llama_cpp 0.12.0 → 0.12.1

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

@@ -70,7 +70,7 @@ static_assert(sizeof(block_q8_1) == 2*sizeof(float) + QK8_1, "wrong q8_1 block s
 // 2-bit quantization
 // weight is represented as x = a * q + b
 // 16 blocks of 16 elements each
-// Effectively 2.5625 bits per weight
+// Effectively 2.625 bits per weight
 typedef struct {
     uint8_t scales[QK_K/16]; // scales and mins, quantized with 4 bits
     uint8_t qs[QK_K/4];      // quants
@@ -165,6 +165,22 @@ typedef struct {
 } block_q8_K;
 static_assert(sizeof(block_q8_K) == sizeof(float) + QK_K + QK_K/16*sizeof(int16_t), "wrong q8_K block size/padding");
 
+// (Almost) "true" 2-bit quantization.
+// Due to the need to use blocks as per ggml dsign, it ends up using
+// 2.0625 bpw because of the 16-bit scale for each block of 256.
+typedef struct {
+    ggml_fp16_t d;
+    uint16_t qs[QK_K/8];
+} block_iq2_xxs;
+static_assert(sizeof(block_iq2_xxs) == sizeof(ggml_fp16_t) + QK_K/8*sizeof(uint16_t), "wrong iq2_xxs block size/padding");
+
+// 2.3125 bpw quants
+typedef struct {
+    ggml_fp16_t d;
+    uint16_t qs[QK_K/8];
+    uint8_t  scales[QK_K/32];
+} block_iq2_xs;
+static_assert(sizeof(block_iq2_xs) == sizeof(ggml_fp16_t) + QK_K/8*sizeof(uint16_t) + QK_K/32, "wrong iq2_xs block size/padding");
 
 // Quantization
 void quantize_row_q4_0_reference(const float * restrict x, block_q4_0 * restrict y, int k);
@@ -180,6 +196,8 @@ void quantize_row_q4_K_reference(const float * restrict x, block_q4_K * restrict
 void quantize_row_q5_K_reference(const float * restrict x, block_q5_K * restrict y, int k);
 void quantize_row_q6_K_reference(const float * restrict x, block_q6_K * restrict y, int k);
 void quantize_row_q8_K_reference(const float * restrict x, block_q8_K * restrict y, int k);
+void quantize_row_iq2_xxs_reference(const float * restrict x, block_iq2_xxs * restrict y, int k);
+void quantize_row_iq2_xs_reference (const float * restrict x, block_iq2_xs  * restrict y, int k);
 
 void quantize_row_q4_0(const float * restrict x, void * restrict y, int k);
 void quantize_row_q4_1(const float * restrict x, void * restrict y, int k);
@@ -194,6 +212,8 @@ void quantize_row_q4_K(const float * restrict x, void * restrict y, int k);
 void quantize_row_q5_K(const float * restrict x, void * restrict y, int k);
 void quantize_row_q6_K(const float * restrict x, void * restrict y, int k);
 void quantize_row_q8_K(const float * restrict x, void * restrict y, int k);
+void quantize_row_iq2_xxs(const float * restrict x, void * restrict y, int k);
+void quantize_row_iq2_xs (const float * restrict x, void * restrict y, int k);
 
 // Dequantization
 void dequantize_row_q4_0(const block_q4_0 * restrict x, float * restrict y, int k);
@@ -209,6 +229,8 @@ void dequantize_row_q4_K(const block_q4_K * restrict x, float * restrict y, int
 void dequantize_row_q5_K(const block_q5_K * restrict x, float * restrict y, int k);
 void dequantize_row_q6_K(const block_q6_K * restrict x, float * restrict y, int k);
 void dequantize_row_q8_K(const block_q8_K * restrict x, float * restrict y, int k);
+void dequantize_row_iq2_xxs(const block_iq2_xxs * restrict x, float * restrict y, int k);
+void dequantize_row_iq2_xs (const block_iq2_xs  * restrict x, float * restrict y, int k);
 
