llama_cpp 0.7.1 → 0.8.0

data/ext/llama_cpp/src/ggml.c

@@ -13537,7 +13537,7 @@ static void ggml_compute_forward_rope_f16(
                         dst_data[n_dims]     = GGML_FP32_TO_FP16(x2*cos_block_theta - x3*sin_block_theta);
                         dst_data[n_dims/2*3] = GGML_FP32_TO_FP16(x2*sin_block_theta + x3*cos_block_theta);
                     }
-                } if (!is_neox) {
+                } else if (!is_neox) {
                     for (int64_t i0 = 0; i0 < ne0; i0 += 2) {
                         const float cos_theta = cosf(theta);
                         const float sin_theta = sinf(theta);
@@ -19170,6 +19170,7 @@ void ggml_graph_export(const struct ggml_cgraph * cgraph, const char * fname) {
 
                             if (idx == -1) {
                                 fprintf(stderr, "%s: failed to find tensor, arg = %d, node = %d\n", __func__, j, i);
+                                fclose(fout);
                                 return;
                             }
 
@@ -20844,7 +20845,7 @@ struct gguf_kv {
 };
 
 struct gguf_header {
-    uint32_t magic;
+    char magic[4];
     uint32_t version;
     uint64_t n_tensors; // GGUFv2
     uint64_t n_kv;      // GGUFv2
@@ -20914,7 +20915,7 @@ static bool gguf_fread_str_v1(FILE * file, struct gguf_str * p, size_t * offset)
 struct gguf_context * gguf_init_empty(void) {
     struct gguf_context * ctx = GGML_ALIGNED_MALLOC(sizeof(struct gguf_context));
 
-    ctx->header.magic     = GGUF_MAGIC;
+    memcpy(ctx->header.magic, GGUF_MAGIC, sizeof(ctx->header.magic));
     ctx->header.version   = GGUF_VERSION;
     ctx->header.n_tensors = 0;
     ctx->header.n_kv      = 0;
@@ -20940,16 +20941,18 @@ struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_p
     // offset from start of file
     size_t offset = 0;
 
-    uint32_t magic = 0;
+    char magic[4];
 
     // check the magic before making allocations
     {
         gguf_fread_el(file, &magic, sizeof(magic), &offset);
 
-        if (magic != GGUF_MAGIC) {
-            fprintf(stderr, "%s: invalid magic number %08x\n", __func__, magic);
-            fclose(file);
-            return NULL;
+        for (uint32_t i = 0; i < sizeof(magic); i++) {
+            if (magic[i] != GGUF_MAGIC[i]) {
+                fprintf(stderr, "%s: invalid magic characters %s.\n", __func__, magic);
+                fclose(file);
+                return NULL;
+            }
         }
     }
 
@@ -20959,7 +20962,8 @@ struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_p
 
     // read the header
     {
-        ctx->header.magic = magic;
+        strncpy(ctx->header.magic, magic, 4);
+
 
         ctx->kv    = NULL;
         ctx->infos = NULL;

data/ext/llama_cpp/src/ggml.h

@@ -231,8 +231,9 @@
 #define GGML_EXIT_SUCCESS 0
 #define GGML_EXIT_ABORTED 1
 
-#define GGUF_MAGIC   0x46554747 // "GGUF"
-#define GGUF_VERSION 2
+#define GGUF_MAGIC "GGUF"
+
+#define GGUF_VERSION 3
 
 #define GGUF_DEFAULT_ALIGNMENT 32
 

data/ext/llama_cpp/src/k_quants.c

@@ -46,7 +46,7 @@ inline static int32_t vaddvq_s32(int32x4_t v) {
 #if defined(_MSC_VER) || defined(__MINGW32__)
 #include <intrin.h>
 #else
-#if !defined(__riscv)
+#if !defined(__riscv) && !defined(__s390__)
 #include <immintrin.h>
 #endif
 #endif
@@ -462,12 +462,9 @@ void quantize_row_q2_K(const float * restrict x, void * restrict vy, int k) {
 }
 
 size_t ggml_quantize_q2_K(const float * restrict src, void * restrict dst, int n, int k, int64_t * restrict hist) {
-    const int nb = k / QK_K;
-
-    // TODO - collect histograms - although, at a second thought, I don't really care about them
-    (void)hist;
+    (void)hist; // TODO: collect histograms
 
-    for (int j = 0; j < nb; j += k) {
+    for (int j = 0; j < n; j += k) {
         block_q2_K * restrict y = (block_q2_K *)dst + j/QK_K;
         quantize_row_q2_K_reference(src + j, y, k);
     }
@@ -678,12 +675,9 @@ void quantize_row_q3_K(const float * restrict x, void * restrict vy, int k) {
 }
 
 size_t ggml_quantize_q3_K(const float * restrict src, void * restrict dst, int n, int k, int64_t * restrict hist) {
-    const int nb = k / QK_K;
-
-    // TODO - collect histograms - although, at a second thought, I don't really care about them
-    (void)hist;
+    (void)hist; // TODO: collect histograms
 
-    for (int j = 0; j < nb; j += k) {
+    for (int j = 0; j < n; j += k) {
         block_q3_K * restrict y = (block_q3_K *)dst + j/QK_K;
         quantize_row_q3_K_reference(src + j, y, k);
     }
@@ -846,9 +840,9 @@ void quantize_row_q4_K(const float * restrict x, void * restrict vy, int k) {
 
 size_t ggml_quantize_q4_K(const float * restrict src, void * restrict dst, int n, int k, int64_t * restrict hist) {
     assert(k % QK_K == 0);
-    const int nb = k / QK_K;
     (void)hist; // TODO: collect histograms
-    for (int j = 0; j < nb; j += k) {
+
+    for (int j = 0; j < n; j += k) {
         block_q4_K * restrict y = (block_q4_K *)dst + j/QK_K;
         quantize_row_q4_K_reference(src + j, y, k);
     }
@@ -1052,9 +1046,9 @@ void quantize_row_q5_K(const float * restrict x, void * restrict vy, int k) {
 
 size_t ggml_quantize_q5_K(const float * restrict src, void * restrict dst, int n, int k, int64_t * restrict hist) {
     assert(k % QK_K == 0);
-    const int nb = k / QK_K;
-    (void)hist;
-    for (int j = 0; j < nb; j += k) {
+    (void)hist; // TODO: collect histograms
+
+    for (int j = 0; j < n; j += k) {
         block_q5_K * restrict y = (block_q5_K *)dst + j/QK_K;
         quantize_row_q5_K_reference(src + j, y, k);
     }
@@ -1200,11 +1194,9 @@ void quantize_row_q6_K(const float * restrict x, void * restrict vy, int k) {
 
 size_t ggml_quantize_q6_K(const float * src, void * dst, int n, int k, int64_t * hist) {
     assert(k % QK_K == 0);
-    const int nb = k / QK_K;
-
-    (void)hist; // TODO
+    (void)hist; // TODO: collect histograms
 
-    for (int j = 0; j < nb; j += k) {
+    for (int j = 0; j < n; j += k) {
         block_q6_K * restrict y = (block_q6_K *)dst + j/QK_K;
         quantize_row_q6_K_reference(src + j, y, k);
     }