llama_cpp 0.14.2 → 0.14.4

data/vendor/tmp/llama.cpp/ggml-vulkan.h

@@ -11,17 +11,6 @@ extern "C" {
 #define GGML_VK_MAX_DEVICES 16
 
 GGML_API void ggml_vk_instance_init(void);
-GGML_API void ggml_vk_init_cpu_assist(void);
-
-GGML_API void ggml_vk_preallocate_buffers_graph_cpu_assist(struct ggml_tensor * node);
-GGML_API void ggml_vk_preallocate_buffers_cpu_assist(void);
-GGML_API void ggml_vk_build_graph_cpu_assist(struct ggml_tensor * node, bool last_node);
-GGML_API bool ggml_vk_compute_forward_cpu_assist(struct ggml_compute_params * params, struct ggml_tensor * tensor);
-#ifdef GGML_VULKAN_CHECK_RESULTS
-void ggml_vk_check_results_1_cpu_assist(struct ggml_compute_params * params, struct ggml_tensor * tensor);
-#endif
-GGML_API void ggml_vk_graph_cleanup_cpu_assist(void);
-GGML_API void ggml_vk_free_cpu_assist(void);
 
 // backend API
 GGML_API GGML_CALL ggml_backend_t ggml_backend_vk_init(size_t dev_num);

data/vendor/tmp/llama.cpp/ggml.c

@@ -3,6 +3,7 @@
 
 #include "ggml-impl.h"
 #include "ggml-quants.h"
+#include "ggml.h"
 
 #if defined(_MSC_VER) || defined(__MINGW32__)
 #include <malloc.h> // using malloc.h with MSC/MINGW
@@ -43,6 +44,10 @@
 
 #if defined(_WIN32)
 
+#define WIN32_LEAN_AND_MEAN
+#ifndef NOMINMAX
+    #define NOMINMAX
+#endif
 #include <windows.h>
 
 typedef volatile LONG atomic_int;
@@ -273,8 +278,6 @@ inline static void * ggml_calloc(size_t num, size_t size) {
 #include <Accelerate/Accelerate.h>
 #if defined(GGML_USE_CLBLAST) // allow usage of CLBlast alongside Accelerate functions
 #include "ggml-opencl.h"
-#elif defined(GGML_USE_VULKAN)
-#include "ggml-vulkan.h"
 #endif
 #elif defined(GGML_USE_OPENBLAS)
 #if defined(GGML_BLAS_USE_MKL)
@@ -282,14 +285,8 @@ inline static void * ggml_calloc(size_t num, size_t size) {
 #else
 #include <cblas.h>
 #endif
-#elif defined(GGML_USE_CUBLAS)
-#include "ggml-cuda.h"
 #elif defined(GGML_USE_CLBLAST)
 #include "ggml-opencl.h"
-#elif defined(GGML_USE_VULKAN)
-#include "ggml-vulkan.h"
-#elif defined(GGML_USE_SYCL)
-#include "ggml-sycl.h"
 #endif
 
 // floating point type used to accumulate sums
@@ -432,6 +429,57 @@ int64_t ggml_cycles_per_ms(void) {
 #define ggml_perf_cycles_per_ms() 0
 #endif
 