 // Dot product
 void ggml_vec_dot_q4_0_q8_0(int n, float * restrict s, const void * restrict vx, const void * restrict vy);
@@ -222,3 +244,5 @@ void ggml_vec_dot_q3_K_q8_K(int n, float * restrict s, const void * restrict vx,
 void ggml_vec_dot_q4_K_q8_K(int n, float * restrict s, const void * restrict vx, const void * restrict vy);
 void ggml_vec_dot_q5_K_q8_K(int n, float * restrict s, const void * restrict vx, const void * restrict vy);
 void ggml_vec_dot_q6_K_q8_K(int n, float * restrict s, const void * restrict vx, const void * restrict vy);
+void ggml_vec_dot_iq2_xxs_q8_K(int n, float * restrict s, const void * restrict vx, const void * restrict vy);
+void ggml_vec_dot_iq2_xs_q8_K (int n, float * restrict s, const void * restrict vx, const void * restrict vy);

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

@@ -132,7 +132,7 @@ void ggml_print_backtrace(void) {
             "-ex", "bt -frame-info source-and-location",
             "-ex", "detach",
             "-ex", "quit",
-            NULL);
+            (char *) NULL);
     } else {
         waitpid(pid, NULL, 0);
     }
@@ -394,6 +394,12 @@ static const size_t CACHE_LINE_SIZE_F32 = CACHE_LINE_SIZE/sizeof(float);
 static void ggml_vec_dot_f32(const int n, float * restrict s, const float * restrict x, const float * restrict y);
 static void ggml_vec_dot_f16(const int n, float * restrict s, ggml_fp16_t * restrict x, ggml_fp16_t * restrict y);
 
+ggml_collect_imatrix_t g_imatrix_collect = NULL;
+
+void ggml_set_imatrix_collection(ggml_collect_imatrix_t imatrix_collect) {
+    g_imatrix_collect = imatrix_collect;
+}
+
 static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
     [GGML_TYPE_I8] = {
         .type_name                = "i8",
@@ -573,6 +579,28 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
         .vec_dot                  = ggml_vec_dot_q6_K_q8_K,
         .vec_dot_type             = GGML_TYPE_Q8_K,
     },
+    [GGML_TYPE_IQ2_XXS] = {
+        .type_name                = "iq2_xxs",
+        .blck_size                = QK_K,
+        .type_size                = sizeof(block_iq2_xxs),
+        .is_quantized             = true,
+        .to_float                 = (ggml_to_float_t) dequantize_row_iq2_xxs,
+        .from_float               = quantize_row_iq2_xxs,
+        .from_float_reference     = (ggml_from_float_t) quantize_row_iq2_xxs_reference,
+        .vec_dot                  = ggml_vec_dot_iq2_xxs_q8_K,
+        .vec_dot_type             = GGML_TYPE_Q8_K,
+    },
+    [GGML_TYPE_IQ2_XS] = {
+        .type_name                = "iq2_xs",
+        .blck_size                = QK_K,
+        .type_size                = sizeof(block_iq2_xs),
+        .is_quantized             = true,
+        .to_float                 = (ggml_to_float_t) dequantize_row_iq2_xs,
+        .from_float               = quantize_row_iq2_xs,
+        .from_float_reference     = (ggml_from_float_t) quantize_row_iq2_xs_reference,
+        .vec_dot                  = ggml_vec_dot_iq2_xs_q8_K,
+        .vec_dot_type             = GGML_TYPE_Q8_K,
+    },
     [GGML_TYPE_Q8_K] = {
         .type_name                = "q8_K",
         .blck_size                = QK_K,
@@ -2111,6 +2139,8 @@ enum ggml_type ggml_ftype_to_ggml_type(enum ggml_ftype ftype) {
         case GGML_FTYPE_MOSTLY_Q4_K:          wtype = GGML_TYPE_Q4_K;  break;
         case GGML_FTYPE_MOSTLY_Q5_K:          wtype = GGML_TYPE_Q5_K;  break;
         case GGML_FTYPE_MOSTLY_Q6_K:          wtype = GGML_TYPE_Q6_K;  break;
+        case GGML_FTYPE_MOSTLY_IQ2_XXS:       wtype = GGML_TYPE_IQ2_XXS;  break;
+        case GGML_FTYPE_MOSTLY_IQ2_XS:        wtype = GGML_TYPE_IQ2_XS;   break;
         case GGML_FTYPE_UNKNOWN:              wtype = GGML_TYPE_COUNT; break;
         case GGML_FTYPE_MOSTLY_Q4_1_SOME_F16: wtype = GGML_TYPE_COUNT; break;
     }
@@ -4299,13 +4329,13 @@ struct ggml_tensor * ggml_set_2d_inplace(
 static struct ggml_tensor * ggml_cpy_impl(
         struct ggml_context * ctx,
         struct ggml_tensor  * a,
-        struct ggml_tensor  * b,
-        bool inplace) {
+        struct ggml_tensor  * b) {
     GGML_ASSERT(ggml_nelements(a) == ggml_nelements(b));
 
