@fugood/llama.node 0.2.2 → 0.3.0

package/src/llama.cpp/{ggml.h → ggml/include/ggml.h}

@@ -254,18 +254,8 @@
 
 #define GGML_PAD(x, n) (((x) + (n) - 1) & ~((n) - 1))
 
-#define GGML_ASSERT(x) \
-    do { \
-        if (!(x)) { \
-            fflush(stdout); \
-            fprintf(stderr, "GGML_ASSERT: %s:%d: %s\n", __FILE__, __LINE__, #x); \
-            ggml_print_backtrace(); \
-            abort(); \
-        } \
-    } while (0)
-
 #ifndef NDEBUG
-#define GGML_UNREACHABLE() GGML_ASSERT(!"statement should not be reached")
+#define GGML_UNREACHABLE() do { fprintf(stderr, "statement should be unreachable\n"); abort(); } while(0)
 #elif defined(__GNUC__)
 #define GGML_UNREACHABLE() __builtin_unreachable()
 #elif defined(_MSC_VER)
@@ -274,6 +264,17 @@
 #define GGML_UNREACHABLE() ((void) 0)
 #endif
 
+#ifdef __cplusplus
+#define GGML_NORETURN [[noreturn]]
+#elif defined(_MSC_VER)
+#define GGML_NORETURN __declspec(noreturn)
+#else
+#define GGML_NORETURN _Noreturn
+#endif
+
+#define GGML_ABORT(...) ggml_abort(__FILE__, __LINE__, __VA_ARGS__)
+#define GGML_ASSERT(x) if (!(x)) GGML_ABORT("GGML_ASSERT(%s) failed", #x)
+
 // used to copy the number of elements and stride in bytes of tensors into local variables.
 // main purpose is to reduce code duplication and improve readability.
 //
@@ -312,10 +313,19 @@
     GGML_TENSOR_LOCALS(int64_t, ne,  dst,  ne) \
     GGML_TENSOR_LOCALS(size_t,  nb,  dst,  nb)
 
+#define GGML_TENSOR_BINARY_OP_LOCALS01 \
+    GGML_TENSOR_LOCALS(int64_t, ne0, src0, ne) \
+    GGML_TENSOR_LOCALS(size_t,  nb0, src0, nb) \
+    GGML_TENSOR_LOCALS(int64_t, ne1, src1, ne) \
+    GGML_TENSOR_LOCALS(size_t,  nb1, src1, nb)
+
 #ifdef  __cplusplus
 extern "C" {
 #endif
 
+    GGML_NORETURN GGML_ATTRIBUTE_FORMAT(3, 4)
+    GGML_API void ggml_abort(const char * file, int line, const char * fmt, ...);
+
     enum ggml_status {
         GGML_STATUS_ALLOC_FAILED = -2,
         GGML_STATUS_FAILED = -1,
@@ -377,6 +387,9 @@ extern "C" {
         GGML_TYPE_F64     = 28,
         GGML_TYPE_IQ1_M   = 29,
         GGML_TYPE_BF16    = 30,
+        GGML_TYPE_Q4_0_4_4 = 31,
+        GGML_TYPE_Q4_0_4_8 = 32,
+        GGML_TYPE_Q4_0_8_8 = 33,
         GGML_TYPE_COUNT,
     };
 
@@ -418,6 +431,9 @@ extern "C" {
         GGML_FTYPE_MOSTLY_IQ4_XS  = 22, // except 1d tensors
         GGML_FTYPE_MOSTLY_IQ1_M   = 23, // except 1d tensors
         GGML_FTYPE_MOSTLY_BF16    = 24, // except 1d tensors
+        GGML_FTYPE_MOSTLY_Q4_0_4_4 = 25, // except 1d tensors
+        GGML_FTYPE_MOSTLY_Q4_0_4_8 = 26, // except 1d tensors
+        GGML_FTYPE_MOSTLY_Q4_0_8_8 = 27, // except 1d tensors
     };
 
     // available tensor operations:
@@ -585,11 +601,7 @@ extern "C" {
         struct ggml_tensor * grad;
         struct ggml_tensor * src[GGML_MAX_SRC];
 
-        // performance
-        int     perf_runs;
-        int64_t perf_cycles;
-        int64_t perf_time_us;
-
+        // source tensor and offset for views
         struct ggml_tensor * view_src;
         size_t               view_offs;
 