+//
+// cross-platform UTF-8 file paths
+//
+
+#ifdef _WIN32
+static wchar_t * ggml_mbstowcs(const char * mbs) {
+    int wlen = MultiByteToWideChar(CP_UTF8, 0, mbs, -1, NULL, 0);
+    if (!wlen) {
+        errno = EINVAL;
+        return NULL;
+    }
+
+    wchar_t * wbuf = GGML_MALLOC(wlen * sizeof(wchar_t));
+    wlen = MultiByteToWideChar(CP_UTF8, 0, mbs, -1, wbuf, wlen);
+    if (!wlen) {
+        GGML_FREE(wbuf);
+        errno = EINVAL;
+        return NULL;
+    }
+
+    return wbuf;
+}
+#endif
+
+FILE * ggml_fopen(const char * fname, const char * mode) {
+#ifdef _WIN32
+    FILE * file = NULL;
+
+    // convert fname (UTF-8)
+    wchar_t * wfname = ggml_mbstowcs(fname);
+    if (wfname) {
+        // convert mode (ANSI)
+        wchar_t * wmode = GGML_MALLOC((strlen(mode) + 1) * sizeof(wchar_t));
+        wchar_t * wmode_p = wmode;
+        do {
+            *wmode_p++ = (wchar_t)*mode;
+        } while (*mode++);
+
+        // open file
+        file = _wfopen(wfname, wmode);
+
+        GGML_FREE(wfname);
+        GGML_FREE(wmode);
+    }
+
+    return file;
+#else
+    return fopen(fname, mode);
+#endif
+}
+
 //
 // cache line
 //
@@ -742,6 +790,18 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
         .vec_dot_type             = GGML_TYPE_Q8_K,
         .nrows                    = 1,
     },
+    [GGML_TYPE_IQ1_M] = {
+        .type_name                = "iq1_m",
+        .blck_size                = QK_K,
+        .type_size                = sizeof(block_iq1_m),
+        .is_quantized             = true,
+        .to_float                 = (ggml_to_float_t) dequantize_row_iq1_m,
+        .from_float               = NULL,
+        .from_float_reference     = NULL,
+        .vec_dot                  = ggml_vec_dot_iq1_m_q8_K,
+        .vec_dot_type             = GGML_TYPE_Q8_K,
+        .nrows                    = 1,
+    },
     [GGML_TYPE_IQ4_NL] = {
         .type_name                = "iq4_nl",
         .blck_size                = QK4_NL,
@@ -931,6 +991,101 @@ inline static float vaddvq_f32(float32x4_t v) {
     #define GGML_F16_VEC_REDUCE         GGML_F32Cx4_REDUCE
 #endif
 
+#elif defined(__AVX512F__)
+
+#define GGML_SIMD
+
+// F32 AVX512
+
+#define GGML_F32_STEP 64
+#define GGML_F32_EPR  16
+
+#define GGML_F32x16         __m512
+#define GGML_F32x16_ZERO    _mm512_setzero_ps()
+#define GGML_F32x16_SET1(x) _mm512_set1_ps(x)
+#define GGML_F32x16_LOAD    _mm512_loadu_ps
+#define GGML_F32x16_STORE   _mm512_storeu_ps
+// _mm512_fmadd_ps is defined in AVX512F so no guard is required
+#define GGML_F32x16_FMA(a, b, c) _mm512_fmadd_ps(b, c, a)
+#define GGML_F32x16_ADD     _mm512_add_ps
+#define GGML_F32x16_MUL     _mm512_mul_ps
+#define GGML_F32x16_REDUCE(res, x)                                    \
+do {                                                                  \
+    int offset = GGML_F32_ARR >> 1;                                   \
+    for (int i = 0; i < offset; ++i) {                                \
+        x[i] = _mm512_add_ps(x[i], x[offset+i]);                      \
+    }                                                                 \
+    offset >>= 1;                                                     \
+    for (int i = 0; i < offset; ++i) {                                \
+        x[i] = _mm512_add_ps(x[i], x[offset+i]);                      \
+    }                                                                 \
+    offset >>= 1;                                                     \
+    for (int i = 0; i < offset; ++i) {                                \
+        x[i] = _mm512_add_ps(x[i], x[offset+i]);                      \
+    }                                                                 \
+    res = _mm512_reduce_add_ps(x[0]);                                 \
+} while (0)
+
+// TODO: is this optimal ?
+
+#define GGML_F32_VEC        GGML_F32x16
+#define GGML_F32_VEC_ZERO   GGML_F32x16_ZERO
+#define GGML_F32_VEC_SET1   GGML_F32x16_SET1
+#define GGML_F32_VEC_LOAD   GGML_F32x16_LOAD
+#define GGML_F32_VEC_STORE  GGML_F32x16_STORE
+#define GGML_F32_VEC_FMA    GGML_F32x16_FMA
+#define GGML_F32_VEC_ADD    GGML_F32x16_ADD
+#define GGML_F32_VEC_MUL    GGML_F32x16_MUL
+#define GGML_F32_VEC_REDUCE GGML_F32x16_REDUCE
+
+// F16 AVX512
+
+// F16 AVX
+
+#define GGML_F16_STEP 64
+#define GGML_F16_EPR  16
+
+// AVX512 has FP16 extension (AVX512_FP16) but I don't have it on my machine so I use FP32 instead
+
+#define GGML_F32Cx16             __m512
+#define GGML_F32Cx16_ZERO        _mm512_setzero_ps()
+#define GGML_F32Cx16_SET1(x)     _mm512_set1_ps(x)
+
+// unlike  _mm256_cvt intrinsics that require F16C, _mm512_cvt is defined in AVX512F
+// so F16C guard isn't required
+#define GGML_F32Cx16_LOAD(x)     _mm512_cvtph_ps(_mm256_loadu_si256((__m256i *)(x)))
+#define GGML_F32Cx16_STORE(x, y) _mm256_storeu_si256((__m256i *)(x), _mm512_cvtps_ph(y, 0))
+
+#define GGML_F32Cx16_FMA(a, b, c) _mm512_fmadd_ps(b, c, a)
+#define GGML_F32Cx16_ADD         _mm512_add_ps
+#define GGML_F32Cx16_MUL         _mm512_mul_ps
+#define GGML_F32Cx16_REDUCE(res, x)                               \
+do {                                                              \
+    int offset = GGML_F32_ARR >> 1;                               \
+    for (int i = 0; i < offset; ++i) {                            \
+        x[i] = _mm512_add_ps(x[i], x[offset+i]);                  \
+    }                                                             \
+    offset >>= 1;                                                 \
+    for (int i = 0; i < offset; ++i) {                            \
+        x[i] = _mm512_add_ps(x[i], x[offset+i]);                  \
+    }                                                             \
+    offset >>= 1;                                                 \
+    for (int i = 0; i < offset; ++i) {                            \
+        x[i] = _mm512_add_ps(x[i], x[offset+i]);                  \
+    }                                                             \
+    res = _mm512_reduce_add_ps(x[0]);                             \
+} while (0)
+
+#define GGML_F16_VEC                GGML_F32Cx16
+#define GGML_F16_VEC_ZERO           GGML_F32Cx16_ZERO
+#define GGML_F16_VEC_SET1           GGML_F32Cx16_SET1
+#define GGML_F16_VEC_LOAD(p, i)     GGML_F32Cx16_LOAD(p)
+#define GGML_F16_VEC_STORE(p, r, i) GGML_F32Cx16_STORE(p, r[i])
+#define GGML_F16_VEC_FMA            GGML_F32Cx16_FMA
+#define GGML_F16_VEC_ADD            GGML_F32Cx16_ADD
+#define GGML_F16_VEC_MUL            GGML_F32Cx16_MUL
+#define GGML_F16_VEC_REDUCE         GGML_F32Cx16_REDUCE
+
 #elif defined(__AVX__)
 