     bool is_node = false;
 
-    if (!inplace && (a->grad || b->grad)) {
+    if (a->grad || b->grad) {
+        // inplace is false and either one have a grad
         is_node = true;
     }
 
@@ -4329,29 +4359,21 @@ struct ggml_tensor * ggml_cpy(
         struct ggml_context * ctx,
         struct ggml_tensor * a,
         struct ggml_tensor * b) {
-    return ggml_cpy_impl(ctx, a, b, false);
-}
-
-struct ggml_tensor * ggml_cpy_inplace(
-        struct ggml_context * ctx,
-        struct ggml_tensor * a,
-        struct ggml_tensor * b) {
-    return ggml_cpy_impl(ctx, a, b, true);
+    return ggml_cpy_impl(ctx, a, b);
 }
 
 // ggml_cont
 
 static struct ggml_tensor * ggml_cont_impl(
         struct ggml_context * ctx,
-        struct ggml_tensor  * a,
-        bool inplace) {
+        struct ggml_tensor  * a) {
     bool is_node = false;
 
-    if (!inplace && a->grad) {
+    if (a->grad) {
         is_node = true;
     }
 
-    struct ggml_tensor * result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
+    struct ggml_tensor * result = ggml_dup_tensor(ctx, a);
     ggml_format_name(result, "%s (cont)", a->name);
 
     result->op   = GGML_OP_CONT;
@@ -4364,13 +4386,7 @@ static struct ggml_tensor * ggml_cont_impl(
 struct ggml_tensor * ggml_cont(
         struct ggml_context * ctx,
         struct ggml_tensor * a) {
-    return ggml_cont_impl(ctx, a, false);
-}
-
-struct ggml_tensor * ggml_cont_inplace(
-        struct ggml_context * ctx,
-        struct ggml_tensor * a) {
-    return ggml_cont_impl(ctx, a, true);
+    return ggml_cont_impl(ctx, a);
 }
 
 // make contiguous, with new shape
@@ -7436,6 +7452,8 @@ static void ggml_compute_forward_add(
         case GGML_TYPE_Q4_K:
         case GGML_TYPE_Q5_K:
         case GGML_TYPE_Q6_K:
+        case GGML_TYPE_IQ2_XXS:
+        case GGML_TYPE_IQ2_XS:
             {
                 ggml_compute_forward_add_q_f32(params, src0, src1, dst);
             } break;
@@ -7700,6 +7718,8 @@ static void ggml_compute_forward_add1(
         case GGML_TYPE_Q4_K:
         case GGML_TYPE_Q5_K:
         case GGML_TYPE_Q6_K:
+        case GGML_TYPE_IQ2_XXS:
+        case GGML_TYPE_IQ2_XS:
             {
                 ggml_compute_forward_add1_q_f32(params, src0, src1, dst);
             } break;
@@ -7814,6 +7834,8 @@ static void ggml_compute_forward_acc(
         case GGML_TYPE_Q4_K:
         case GGML_TYPE_Q5_K:
         case GGML_TYPE_Q6_K:
+        case GGML_TYPE_IQ2_XXS:
+        case GGML_TYPE_IQ2_XS:
         default:
             {
                 GGML_ASSERT(false);
@@ -9704,10 +9726,10 @@ static void ggml_compute_forward_group_norm(
 #if defined(GGML_USE_ACCELERATE) || defined(GGML_USE_OPENBLAS)
 // helper function to determine if it is better to use BLAS or not
 // for large matrices, BLAS is faster
-static bool ggml_compute_forward_mul_mat_use_blas(
-        const struct ggml_tensor * src0,
-        const struct ggml_tensor * src1,
-              struct ggml_tensor * dst) {
+static bool ggml_compute_forward_mul_mat_use_blas(struct ggml_tensor * dst) {
+    const struct ggml_tensor * src0 = dst->src[0];
+    const struct ggml_tensor * src1 = dst->src[1];
+
     //const int64_t ne00 = src0->ne[0];
     //const int64_t ne01 = src0->ne[1];
 