@@ -599,7 +611,7 @@ extern "C" {
 
         void * extra; // extra things e.g. for ggml-cuda.cu
 
-        char padding[8];
+        // char padding[4];
     };
 
     static const size_t GGML_TENSOR_SIZE = sizeof(struct ggml_tensor);
@@ -628,8 +640,11 @@ extern "C" {
         GGML_CGRAPH_EVAL_ORDER_COUNT
     };
 
+    typedef uint32_t ggml_bitset_t;
+
     struct ggml_hash_set {
         size_t size;
+        ggml_bitset_t * used;
         struct ggml_tensor ** keys;
     };
 
@@ -643,14 +658,9 @@ extern "C" {
         struct ggml_tensor ** grads;
         struct ggml_tensor ** leafs;
 
-        struct ggml_hash_set visited_hash_table;
+        struct ggml_hash_set visited_hash_set;
 
         enum ggml_cgraph_eval_order order;
-
-        // performance
-        int     perf_runs;
-        int64_t perf_cycles;
-        int64_t perf_time_us;
     };
 
     // scratch buffer
@@ -667,28 +677,6 @@ extern "C" {
         bool   no_alloc;   // don't allocate memory for the tensor data
     };
 
-
-    // compute types
-
-    // NOTE: the INIT or FINALIZE pass is not scheduled unless explicitly enabled.
-    // This behavior was changed since https://github.com/ggerganov/llama.cpp/pull/1995.
-    enum ggml_task_type {
-        GGML_TASK_TYPE_INIT = 0,
-        GGML_TASK_TYPE_COMPUTE,
-        GGML_TASK_TYPE_FINALIZE,
-    };
-
-    struct ggml_compute_params {
-        enum ggml_task_type type;
-
-        // ith = thread index, nth = number of threads
-        int ith, nth;
-
-        // work buffer for all threads
-        size_t wsize;
-        void * wdata;
-    };
-
     // numa strategies
     enum ggml_numa_strategy {
         GGML_NUMA_STRATEGY_DISABLED   = 0,
@@ -717,8 +705,6 @@ extern "C" {
     GGML_API int64_t ggml_cycles(void);
     GGML_API int64_t ggml_cycles_per_ms(void);
 
-    GGML_API void    ggml_print_backtrace(void);
-
     // accepts a UTF-8 path, even on Windows
     GGML_API FILE *  ggml_fopen(const char * fname, const char * mode);
 
@@ -733,9 +719,9 @@ extern "C" {
     GGML_API GGML_CALL size_t  ggml_nbytes      (const struct ggml_tensor * tensor);
     GGML_API           size_t  ggml_nbytes_pad  (const struct ggml_tensor * tensor); // same as ggml_nbytes() but padded to GGML_MEM_ALIGN
 
-    GGML_API GGML_CALL int    ggml_blck_size(enum ggml_type type);
-    GGML_API GGML_CALL size_t ggml_type_size(enum ggml_type type);             // size in bytes for all elements in a block
-    GGML_API GGML_CALL size_t ggml_row_size (enum ggml_type type, int64_t ne); // size in bytes for all elements in a row
+    GGML_API GGML_CALL int64_t ggml_blck_size(enum ggml_type type);
+    GGML_API GGML_CALL size_t  ggml_type_size(enum ggml_type type);             // size in bytes for all elements in a block
+    GGML_API GGML_CALL size_t  ggml_row_size (enum ggml_type type, int64_t ne); // size in bytes for all elements in a row
 
     GGML_DEPRECATED(
     GGML_API double ggml_type_sizef(enum ggml_type type), // ggml_type_size()/ggml_blck_size() as float
@@ -756,7 +742,6 @@ extern "C" {
     GGML_API enum ggml_type ggml_ftype_to_ggml_type(enum ggml_ftype ftype);
 
     GGML_API GGML_CALL bool ggml_is_transposed(const struct ggml_tensor * tensor);
-    GGML_API GGML_CALL bool ggml_is_contiguous(const struct ggml_tensor * tensor);
     GGML_API GGML_CALL bool ggml_is_permuted  (const struct ggml_tensor * tensor);
     GGML_API GGML_CALL bool ggml_is_empty     (const struct ggml_tensor * tensor);
     GGML_API           bool ggml_is_scalar    (const struct ggml_tensor * tensor);
@@ -765,9 +750,16 @@ extern "C" {
     GGML_API           bool ggml_is_3d        (const struct ggml_tensor * tensor);
     GGML_API           int  ggml_n_dims       (const struct ggml_tensor * tensor); // returns 1 for scalars
 
