llama_cpp 0.14.0 → 0.14.2

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

@@ -470,6 +470,19 @@ static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
         .type_size                = sizeof(int32_t),
         .is_quantized             = false,
     },
+    [GGML_TYPE_I64] = {
+        .type_name                = "i64",
+        .blck_size                = 1,
+        .type_size                = sizeof(int64_t),
+        .is_quantized             = false,
+    },
+    [GGML_TYPE_F64] = {
+        .type_name                = "f64",
+        .blck_size                = 1,
+        .type_size                = sizeof(double),
+        .is_quantized             = false,
+        .nrows                    = 1,
+    },
     [GGML_TYPE_F32] = {
         .type_name                = "f32",
         .blck_size                = 1,
@@ -857,7 +870,7 @@ inline static float vaddvq_f32(float32x4_t v) {
     #define GGML_F16x8              float16x8_t
     #define GGML_F16x8_ZERO         vdupq_n_f16(0.0f)
     #define GGML_F16x8_SET1(x)      vdupq_n_f16(x)
-    #define GGML_F16x8_LOAD(x)      vld1q_f16((const __fp16 *)(x))
+    #define GGML_F16x8_LOAD(x)      vld1q_f16((const ggml_fp16_internal_t *)(x))
     #define GGML_F16x8_STORE        vst1q_f16
     #define GGML_F16x8_FMA(a, b, c) vfmaq_f16(a, b, c)
     #define GGML_F16x8_ADD          vaddq_f16
@@ -900,7 +913,7 @@ inline static float vaddvq_f32(float32x4_t v) {
     #define GGML_F32Cx4              float32x4_t
     #define GGML_F32Cx4_ZERO         vdupq_n_f32(0.0f)
     #define GGML_F32Cx4_SET1(x)      vdupq_n_f32(x)
-    #define GGML_F32Cx4_LOAD(x)      vcvt_f32_f16(vld1_f16((const __fp16 *)(x)))
+    #define GGML_F32Cx4_LOAD(x)      vcvt_f32_f16(vld1_f16((const ggml_fp16_internal_t *)(x)))
     #define GGML_F32Cx4_STORE(x, y)  vst1_f16(x, vcvt_f16_f32(y))
     #define GGML_F32Cx4_FMA(a, b, c) vfmaq_f32(a, b, c)
     #define GGML_F32Cx4_ADD          vaddq_f32
@@ -1841,6 +1854,8 @@ static const char * GGML_OP_NAME[GGML_OP_COUNT] = {
     "FLASH_ATTN",
     "FLASH_FF",
     "FLASH_ATTN_BACK",
+    "SSM_CONV",
+    "SSM_SCAN",
     "WIN_PART",
     "WIN_UNPART",
     "GET_REL_POS",
@@ -1863,7 +1878,7 @@ static const char * GGML_OP_NAME[GGML_OP_COUNT] = {
     "CROSS_ENTROPY_LOSS_BACK",
 };
 
-static_assert(GGML_OP_COUNT == 74, "GGML_OP_COUNT != 74");
+static_assert(GGML_OP_COUNT == 76, "GGML_OP_COUNT != 76");
 
 static const char * GGML_OP_SYMBOL[GGML_OP_COUNT] = {
     "none",
@@ -1929,6 +1944,8 @@ static const char * GGML_OP_SYMBOL[GGML_OP_COUNT] = {
     "flash_attn(x)",
     "flash_ff(x)",
     "flash_attn_back(x)",
+    "ssm_conv(x)",
+    "ssm_scan(x)",
     "win_part(x)",
     "win_unpart(x)",
     "get_rel_pos(x)",
@@ -1951,7 +1968,7 @@ static const char * GGML_OP_SYMBOL[GGML_OP_COUNT] = {
     "cross_entropy_loss_back(x,y)",
 };
 
-static_assert(GGML_OP_COUNT == 74, "GGML_OP_COUNT != 74");
+static_assert(GGML_OP_COUNT == 76, "GGML_OP_COUNT != 76");
 
 static_assert(GGML_OP_POOL_COUNT == 2, "GGML_OP_POOL_COUNT != 2");
 
@@ -6154,6 +6171,108 @@ struct ggml_tensor * ggml_flash_attn_back(
     return result;
 }
 
