candidates->data + candidates->size, comp);
-        } else {
+        if (k <= 128) {
            std::partial_sort(candidates->data, candidates->data + k, candidates->data + candidates->size, comp);
+        } else {
+            constexpr int   nbuckets     = 128;
+            constexpr float bucket_low   = -10.0f;
+            constexpr float bucket_high  =  10.0f;
+            constexpr float bucket_scale = nbuckets/(bucket_high - bucket_low);
+            constexpr float bucker_inter = -bucket_low * bucket_scale;
+
+            std::vector<int> bucket_idx(candidates->size);
+            std::vector<int> histo(nbuckets, 0);
+
+            for (int i = 0; i < (int)candidates->size; ++i) {
+                const float val = candidates->data[i].logit;
+                int ib = int(bucket_scale * val + bucker_inter); //nbuckets * (val - bucket_low) / (bucket_high - bucket_low);
+                ib = std::max(0, std::min(nbuckets-1, ib));
+                bucket_idx[i] = ib;
+                ++histo[ib];
+            }
+            int nhave = 0;
+            int ib = nbuckets - 1;
+            for ( ; ib >= 0; --ib) {
+                nhave += histo[ib];
+                if (nhave >= k) break;
+            }
+            std::vector<llama_token_data> tmp_tokens(nhave);
+            auto ptr = tmp_tokens.data();
+            std::vector<llama_token_data*> bucket_ptrs;
+            bucket_ptrs.reserve(nbuckets - ib);
+            for (int j = nbuckets - 1; j >= ib; --j) {
+                bucket_ptrs.push_back(ptr);
+                ptr += histo[j];
+            }
+            for (int i = 0; i < (int)candidates->size; ++i) {
+                int j = bucket_idx[i];
+                if (j >= ib) {
+                    *bucket_ptrs[nbuckets-1-j]++ = candidates->data[i];
+                }
+            }
+
+            ptr = tmp_tokens.data();
+            int ndone = 0;
+            for (int j = nbuckets-1; j > ib; --j) {
+                std::sort(ptr, ptr + histo[j], comp);
+                ptr += histo[j];
+                ndone += histo[j];
+            }
+            std::partial_sort(ptr, ptr + k - ndone, ptr + histo[ib], comp);
+
+            std::memcpy(candidates->data, tmp_tokens.data(), k*sizeof(llama_token_data));
+
        }
        candidates->sorted = true;
    }
@@ -0,0 +1,50 @@
+#!/bin/bash
+set -euo pipefail
+this=$(realpath "$0"); readonly this
+cd "$(dirname "$this")"
+shellcheck "$this"
+
+if (( $# != 1 && $# != 2  )); then
+    cat >&2 <<'EOF'
+usage:
+    ci-run.sh <tmp_dir> [<cache_dir>]
+
+This script wraps ci/run.sh:
+* If <tmp_dir> is a ramdisk, you can reduce writes to your SSD. If <tmp_dir> is not a ramdisk, keep in mind that total writes will increase by the size of <cache_dir>.
+    (openllama_3b_v2: quantized models are about 30GB)
+* Persistent model and data files are synced to and from <cache_dir>,
+    excluding generated .gguf files.
+    (openllama_3b_v2: persistent files are about 6.6GB)
+* <cache_dir> defaults to  ~/.cache/llama.cpp
+EOF
+    exit 1
+fi
+
+cd .. # => llama.cpp repo root
+
+tmp="$1"
+mkdir -p "$tmp"
+tmp=$(realpath "$tmp")
+echo >&2 "Using tmp=$tmp"
+
+cache="${2-$HOME/.cache/llama.cpp}"
+mkdir -p "$cache"
+cache=$(realpath "$cache")
+echo >&2 "Using cache=$cache"
+
+_sync() {
+    local from="$1"; shift
+    local to="$1"; shift
+
+    echo >&2 "Syncing from $from to $to"
+    mkdir -p "$from" "$to"
+    rsync -a "$from" "$to" --delete-during "$@"
+}
+
+_sync "$(realpath .)/" "$tmp/llama.cpp"
+_sync "$cache/ci-mnt/models/" "$tmp/llama.cpp/ci-mnt/models/"
+
+cd "$tmp/llama.cpp"
+bash ci/run.sh ci-out ci-mnt
+
+_sync 'ci-mnt/models/' "$cache/ci-mnt/models/" --exclude='*.gguf' -P
@@ -0,0 +1,2 @@
+*
+!*.*
@@ -1,6 +1,6 @@
 function(llama_build_executable source)
    get_filename_component(TEST_TARGET ${source} NAME_WE)
-    add_executable(${TEST_TARGET} ${source})
+    add_executable(${TEST_TARGET} ${source} get-model.cpp)
    install(TARGETS ${TEST_TARGET} RUNTIME)
    target_link_libraries(${TEST_TARGET} PRIVATE common)
 endfunction()
@@ -8,14 +8,20 @@ endfunction()
 function(llama_test_executable name source)
    get_filename_component(TEST_TARGET ${source} NAME_WE)
    add_test(NAME ${name} COMMAND $<TARGET_FILE:${TEST_TARGET}> ${ARGN})
+    set_property(TEST ${name} PROPERTY LABELS "main")
 endfunction()