 #define GGML_SIMD
@@ -2392,6 +2547,7 @@ enum ggml_type ggml_ftype_to_ggml_type(enum ggml_ftype ftype) {
         case GGML_FTYPE_MOSTLY_IQ2_XS:        wtype = GGML_TYPE_IQ2_XS;   break;
         case GGML_FTYPE_MOSTLY_IQ3_XXS:       wtype = GGML_TYPE_IQ3_XXS;  break;
         case GGML_FTYPE_MOSTLY_IQ1_S:         wtype = GGML_TYPE_IQ1_S;    break;
+        case GGML_FTYPE_MOSTLY_IQ1_M:         wtype = GGML_TYPE_IQ1_M;    break;
         case GGML_FTYPE_MOSTLY_IQ4_NL:        wtype = GGML_TYPE_IQ4_NL;   break;
         case GGML_FTYPE_MOSTLY_IQ4_XS:        wtype = GGML_TYPE_IQ4_XS;   break;
         case GGML_FTYPE_MOSTLY_IQ3_S:         wtype = GGML_TYPE_IQ3_S;    break;
@@ -2447,6 +2603,16 @@ static inline bool ggml_is_padded_1d(const struct ggml_tensor * tensor) {
         tensor->nb[3] == tensor->nb[2]*tensor->ne[2];
 }
 
+GGML_CALL bool ggml_is_empty(const struct ggml_tensor * tensor) {
+    for (int i = 0; i < GGML_MAX_DIMS; ++i) {
+        if (tensor->ne[i] == 0) {
+            // empty if any dimension has no elements
+            return true;
+        }
+    }
+    return false;
+}
+
 bool ggml_are_same_shape(const struct ggml_tensor * t0, const struct ggml_tensor * t1) {
     static_assert(GGML_MAX_DIMS == 4, "GGML_MAX_DIMS is not 4 - update this function");
 
@@ -2461,7 +2627,7 @@ bool ggml_are_same_shape(const struct ggml_tensor * t0, const struct ggml_tensor
 static inline bool ggml_can_repeat(const struct ggml_tensor * t0, const struct ggml_tensor * t1) {
     static_assert(GGML_MAX_DIMS == 4, "GGML_MAX_DIMS is not 4 - update this function");
 