+// ggml_ssm_conv
+
+struct ggml_tensor * ggml_ssm_conv(
+        struct ggml_context * ctx,
+        struct ggml_tensor  * s,
+        struct ggml_tensor  * x,
+        struct ggml_tensor  * c,
+        struct ggml_tensor  * sq) {
+    GGML_ASSERT(ggml_is_3d(s));
+    GGML_ASSERT(ggml_is_matrix(x));
+    GGML_ASSERT(ggml_is_matrix(c));
+    GGML_ASSERT(ggml_is_matrix(sq));
+    GGML_ASSERT(sq->type == GGML_TYPE_I32);
+
+    const int64_t d_conv   = c->ne[0];
+    const int64_t d_inner  = c->ne[1];
+    const int64_t n_tokens = x->ne[1];
+    const int64_t n_kv     = s->ne[2];
+
+    GGML_ASSERT( s->ne[0] == d_conv - 1);
+    GGML_ASSERT( s->ne[1] == d_inner);
+    GGML_ASSERT( x->ne[0] == d_inner);
+    GGML_ASSERT(sq->ne[0] == n_kv);
+    GGML_ASSERT(sq->ne[1] == n_tokens);
+
+    bool is_node = false;
+
+    if (s->grad || x->grad || c->grad || sq->grad) {
+        GGML_ASSERT(false); // TODO: implement
+        is_node = true;
+    }
+
+    // 2-in-1 concatenated x and conv_states, {d_inner, n_tokens} with {d_conv, d_inner, n_kv}
+    struct ggml_tensor * result = ggml_new_tensor_1d(ctx, GGML_TYPE_F32, (d_inner*n_tokens) + (d_conv*d_inner*n_kv));
+
+    result->op   = GGML_OP_SSM_CONV;
+    result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
+    result->src[0] = s;
+    result->src[1] = x;
+    result->src[2] = c;
+    result->src[3] = sq;
+
+    return result;
+}
+
+// ggml_ssm_scan
+
+struct ggml_tensor * ggml_ssm_scan(
+        struct ggml_context * ctx,
+        struct ggml_tensor  * s,
+        struct ggml_tensor  * x,
+        struct ggml_tensor  * dt,
+        struct ggml_tensor  * A,
+        struct ggml_tensor  * B,
+        struct ggml_tensor  * C,
+        struct ggml_tensor  * sq) {
+    GGML_ASSERT(ggml_is_contiguous(s));
+    GGML_ASSERT(ggml_is_contiguous(x));
+    GGML_ASSERT(ggml_is_contiguous(dt));
+    GGML_ASSERT(ggml_is_contiguous(A));
+    GGML_ASSERT(sq->type == GGML_TYPE_I32);
+    GGML_ASSERT(B->nb[0] == ggml_type_size(B->type));
+    GGML_ASSERT(C->nb[0] == ggml_type_size(C->type));
+    GGML_ASSERT(ggml_are_same_shape(x, dt));
+
+    {
+        const int64_t d_state  = s->ne[0];
+        const int64_t d_inner  = s->ne[1];
+        const int64_t n_tokens = x->ne[1];
+
+        GGML_ASSERT(x->ne[0] == d_inner);
+        GGML_ASSERT(A->ne[0] == d_state);
+        GGML_ASSERT(A->ne[1] == d_inner);
+        GGML_ASSERT(B->ne[0] == d_state);
+        GGML_ASSERT(B->ne[1] == n_tokens);
+        GGML_ASSERT(C->ne[0] == d_state);
+        GGML_ASSERT(C->ne[1] == n_tokens);
+    }
+
+    bool is_node = false;
+
+    if (s->grad || x->grad || dt->grad || A->grad || B->grad || C->grad || sq->grad) {
+        GGML_ASSERT(false); // TODO: implement
+        is_node = true;
+    }
+
+    // 2-in-1 concatenated y and ssm_states, {d_inner, n_tokens} with {d_state, d_inner, n_kv}
+    struct ggml_tensor * result = ggml_new_tensor_1d(ctx, GGML_TYPE_F32, ggml_nelements(x) + ggml_nelements(s));
+
+    result->op   = GGML_OP_SSM_SCAN;
+    result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
+    result->src[0] = s;
+    result->src[1] = x;
+    result->src[2] = dt;
+    result->src[3] = A;
+    result->src[4] = B;
+    result->src[5] = C;
+    result->src[6] = sq;
+
+    return result;
+}
+
 // ggml_win_part
 
 struct ggml_tensor * ggml_win_part(
@@ -11454,8 +11573,6 @@ static void ggml_compute_forward_get_rows_q(
     const struct ggml_tensor * src0 = dst->src[0];
     const struct ggml_tensor * src1 = dst->src[1];
 