+    GGML_API GGML_CALL bool ggml_is_contiguous  (const struct ggml_tensor * tensor);
+    GGML_API GGML_CALL bool ggml_is_contiguous_0(const struct ggml_tensor * tensor); // same as ggml_is_contiguous()
+    GGML_API GGML_CALL bool ggml_is_contiguous_1(const struct ggml_tensor * tensor); // contiguous for dims >= 1
+    GGML_API GGML_CALL bool ggml_is_contiguous_2(const struct ggml_tensor * tensor); // contiguous for dims >= 2
+
     GGML_API bool ggml_are_same_shape (const struct ggml_tensor * t0, const struct ggml_tensor * t1);
     GGML_API bool ggml_are_same_stride(const struct ggml_tensor * t0, const struct ggml_tensor * t1);
 
+    GGML_API bool ggml_can_repeat(const struct ggml_tensor * t0, const struct ggml_tensor * t1);
+
     // use this to compute the memory overhead of a tensor
     GGML_API size_t ggml_tensor_overhead(void);
 
@@ -1461,7 +1453,6 @@ extern "C" {
     // rotary position embedding
     // if mode & 1 == 1, skip n_past elements (NOT SUPPORTED)
     // if mode & 2 == 1, GPT-NeoX style
-    // if mode & 4 == 1, ChatGLM style
     //
     // b is an int32 vector with size a->ne[2], it contains the positions
     // c is freq factors (e.g. phi3-128k), (optional)
@@ -1470,8 +1461,7 @@ extern "C" {
             struct ggml_tensor  * a,
             struct ggml_tensor  * b,
             int                   n_dims,
-            int                   mode,
-            int                   n_ctx);
+            int                   mode);
 
     // in-place, returns view(a)
     GGML_API struct ggml_tensor * ggml_rope_inplace(
@@ -1479,8 +1469,7 @@ extern "C" {
             struct ggml_tensor  * a,
             struct ggml_tensor  * b,
             int                   n_dims,
-            int                   mode,
-            int                   n_ctx);
+            int                   mode);
 
     // custom RoPE
     GGML_API struct ggml_tensor * ggml_rope_ext(
@@ -1490,8 +1479,7 @@ extern "C" {
             struct ggml_tensor  * c,
             int                   n_dims,
             int                   mode,
-            int                   n_ctx,
-            int                   n_orig_ctx,
+            int                   n_ctx_orig,
             float                 freq_base,
             float                 freq_scale,
             float                 ext_factor,
@@ -1507,8 +1495,7 @@ extern "C" {
             struct ggml_tensor  * c,
             int                   n_dims,
             int                   mode,
-            int                   n_ctx,
-            int                   n_orig_ctx,
+            int                   n_ctx_orig,
             float                 freq_base,
             float                 freq_scale,
             float                 ext_factor,
@@ -1522,8 +1509,7 @@ extern "C" {
             struct ggml_tensor  * b,
             int                   n_dims,
             int                   mode,
-            int                   n_ctx,
-            int                   n_orig_ctx,
+            int                   n_ctx_orig,
             float                 freq_base,
             float                 freq_scale,
             float                 ext_factor,
@@ -1538,8 +1524,7 @@ extern "C" {
             struct ggml_tensor  * b,
             int                   n_dims,
             int                   mode,
-            int                   n_ctx,
-            int                   n_orig_ctx,
+            int                   n_ctx_orig,
             float                 freq_base,
             float                 freq_scale,
             float                 ext_factor,
@@ -1550,7 +1535,7 @@ extern "C" {
 
     // compute correction dims for YaRN RoPE scaling
     GGML_CALL void ggml_rope_yarn_corr_dims(
-        int n_dims, int n_orig_ctx, float freq_base, float beta_fast, float beta_slow, float dims[2]);
+        int n_dims, int n_ctx_orig, float freq_base, float beta_fast, float beta_slow, float dims[2]);
 
     // rotary position embedding backward, i.e compute dx from dy
     // a - dy
@@ -1561,16 +1546,13 @@ extern "C" {
             struct ggml_tensor  * c,
             int                   n_dims,
             int                   mode,
-            int                   n_ctx,
-            int                   n_orig_ctx,
+            int                   n_ctx_orig,
             float                 freq_base,
             float                 freq_scale,
             float                 ext_factor,
             float                 attn_factor,
             float                 beta_fast,
-            float                 beta_slow,
-            float                 xpos_base,
-            bool                  xpos_down);
+            float                 beta_slow);
 