@@ -9747,6 +9769,10 @@ static void ggml_compute_forward_mul_mat(
     const int ith = params->ith;
     const int nth = params->nth;
 
+    if (ith == 1 && g_imatrix_collect) {
+        g_imatrix_collect(src0, src1);
+    }
+
     const enum ggml_type type = src0->type;
 
     const bool src1_cont = ggml_is_contiguous(src1);
@@ -9787,7 +9813,7 @@ static void ggml_compute_forward_mul_mat(
 #endif
 
 #if defined(GGML_USE_ACCELERATE) || defined(GGML_USE_OPENBLAS)
-    if (ggml_compute_forward_mul_mat_use_blas(src0, src1, dst)) {
+    if (ggml_compute_forward_mul_mat_use_blas(dst)) {
         if (params->ith != 0) {
             return;
         }
@@ -10050,6 +10076,10 @@ static void ggml_compute_forward_mul_mat_id(
 
         const struct ggml_tensor * src0_cur = dst->src[cur_a + 2];
 
+        if (ith == 1 && g_imatrix_collect) {
+            g_imatrix_collect(src0_cur, src1);
+        }
+
         const void * wdata    = (src1->type == vec_dot_type) ? src1->data : params->wdata;
         const size_t row_size = ggml_row_size(vec_dot_type, ne10);
 
@@ -10455,6 +10485,8 @@ static void ggml_compute_forward_out_prod(
         case GGML_TYPE_Q4_K:
         case GGML_TYPE_Q5_K:
         case GGML_TYPE_Q6_K:
+        case GGML_TYPE_IQ2_XXS:
+        case GGML_TYPE_IQ2_XS:
             {
                 ggml_compute_forward_out_prod_q_f32(params, src0, src1, dst);
             } break;
@@ -10629,6 +10661,8 @@ static void ggml_compute_forward_set(
         case GGML_TYPE_Q4_K:
         case GGML_TYPE_Q5_K:
         case GGML_TYPE_Q6_K:
+        case GGML_TYPE_IQ2_XXS:
+        case GGML_TYPE_IQ2_XS:
         default:
             {
                 GGML_ASSERT(false);
@@ -10823,6 +10857,8 @@ static void ggml_compute_forward_get_rows(
         case GGML_TYPE_Q4_K:
         case GGML_TYPE_Q5_K:
         case GGML_TYPE_Q6_K:
+        case GGML_TYPE_IQ2_XXS:
+        case GGML_TYPE_IQ2_XS:
             {
                 ggml_compute_forward_get_rows_q(params, src0, src1, dst);
             } break;
@@ -11459,6 +11495,8 @@ static void ggml_compute_forward_alibi(
         case GGML_TYPE_Q4_K:
         case GGML_TYPE_Q5_K:
         case GGML_TYPE_Q6_K:
+        case GGML_TYPE_IQ2_XXS:
+        case GGML_TYPE_IQ2_XS:
         case GGML_TYPE_Q8_K:
         case GGML_TYPE_I8:
         case GGML_TYPE_I16:
@@ -11533,6 +11571,8 @@ static void ggml_compute_forward_clamp(
         case GGML_TYPE_Q4_K:
         case GGML_TYPE_Q5_K:
         case GGML_TYPE_Q6_K:
+        case GGML_TYPE_IQ2_XXS:
+        case GGML_TYPE_IQ2_XS:
         case GGML_TYPE_Q8_K:
         case GGML_TYPE_I8:
         case GGML_TYPE_I16:
@@ -16301,24 +16341,6 @@ static int ggml_get_n_tasks(struct ggml_tensor * node, int n_threads) {
 