-    return
+    return ggml_is_empty(t0) ? ggml_is_empty(t1) :
         (t1->ne[0]%t0->ne[0] == 0) &&
         (t1->ne[1]%t0->ne[1] == 0) &&
         (t1->ne[2]%t0->ne[2] == 0) &&
@@ -2545,14 +2711,8 @@ struct ggml_context * ggml_init(struct ggml_init_params params) {
             GGML_PRINT_DEBUG("%s: g_state initialized in %f ms\n", __func__, (t_end - t_start)/1000.0f);
         }
 
-#if defined(GGML_USE_CUBLAS)
-        ggml_init_cublas();
-#elif defined(GGML_USE_CLBLAST)
+#if defined(GGML_USE_CLBLAST)
         ggml_cl_init();
-#elif defined(GGML_USE_VULKAN)
-        ggml_vk_init_cpu_assist();
-#elif defined(GGML_USE_SYCL)
-        ggml_init_sycl();
 #endif
 
         ggml_setup_op_has_task_pass();
@@ -2772,7 +2932,7 @@ static struct ggml_tensor * ggml_new_tensor_impl(
         data_size *= ne[i];
     }
 
-    GGML_ASSERT(view_src == NULL || data_size + view_offs <= ggml_nbytes(view_src));
+    GGML_ASSERT(view_src == NULL || data_size == 0 || data_size + view_offs <= ggml_nbytes(view_src));
 
     void * data = view_src != NULL ? view_src->data : NULL;
     if (data != NULL) {
@@ -4413,45 +4573,38 @@ void ggml_mul_mat_set_prec(
 
 // ggml_mul_mat_id
 
+// NOTE: id will be removed in the future and instead all the experts listed in ids will be computed
+//       this will allow computing all the used experts in a single matrix multiplication
 struct ggml_tensor * ggml_mul_mat_id(
         struct ggml_context * ctx,
-        struct ggml_tensor  * const as[],
-        int                   n_as,
+        struct ggml_tensor  * as,
         struct ggml_tensor  * ids,
         int                   id,
         struct ggml_tensor  * b) {
 
     GGML_ASSERT(ids->type == GGML_TYPE_I32);
-    GGML_ASSERT(ids->ne[2] == 1 && ids->ne[3] == 1);
-    GGML_ASSERT(ids->ne[1] == b->ne[1]);
+    GGML_ASSERT(ids->ne[2] == 1 && ids->ne[3] == 1); // ids is 2d
+    GGML_ASSERT(ids->ne[1] == b->ne[1]); // must have an expert per b row
     GGML_ASSERT(ids->ne[2] == b->ne[2] && ids->ne[3] == b->ne[3]);
-    GGML_ASSERT(n_as > 0 && n_as <= GGML_MAX_SRC - 2);
-    GGML_ASSERT(id >= 0 && id < ids->ne[0]);
+    GGML_ASSERT(id >= 0 && id < ids->ne[0]); // valid id
+    GGML_ASSERT(as->ne[0] == b->ne[0]); // can_mul_mat
 
     bool is_node = false;
 
-    if (as[0]->grad || b->grad) {
+    if (as->grad || b->grad) {
         is_node = true;
     }
 
-    const int64_t ne[4] = { as[0]->ne[1], b->ne[1], b->ne[2], b->ne[3] };
+    const int64_t ne[4] = { as->ne[1], b->ne[1], b->ne[2], b->ne[3] };
     struct ggml_tensor * result = ggml_new_tensor(ctx, GGML_TYPE_F32, 4, ne);
 
     ggml_set_op_params_i32(result, 0, id);
-    ggml_set_op_params_i32(result, 1, n_as);
 
     result->op   = GGML_OP_MUL_MAT_ID;
     result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
-    result->src[0] = ids;
+    result->src[0] = as;
     result->src[1] = b;
-
-    for (int i = 0; i < n_as; i++) {
-        struct ggml_tensor * a = as[i];
-        GGML_ASSERT(ggml_are_same_shape(as[0], a));
-        GGML_ASSERT(ggml_can_mul_mat(a, b));
-        GGML_ASSERT(!ggml_is_transposed(a));
-        result->src[i + 2] = a;
-    }
+    result->src[2] = ids;
 