 function(llama_build_and_test_executable source)
+    llama_build_and_test_executable_with_label(${source} "main")
+endfunction()
+
+function(llama_build_and_test_executable_with_label source label)
    get_filename_component(TEST_TARGET ${source} NAME_WE)
-    add_executable(${TEST_TARGET} ${source})
+    add_executable(${TEST_TARGET} ${source} get-model.cpp)
    install(TARGETS ${TEST_TARGET} RUNTIME)
    target_link_libraries(${TEST_TARGET} PRIVATE common)
    add_test(NAME ${TEST_TARGET} COMMAND $<TARGET_FILE:${TEST_TARGET}> ${ARGN})
+    set_property(TEST ${TEST_TARGET} PROPERTY LABELS ${label})
 endfunction()

 # llama_build_and_test_executable(test-double-float.cpp) # SLOW
@@ -49,10 +55,12 @@ llama_build_and_test_executable(test-llama-grammar.cpp)
 llama_build_and_test_executable(test-grad0.cpp)
 # llama_build_and_test_executable(test-opt.cpp) # SLOW
 llama_build_and_test_executable(test-backend-ops.cpp)
-llama_build_and_test_executable(test-autorelease.cpp)

 llama_build_and_test_executable(test-rope.cpp)

+llama_build_and_test_executable_with_label(test-model-load-cancel.cpp "model")
+llama_build_and_test_executable_with_label(test-autorelease.cpp "model")
+
 # dummy executable - not installed
 get_filename_component(TEST_TARGET test-c.c NAME_WE)
 add_executable(${TEST_TARGET} test-c.c)
@@ -0,0 +1,21 @@
+#include <cstdio>
+#include <cstdlib>
+#include <cstring>
+
+#include "get-model.h"
+
+char * get_model_or_exit(int argc, char *argv[]) {
+    char * model_path;
+    if (argc > 1) {
+        model_path = argv[1];
+
+    } else {
+        model_path = getenv("LLAMACPP_TEST_MODELFILE");
+        if (!model_path || strlen(model_path) == 0) {
+            fprintf(stderr, "\033[33mWARNING: No model file provided. Skipping this test. Set LLAMACPP_TEST_MODELFILE=<gguf_model_path> to silence this warning and run this test.\n\033[0m");
+            exit(EXIT_SUCCESS);
+        }
+    }
+
+    return model_path;
+}
@@ -0,0 +1,2 @@
+#pragma once
+char * get_model_or_exit(int, char*[]);
@@ -5,19 +5,15 @@
 #include <thread>