                 //n_tasks = MIN(n_threads, MAX(1, nr0/128));
                 //printf("nr0 = %8d, nr1 = %8d, nr0*nr1 = %8d, n_tasks%d\n", nr0, nr1, nr0*nr1, n_tasks);
-
-#if defined(GGML_USE_CUBLAS)
-                if (ggml_cuda_can_mul_mat(node->src[0], node->src[1], node)) {
-                    n_tasks = 1; // TODO: this actually is doing nothing
-                                 //       the threads are still spinning
-                }
-#elif defined(GGML_USE_CLBLAST)
-                if (ggml_cl_can_mul_mat(node->src[0], node->src[1], node)) {
-                    n_tasks = 1; // TODO: this actually is doing nothing
-                                 //       the threads are still spinning
-                }
-#endif
-#if defined(GGML_USE_ACCELERATE) || defined(GGML_USE_OPENBLAS)
-                if (ggml_compute_forward_mul_mat_use_blas(node->src[0], node->src[1], node)) {
-                    n_tasks = 1; // TODO: this actually is doing nothing
-                                 //       the threads are still spinning
-                }
-#endif
             } break;
         case GGML_OP_MUL_MAT_ID:
             {
@@ -16491,6 +16513,7 @@ static thread_ret_t ggml_graph_compute_thread(void * data) {
             state->shared->node_n += 1;
             return (thread_ret_t) GGML_EXIT_ABORTED;
         }
+
         if (atomic_fetch_sub(&state->shared->n_active, 1) == 1) {
             // all other threads are finished and spinning
             // do finalize and init here so we don't have synchronize again
@@ -16556,14 +16579,18 @@ static thread_ret_t ggml_graph_compute_thread(void * data) {
         } else {
             // wait for other threads to finish
             const int last = node_n;
+
+            const bool do_yield = last < 0 || cgraph->nodes[last]->op == GGML_OP_MUL_MAT;
+
             while (true) {
                 // TODO: this sched_yield can have significant impact on the performance - either positive or negative
                 //       depending on the workload and the operating system.
                 //       since it is not clear what is the best approach, it should potentially become user-configurable
                 //       ref: https://github.com/ggerganov/ggml/issues/291
-#if defined(GGML_USE_ACCELERATE) || defined(GGML_USE_OPENBLAS)
-                sched_yield();
-#endif
+                // UPD:  adding the do_yield flag seems to resolve the issue universally
+                if (do_yield) {
+                    sched_yield();
+                }
 
                 node_n = atomic_load(&state->shared->node_n);
                 if (node_n != last) break;
@@ -16642,7 +16669,7 @@ struct ggml_cplan ggml_graph_plan(struct ggml_cgraph * cgraph, int n_threads) {
                     } else
 #endif
 #if defined(GGML_USE_ACCELERATE) || defined(GGML_USE_OPENBLAS)
-                    if (ggml_compute_forward_mul_mat_use_blas(node->src[0], node->src[1], node)) {
+                    if (ggml_compute_forward_mul_mat_use_blas(node)) {
                         if (node->src[0]->type != GGML_TYPE_F32) {
                             // here we need memory just for single 2D matrix from src0
                             cur = ggml_type_size(GGML_TYPE_F32)*(node->src[0]->ne[0]*node->src[0]->ne[1]);
@@ -18661,6 +18688,18 @@ size_t ggml_quantize_chunk(enum ggml_type type, const float * src, void * dst, i
                 block_q6_K * block = (block_q6_K*)dst + start / QK_K;
                 result = ggml_quantize_q6_K(src + start, block, n, n, hist);
             } break;
+        case GGML_TYPE_IQ2_XXS:
+            {
+                GGML_ASSERT(start % QK_K == 0);
+                block_iq2_xxs * block = (block_iq2_xxs*)dst + start / QK_K;
+                result = ggml_quantize_iq2_xxs(src + start, block, n, n, hist);
+            } break;
+        case GGML_TYPE_IQ2_XS:
+            {
+                GGML_ASSERT(start % QK_K == 0);
+                block_iq2_xs * block = (block_iq2_xs*)dst + start / QK_K;
+                result = ggml_quantize_iq2_xs(src + start, block, n, n, hist);
+            } break;
         case GGML_TYPE_F16:
             {
                 int elemsize = sizeof(ggml_fp16_t);
@@ -19016,8 +19055,8 @@ struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_p
                 (int64_t) info->ne[3];
 