     return result;
 }
@@ -7992,6 +8145,7 @@ static void ggml_compute_forward_add(
         case GGML_TYPE_IQ2_XS:
         case GGML_TYPE_IQ3_XXS:
         case GGML_TYPE_IQ1_S:
+        case GGML_TYPE_IQ1_M:
         case GGML_TYPE_IQ4_NL:
         case GGML_TYPE_IQ4_XS:
         case GGML_TYPE_IQ3_S:
@@ -8274,6 +8428,7 @@ static void ggml_compute_forward_add1(
         case GGML_TYPE_IQ2_XS:
         case GGML_TYPE_IQ3_XXS:
         case GGML_TYPE_IQ1_S:
+        case GGML_TYPE_IQ1_M:
         case GGML_TYPE_IQ4_NL:
         case GGML_TYPE_IQ4_XS:
         case GGML_TYPE_IQ3_S:
@@ -8401,6 +8556,7 @@ static void ggml_compute_forward_acc(
         case GGML_TYPE_IQ2_XS:
         case GGML_TYPE_IQ3_XXS:
         case GGML_TYPE_IQ1_S:
+        case GGML_TYPE_IQ1_M:
         case GGML_TYPE_IQ4_NL:
         case GGML_TYPE_IQ4_XS:
         case GGML_TYPE_IQ3_S:
@@ -10785,10 +10941,9 @@ static void ggml_compute_forward_mul_mat_id(
         const struct ggml_compute_params * params,
               struct ggml_tensor * dst) {
 
-    const struct ggml_tensor * ids = dst->src[0];
+    const struct ggml_tensor * src0 = dst->src[0];
     const struct ggml_tensor * src1 = dst->src[1];
-
-    const struct ggml_tensor * src0 = dst->src[2]; // only for GGML_TENSOR_BINARY_OP_LOCALS
+    const struct ggml_tensor * ids = dst->src[2];
 
     GGML_TENSOR_BINARY_OP_LOCALS
 
@@ -10818,13 +10973,13 @@ static void ggml_compute_forward_mul_mat_id(
     GGML_ASSERT(nb1 <= nb2);
     GGML_ASSERT(nb2 <= nb3);
 
-    // broadcast factors
-    const int64_t r2 = ne12/ne02;
-    const int64_t r3 = ne13/ne03;
+    // broadcast is not supported with mmid
+    assert(ne12 == 1);
+    assert(ne13 == 1);
 
     // row groups
     const int id   = ggml_get_op_params_i32(dst, 0);
-    const int n_as = ggml_get_op_params_i32(dst, 1);
+    const int n_as = src0->ne[2];
 
     char * wdata_src1_end = (src1->type == vec_dot_type) ?
             (char *) params->wdata :
@@ -10884,7 +11039,7 @@ static void ggml_compute_forward_mul_mat_id(
             continue;
         }
 
-        const struct ggml_tensor * src0_cur = dst->src[cur_a + 2];
+        size_t src0_offset = cur_a*src0->nb[2];
 
         const void * wdata    = (src1->type == vec_dot_type) ? src1->data : params->wdata;
         const size_t row_size = ggml_row_size(vec_dot_type, ne10);
@@ -10919,9 +11074,6 @@ static void ggml_compute_forward_mul_mat_id(
             continue;
         }
 
-        assert(ne12 % ne02 == 0);
-        assert(ne13 % ne03 == 0);
-
         // block-tiling attempt
         const int64_t blck_0 = 16;
         const int64_t blck_1 = 16;
@@ -10938,14 +11090,14 @@ static void ggml_compute_forward_mul_mat_id(
                     const int64_t  i11 = MMID_MATRIX_ROW(cur_a, _i11);
 
                     // broadcast src0 into src1
-                    const int64_t i03 = i13/r3;
-                    const int64_t i02 = i12/r2;
+                    //const int64_t i03 = i13/r3;
+                    //const int64_t i02 = i12/r2;
 
                     const int64_t i1 = i11;
                     const int64_t i2 = i12;
                     const int64_t i3 = i13;
 
-                    const char * src0_row = (const char *) src0_cur->data + (0 + i02*nb02 + i03*nb03);
+                    const char * src0_row = (const char *) src0->data + src0_offset;
 
                     // desc: when src1 is not a contiguous memory block we have to calculate the offset using the strides
                     //       if it is, then we have either copied the data to params->wdata and made it contiguous or we are using
@@ -11010,7 +11162,6 @@ static void ggml_compute_forward_out_prod_f32(
     // nb01 >= nb00 - src0 is not transposed
     //   compute by src0 rows
 