     // clamp
     // in-place, returns view(a)
@@ -2028,8 +2010,8 @@ extern "C" {
 
     // ggml_graph_plan() has to be called before ggml_graph_compute()
     // when plan.work_size > 0, caller must allocate memory for plan.work_data
-    GGML_API struct ggml_cplan ggml_graph_plan            (const struct ggml_cgraph * cgraph, int n_threads /*= GGML_DEFAULT_N_THREADS*/);
-    GGML_API enum ggml_status  ggml_graph_compute         (      struct ggml_cgraph * cgraph, struct ggml_cplan * cplan);
+    GGML_API struct ggml_cplan ggml_graph_plan   (const struct ggml_cgraph * cgraph, int n_threads /*= GGML_DEFAULT_N_THREADS*/);
+    GGML_API enum ggml_status  ggml_graph_compute(      struct ggml_cgraph * cgraph, struct ggml_cplan * cplan);
     // same as ggml_graph_compute() but the work data is allocated as a part of the context
     // note: the drawback of this API is that you must have ensured that the context has enough memory for the work data
     GGML_API enum ggml_status  ggml_graph_compute_with_ctx(struct ggml_context * ctx, struct ggml_cgraph * cgraph, int n_threads);
@@ -2413,15 +2395,17 @@ extern "C" {
     GGML_API int ggml_cpu_has_wasm_simd  (void);
     GGML_API int ggml_cpu_has_blas       (void);
     GGML_API int ggml_cpu_has_cuda       (void);
-    GGML_API int ggml_cpu_has_clblast    (void);
     GGML_API int ggml_cpu_has_vulkan     (void);
     GGML_API int ggml_cpu_has_kompute    (void);
     GGML_API int ggml_cpu_has_gpublas    (void);
     GGML_API int ggml_cpu_has_sse3       (void);
     GGML_API int ggml_cpu_has_ssse3      (void);
     GGML_API int ggml_cpu_has_sycl       (void);
+    GGML_API int ggml_cpu_has_rpc        (void);
     GGML_API int ggml_cpu_has_vsx        (void);
     GGML_API int ggml_cpu_has_matmul_int8(void);
+    GGML_API int ggml_cpu_has_cann       (void);
+    GGML_API int ggml_cpu_has_llamafile  (void);
 
     //
     // Internal types and functions exposed for tests and benchmarks
@@ -2435,20 +2419,31 @@ extern "C" {
 #endif
     typedef void (*ggml_to_float_t)  (const void  * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
     typedef void (*ggml_from_float_t)(const float * GGML_RESTRICT x, void  * GGML_RESTRICT y, int64_t k);
-    typedef void (*ggml_vec_dot_t)   (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT x, size_t bx,
-                                      const void * GGML_RESTRICT y, size_t by, int nrc);
+    typedef void (*ggml_from_float_to_mat_t)
+                                     (const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t nr, int64_t k, int64_t bs);
+    typedef void (*ggml_vec_dot_t)  (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT x, size_t bx,
+                                       const void * GGML_RESTRICT y, size_t by, int nrc);
+    typedef void (*ggml_gemv_t)     (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT x,
+                                       const void * GGML_RESTRICT y, int nr, int nc);
+    typedef void (*ggml_gemm_t)     (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT x,
+                                       const void * GGML_RESTRICT y, int nr, int nc);
 
     typedef struct {
-        const char      * type_name;
-        int               blck_size;
-        size_t            type_size;
-        bool              is_quantized;
-        ggml_to_float_t   to_float;
-        ggml_from_float_t from_float;
-        ggml_from_float_t from_float_reference;
-        ggml_vec_dot_t    vec_dot;
-        enum ggml_type    vec_dot_type;
-        int64_t           nrows; // number of rows to process simultaneously;
+        const char             * type_name;
+        int64_t                  blck_size;
+        int64_t                  blck_size_interleave; // interleave elements in blocks
+        size_t                   type_size;
+        bool                     is_quantized;
+        ggml_to_float_t          to_float;
+        ggml_from_float_t        from_float;
+        ggml_from_float_t        from_float_ref;
+        ggml_from_float_to_mat_t from_float_to_mat;
+        ggml_vec_dot_t           vec_dot;
+        enum ggml_type           vec_dot_type;
+        int64_t                  nrows; // number of rows to process simultaneously
+        int64_t                  ncols; // number of columns to process simultaneously
+        ggml_gemv_t              gemv;
+        ggml_gemm_t              gemm;
     } ggml_type_traits_t;
 
     GGML_API ggml_type_traits_t ggml_internal_get_type_traits(enum ggml_type type);