             if (ne % ggml_blck_size(info->type) != 0) {
-                fprintf(stderr, "%s: tensor '%s' number of elements (%" PRId64 ") is not a multiple of block size (%d)\n",
-                        __func__, info->name.data, ne, ggml_blck_size(info->type));
+                fprintf(stderr, "%s: tensor '%s' of type %d (%s) number of elements (%" PRId64 ") is not a multiple of block size (%d)\n",
+                        __func__, info->name.data, (int)info->type, ggml_type_name(info->type), ne, ggml_blck_size(info->type));
                 fclose(file);
                 gguf_free(ctx);
                 return NULL;

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

@@ -218,7 +218,9 @@
 #define GGML_MAX_PARAMS         2048
 #define GGML_MAX_CONTEXTS       64
 #define GGML_MAX_SRC            10
+#ifndef GGML_MAX_NAME
 #define GGML_MAX_NAME           64
+#endif
 #define GGML_MAX_OP_PARAMS      64
 #define GGML_DEFAULT_N_THREADS  4
 #define GGML_DEFAULT_GRAPH_SIZE 2048
@@ -339,6 +341,8 @@ extern "C" {
         GGML_TYPE_Q5_K = 13,
         GGML_TYPE_Q6_K = 14,
         GGML_TYPE_Q8_K = 15,
+        GGML_TYPE_IQ2_XXS = 16,
+        GGML_TYPE_IQ2_XS  = 17,
         GGML_TYPE_I8,
         GGML_TYPE_I16,
         GGML_TYPE_I32,
@@ -373,6 +377,8 @@ extern "C" {
         GGML_FTYPE_MOSTLY_Q4_K = 12, // except 1d tensors
         GGML_FTYPE_MOSTLY_Q5_K = 13, // except 1d tensors
         GGML_FTYPE_MOSTLY_Q6_K = 14, // except 1d tensors
+        GGML_FTYPE_MOSTLY_IQ2_XXS = 15, // except 1d tensors
+        GGML_FTYPE_MOSTLY_IQ2_XS  = 16, // except 1d tensors
     };
 
     // available tensor operations:
@@ -1159,22 +1165,11 @@ extern "C" {
             struct ggml_tensor  * a,
             struct ggml_tensor  * b);
 
-    // a -> b, in-place, return view(b)
-    GGML_API struct ggml_tensor * ggml_cpy_inplace(
-            struct ggml_context * ctx,
-            struct ggml_tensor  * a,
-            struct ggml_tensor  * b);
-
     // make contiguous
     GGML_API struct ggml_tensor * ggml_cont(
             struct ggml_context * ctx,
             struct ggml_tensor  * a);
 
-    // make contiguous, in-place
-    GGML_API struct ggml_tensor * ggml_cont_inplace(
-            struct ggml_context * ctx,
-            struct ggml_tensor  * a);
-
     // make contiguous, with new shape
     GGML_API struct ggml_tensor * ggml_cont_1d(
             struct ggml_context * ctx,
@@ -2067,9 +2062,17 @@ extern "C" {
     GGML_API size_t ggml_quantize_q4_K(const float * src, void * dst, int n, int k, int64_t * hist);
     GGML_API size_t ggml_quantize_q5_K(const float * src, void * dst, int n, int k, int64_t * hist);
     GGML_API size_t ggml_quantize_q6_K(const float * src, void * dst, int n, int k, int64_t * hist);
+    GGML_API size_t ggml_quantize_iq2_xxs(const float * src, void * dst, int n, int k, int64_t * hist);
+    GGML_API size_t ggml_quantize_iq2_xs (const float * src, void * dst, int n, int k, int64_t * hist);
 
     GGML_API size_t ggml_quantize_chunk(enum ggml_type type, const float * src, void * dst, int start, int n, int64_t * hist);
 
+    //
+    // Importance matrix
+    //
+    typedef void(*ggml_collect_imatrix_t)(const struct ggml_tensor * src0, const struct ggml_tensor * src1);
+    GGML_API void ggml_set_imatrix_collection(ggml_collect_imatrix_t imatrix_collect);
+
     //
     // gguf
     //