-    // TODO: #if defined(GGML_USE_CUBLAS) ggml_cuda_out_prod
     // TODO: #if defined(GGML_USE_CLBLAST)
 
 #if defined(GGML_USE_ACCELERATE) || defined(GGML_USE_OPENBLAS)
@@ -11210,7 +11361,6 @@ static void ggml_compute_forward_out_prod_q_f32(
     // nb01 >= nb00 - src0 is not transposed
     //   compute by src0 rows
 
-    // TODO: #if defined(GGML_USE_CUBLAS) ggml_cuda_out_prod
     // TODO: #if defined(GGML_USE_ACCELERATE) || defined(GGML_USE_OPENBLAS) || defined(GGML_USE_CLBLAST)
 
     if (params->type == GGML_TASK_TYPE_INIT) {
@@ -11306,6 +11456,7 @@ static void ggml_compute_forward_out_prod(
         case GGML_TYPE_IQ2_XS:
         case GGML_TYPE_IQ3_XXS:
         case GGML_TYPE_IQ1_S:
+        case GGML_TYPE_IQ1_M:
         case GGML_TYPE_IQ4_NL:
         case GGML_TYPE_IQ4_XS:
         case GGML_TYPE_IQ3_S:
@@ -11497,6 +11648,7 @@ static void ggml_compute_forward_set(
         case GGML_TYPE_IQ2_XS:
         case GGML_TYPE_IQ3_XXS:
         case GGML_TYPE_IQ1_S:
+        case GGML_TYPE_IQ1_M:
         case GGML_TYPE_IQ4_NL:
         case GGML_TYPE_IQ4_XS:
         case GGML_TYPE_IQ3_S:
@@ -11720,6 +11872,7 @@ static void ggml_compute_forward_get_rows(
         case GGML_TYPE_IQ2_XS:
         case GGML_TYPE_IQ3_XXS:
         case GGML_TYPE_IQ1_S:
+        case GGML_TYPE_IQ1_M:
         case GGML_TYPE_IQ4_NL:
         case GGML_TYPE_IQ4_XS:
         case GGML_TYPE_IQ3_S:
@@ -12423,6 +12576,7 @@ static void ggml_compute_forward_alibi(
         case GGML_TYPE_IQ2_XS:
         case GGML_TYPE_IQ3_XXS:
         case GGML_TYPE_IQ1_S:
+        case GGML_TYPE_IQ1_M:
         case GGML_TYPE_IQ4_NL:
         case GGML_TYPE_IQ4_XS:
         case GGML_TYPE_IQ3_S:
@@ -12511,6 +12665,7 @@ static void ggml_compute_forward_clamp(
         case GGML_TYPE_IQ2_XS:
         case GGML_TYPE_IQ3_XXS:
         case GGML_TYPE_IQ1_S:
+        case GGML_TYPE_IQ1_M:
         case GGML_TYPE_IQ4_NL:
         case GGML_TYPE_IQ4_XS:
         case GGML_TYPE_IQ3_S:
@@ -15952,37 +16107,10 @@ static void ggml_compute_forward_cross_entropy_loss_back(
 static void ggml_compute_forward(struct ggml_compute_params * params, struct ggml_tensor * tensor) {
     GGML_ASSERT(params);
 
-    if (tensor->op == GGML_OP_NONE) {
-        return;
-    }
-
-#ifdef GGML_USE_CUBLAS
-    bool skip_cpu = ggml_cuda_compute_forward(params, tensor);
-    if (skip_cpu) {
+    if (tensor->op == GGML_OP_NONE || ggml_is_empty(tensor)) {
         return;
     }
-    GGML_ASSERT(tensor->src[0] == NULL || tensor->src[0]->backend == GGML_BACKEND_TYPE_CPU);
-    GGML_ASSERT(tensor->src[1] == NULL || tensor->src[1]->backend == GGML_BACKEND_TYPE_CPU);
-#elif defined(GGML_USE_VULKAN)
-    const bool skip_cpu = ggml_vk_compute_forward_cpu_assist(params, tensor);
-#ifdef GGML_VULKAN_CHECK_RESULTS
-    if (skip_cpu) {
-        ggml_vk_check_results_1_cpu_assist(params, tensor);
-    }
-#endif
-    if (skip_cpu) {
-        return;
-    }
-    GGML_ASSERT(tensor->src[0] == NULL || tensor->src[0]->backend == GGML_BACKEND_TYPE_CPU);
-    GGML_ASSERT(tensor->src[1] == NULL || tensor->src[1]->backend == GGML_BACKEND_TYPE_CPU);
-#endif // GGML_USE_CUBLAS
 