 #include "llama.h"
+#include "get-model.h"

 // This creates a new context inside a pthread and then tries to exit cleanly.
 int main(int argc, char ** argv) {
-    if (argc < 2) {
-        printf("Usage: %s model.gguf\n", argv[0]);
-        return 0; // intentionally return success
-    }
+    auto * model_path = get_model_or_exit(argc, argv);

-    const std::string fname = argv[1];
-
-    std::thread([&fname]() {
+    std::thread([&model_path]() {
        llama_backend_init(false);
-        auto * model = llama_load_model_from_file(fname.c_str(), llama_model_default_params());
+        auto * model = llama_load_model_from_file(model_path, llama_model_default_params());
        auto * ctx = llama_new_context_with_model(model, llama_context_default_params());
        llama_free(ctx);
        llama_free_model(model);
@@ -0,0 +1,27 @@
+#include "llama.h"
+#include "get-model.h"
+
+#include <cstdlib>
+
+int main(int argc, char *argv[] ) {
+    auto * model_path = get_model_or_exit(argc, argv);
+    auto * file = fopen(model_path, "r");
+    if (file == nullptr) {
+        fprintf(stderr, "no model at '%s' found\n", model_path);
+        return EXIT_FAILURE;
+    }
+
+    fprintf(stderr, "using '%s'\n", model_path);
+    fclose(file);
+
+    llama_backend_init(false);
+    auto params = llama_model_params{};
+    params.use_mmap = false;
+    params.progress_callback = [](float progress, void * ctx){
+        (void) ctx;
+        return progress > 0.50;
+    };
+    auto * model = llama_load_model_from_file(model_path, params);
+    llama_backend_free();
+    return model == nullptr ? EXIT_SUCCESS : EXIT_FAILURE;
+}
Author	SHA1	Message	Date
crasm	413e7b0559	ci : add model tests + script wrapper (#4586 ) * scripts : add lib.sh and lib_test.sh * scripts : stub out new ci-run.sh script * scripts : switch to PascalCase for functions This looks a little odd at first, but I find it very useful as a convention to know if a command is part of our code vs a builtin. * scripts : add some fancy conversion from snake_case to PascalCase * Add venv to ci/run.sh * Revert scripts work * scripts : add wrapper script for local use of ci/run.sh * Simplify .gitignore for tests, clang-tidy fixes * Label all ctest tests * ci : ctest uses -L main * Attempt at writing ctest_with_model * Update test-model-load-cancel * ci : add ctest_with_model for debug and release ggml-ci * Fix gg_get_model function ggml-ci * got stuck on CMake * Add get_model.cpp to tests/CMakeLists.txt ggml-ci * Fix README.md output for ctest_with_model ggml-ci * workflows : use `-L main` for all ctest ggml-ci * Fixes * GG_RUN_CTEST_MODELFILE => LLAMACPP_TESTMODELFILE * Always show warning rather than failing if model file variable is not set * scripts : update usage text for ci-run.sh	2024-01-26 14:18:00 +02:00
Paul Tsochantaris	6dd3c28c9c	metal : remove unused `n_buffers` and `buffers` (#5129 )	2024-01-26 14:16:07 +02:00
Riceball LEE	38b431de23	gguf : fix "general.alignment" type in gguf_reader.py (#5136 )	2024-01-26 11:10:28 +02:00
Georgi Gerganov	aad0b01d73	readme : update hot topics	2024-01-26 10:52:33 +02:00
Kawrakow	1182cf4d4f	Another bucket sort (#5109 ) * Initial bucket sort * Bucket sort: slightly better version * Bucket sort: another minor improvement --------- Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>	2024-01-26 09:14:39 +02:00
XiaotaoChen	fe54033b69	readme : add MobileVLM 1.7B/3B to the supported models list (#5107 ) Co-authored-by: Chenxiaotao03 <chenxiaotao03@meituan.com>	2024-01-25 22:14:32 +02:00