-#ifdef GGML_USE_SYCL
-    bool skip_cpu = ggml_sycl_compute_forward(params, tensor);
-    if (skip_cpu) {
-        return;
-    }
-#endif // GGML_USE_SYCL
     switch (tensor->op) {
         case GGML_OP_DUP:
             {
@@ -17834,6 +17962,12 @@ static void ggml_graph_compute_perf_stats_node(struct ggml_tensor * node, const
 static int ggml_get_n_tasks(struct ggml_tensor * node, int n_threads, int n_cur_threads) {
     int n_tasks = 0;
 
+    if (ggml_is_empty(node)) {
+        // no need to multi-thread a no-op
+        n_tasks = 1;
+        return n_tasks;
+    }
+
     switch (node->op) {
         case GGML_OP_CPY:
         case GGML_OP_DUP:
@@ -18319,13 +18453,13 @@ struct ggml_cplan ggml_graph_plan(const struct ggml_cgraph * cgraph, int n_threa
             case GGML_OP_MUL_MAT_ID:
                 {
                     cur = 0;
-                    const struct ggml_tensor * src0 = node->src[2];
+                    const struct ggml_tensor * src0 = node->src[0];
                     const struct ggml_tensor * src1 = node->src[1];
                     const enum ggml_type vec_dot_type = type_traits[src0->type].vec_dot_type;
                     if (src1->type != vec_dot_type) {
                         cur += ggml_row_size(vec_dot_type, ggml_nelements(src1));
                     }
-                    const int n_as = ggml_get_op_params_i32(node, 1);
+                    const int n_as = src0->ne[2];
                     cur += GGML_PAD(cur, sizeof(int64_t));       // align
                     cur += n_as * sizeof(int64_t);               // matrix_row_counts
                     cur += n_as * src1->ne[1] * sizeof(int64_t); // matrix_rows
@@ -18452,17 +18586,6 @@ enum ggml_status ggml_graph_compute(struct ggml_cgraph * cgraph, struct ggml_cpl
         }
     }
 
-#ifdef GGML_USE_VULKAN
-    for (int i = 0; i < cgraph->n_nodes; i++) {
-        ggml_vk_preallocate_buffers_graph_cpu_assist(cgraph->nodes[i]);
-    }
-    ggml_vk_preallocate_buffers_cpu_assist();
-
-    for (int i = 0; i < cgraph->n_nodes; i++) {
-        ggml_vk_build_graph_cpu_assist(cgraph->nodes[i], i == cgraph->n_nodes - 1);
-    }
-#endif
-
     const int n_threads = cplan->n_threads;
 
     struct ggml_compute_state_shared state_shared = {
@@ -18519,10 +18642,6 @@ enum ggml_status ggml_graph_compute(struct ggml_cgraph * cgraph, struct ggml_cpl
         }
     }
 
-#ifdef GGML_USE_VULKAN
-    ggml_vk_graph_cleanup_cpu_assist();
-#endif
-
     // performance stats (graph)
     {
         int64_t perf_cycles_cur  = ggml_perf_cycles()  - perf_start_cycles;
@@ -18657,7 +18776,7 @@ void ggml_graph_export(const struct ggml_cgraph * cgraph, const char * fname) {
 
     // write binary data
     {
-        FILE * fout = fopen(fname, "wb");
+        FILE * fout = ggml_fopen(fname, "wb");
 
         if (!fout) {
             fprintf(stderr, "%s: failed to open %s\n", __func__, fname);
@@ -18795,7 +18914,7 @@ struct ggml_cgraph * ggml_graph_import(const char * fname, struct ggml_context *
 
     // read file into data
     {
-        FILE * fin = fopen(fname, "rb");
+        FILE * fin = ggml_fopen(fname, "rb");
         if (!fin) {
             fprintf(stderr, "%s: failed to open %s\n", __func__, fname);
             return result;
@@ -19131,7 +19250,7 @@ static void ggml_graph_dump_dot_leaf_edge(FILE * fp, struct ggml_tensor * node,
 void ggml_graph_dump_dot(const struct ggml_cgraph * gb, const struct ggml_cgraph * gf, const char * filename) {
     char color[16];
 
-    FILE * fp = fopen(filename, "w");
+    FILE * fp = ggml_fopen(filename, "w");
     GGML_ASSERT(fp);
 
     fprintf(fp, "digraph G {\n");
@@ -20178,7 +20297,8 @@ void ggml_quantize_init(enum ggml_type type) {
         case GGML_TYPE_IQ2_XXS:
         case GGML_TYPE_IQ2_XS:
         case GGML_TYPE_IQ2_S:
-        case GGML_TYPE_IQ1_S:   iq2xs_init_impl(type); break;
+        case GGML_TYPE_IQ1_S:
+        case GGML_TYPE_IQ1_M:   iq2xs_init_impl(type); break;
         case GGML_TYPE_IQ3_XXS: iq3xs_init_impl(256); break;
         case GGML_TYPE_IQ3_S:   iq3xs_init_impl(512); break;
         default: // nothing
@@ -20203,7 +20323,8 @@ bool ggml_quantize_requires_imatrix(enum ggml_type type) {
     return
         type == GGML_TYPE_IQ2_XXS ||
         type == GGML_TYPE_IQ2_XS  ||
-        type == GGML_TYPE_IQ1_S;
+        type == GGML_TYPE_IQ1_S;//   ||
+        //type == GGML_TYPE_IQ1_M;
 }
 
 size_t ggml_quantize_chunk(
@@ -20247,6 +20368,7 @@ size_t ggml_quantize_chunk(
         case GGML_TYPE_IQ3_S:   result = quantize_iq3_s  (src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
         case GGML_TYPE_IQ2_S:   result = quantize_iq2_s  (src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
         case GGML_TYPE_IQ1_S:   result = quantize_iq1_s  (src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
+        case GGML_TYPE_IQ1_M:   result = quantize_iq1_m  (src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
         case GGML_TYPE_IQ4_NL:  result = quantize_iq4_nl (src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
 #if QK_K == 64
         case GGML_TYPE_IQ4_XS:  result = quantize_iq4_nl (src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
@@ -20449,7 +20571,7 @@ struct gguf_context * gguf_init_empty(void) {
 }
 
 struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_params params) {
-    FILE * file = fopen(fname, "rb");
+    FILE * file = ggml_fopen(fname, "rb");
     if (!file) {
         return NULL;
     }
@@ -21404,7 +21526,7 @@ static void gguf_write_to_buf(const struct gguf_context * ctx, struct gguf_buf *
 }
 
 void gguf_write_to_file(const struct gguf_context * ctx, const char * fname, bool only_meta) {
-    FILE * file = fopen(fname, "wb");
+    FILE * file = ggml_fopen(fname, "wb");
     if (!file) {
         GGML_ASSERT(false && "failed to open file for writing");
     }
@@ -21546,15 +21668,15 @@ int ggml_cpu_has_wasm_simd(void) {
 }
 
 int ggml_cpu_has_blas(void) {
-#if defined(GGML_USE_ACCELERATE) || defined(GGML_USE_OPENBLAS) || defined(GGML_USE_CUBLAS) || defined(GGML_USE_VULKAN) || defined(GGML_USE_CLBLAST) || defined(GGML_USE_SYCL)
+#if defined(GGML_USE_ACCELERATE) || defined(GGML_USE_OPENBLAS) || defined(GGML_USE_CUDA) || defined(GGML_USE_VULKAN) || defined(GGML_USE_CLBLAST) || defined(GGML_USE_SYCL)
     return 1;
 #else
     return 0;
 #endif
 }
 
-int ggml_cpu_has_cublas(void) {
-#if defined(GGML_USE_CUBLAS)
+int ggml_cpu_has_cuda(void) {
+#if defined(GGML_USE_CUDA)
     return 1;
 #else
     return 0;
@@ -21594,7 +21716,7 @@ int ggml_cpu_has_sycl(void) {
 }
 
 int ggml_cpu_has_gpublas(void) {
-    return ggml_cpu_has_cublas() || ggml_cpu_has_clblast() || ggml_cpu_has_vulkan() || ggml_cpu_has_kompute() ||
+    return ggml_cpu_has_cuda() || ggml_cpu_has_clblast() || ggml_cpu_has_vulkan() || ggml_cpu_has_kompute() ||
            ggml_cpu_has_sycl();
 }