-    assert(params->ith == 0);
-
     if (params->type == GGML_TASK_TYPE_INIT || params->type == GGML_TASK_TYPE_FINALIZE) {
         return;
     }
@@ -11463,7 +11580,7 @@ static void ggml_compute_forward_get_rows_q(
     GGML_TENSOR_BINARY_OP_LOCALS
 
     const int64_t nc = ne00;
-    const int64_t nr = ggml_nelements(src1); GGML_UNUSED(nr);
+    const int64_t nr = ggml_nelements(src1);
 
     const enum ggml_type type = src0->type;
     ggml_to_float_t const dequantize_row_q = type_traits[type].to_float;
@@ -11473,17 +11590,25 @@ static void ggml_compute_forward_get_rows_q(
     assert(nb00 == ggml_type_size(type));
     assert(ggml_nrows(dst) == nr);
 
-    // TODO: multi-thread
-    for (int64_t i12 = 0; i12 < ne12; ++i12) {
-        for (int64_t i11 = 0; i11 < ne11; ++i11) {
-            for (int64_t i10 = 0; i10 < ne10; ++i10) {
-                const int64_t i01 = *(int32_t *) ((char *) src1->data + i10*nb10 + i11*nb11 + i12*nb12);
+    const int ith = params->ith;
+    const int nth = params->nth;
 
-                dequantize_row_q(
-                        (const void *) ((char *) src0->data + i01*nb01 + i11*nb02 + i12*nb03),
-                             (float *) ((char *)  dst->data + i10*nb1  + i11*nb2  + i12*nb3), nc);
-            }
-        }
+    // rows per thread
+    const int dr = (nr + nth - 1)/nth;
+
+    // row range for this thread
+    const int ir0 = dr*ith;
+    const int ir1 = MIN(ir0 + dr, nr);
+
+    for (int64_t i = ir0; i < ir1; ++i) {
+        const int64_t i12 = i/(ne11*ne10);
+        const int64_t i11 = (i - i12*ne11*ne10)/ne10;
+        const int64_t i10 = (i - i12*ne11*ne10 - i11*ne10);
+        const int64_t i01 = *(int32_t *) ((char *) src1->data + i10*nb10 + i11*nb11 + i12*nb12);
+
+        dequantize_row_q(
+                (const void *) ((char *) src0->data + i01*nb01 + i11*nb02 + i12*nb03),
+                     (float *) ((char *)  dst->data + i10*nb1  + i11*nb2  + i12*nb3), nc);
     }
 }
 
@@ -11494,8 +11619,6 @@ static void ggml_compute_forward_get_rows_f16(
     const struct ggml_tensor * src0 = dst->src[0];
     const struct ggml_tensor * src1 = dst->src[1];
 
-    assert(params->ith == 0);
-
     if (params->type == GGML_TASK_TYPE_INIT || params->type == GGML_TASK_TYPE_FINALIZE) {
         return;
     }
@@ -11503,24 +11626,32 @@ static void ggml_compute_forward_get_rows_f16(
     GGML_TENSOR_BINARY_OP_LOCALS
 
     const int64_t nc = ne00;
-    const int64_t nr = ggml_nelements(src1); GGML_UNUSED(nr);
+    const int64_t nr = ggml_nelements(src1);
 
     assert(ne0  == nc);
     assert(ne02 == ne11);
     assert(nb00 == sizeof(ggml_fp16_t));
     assert(ggml_nrows(dst) == nr);
 
-    // TODO: multi-thread
-    for (int64_t i12 = 0; i12 < ne12; ++i12) {
-        for (int64_t i11 = 0; i11 < ne11; ++i11) {
-            for (int64_t i10 = 0; i10 < ne10; ++i10) {
-                const int64_t i01 = *(int32_t *) ((char *) src1->data + i10*nb10 + i11*nb11 + i12*nb12);
+    const int ith = params->ith;
+    const int nth = params->nth;
+
+    // rows per thread
+    const int dr = (nr + nth - 1)/nth;
 
-                ggml_fp16_to_fp32_row(
-                        (const void *) ((char *) src0->data + i01*nb01 + i11*nb02 + i12*nb03),
-                             (float *) ((char *)  dst->data + i10*nb1  + i11*nb2  + i12*nb3), nc);
-            }
-        }
+    // row range for this thread
+    const int ir0 = dr*ith;
+    const int ir1 = MIN(ir0 + dr, nr);
+
+    for (int64_t i = ir0; i < ir1; ++i) {
+        const int64_t i12 = i/(ne11*ne10);
+        const int64_t i11 = (i - i12*ne11*ne10)/ne10;
+        const int64_t i10 = (i - i12*ne11*ne10 - i11*ne10);
+        const int64_t i01 = *(int32_t *) ((char *) src1->data + i10*nb10 + i11*nb11 + i12*nb12);
+
+        ggml_fp16_to_fp32_row(
+                (const void *) ((char *) src0->data + i01*nb01 + i11*nb02 + i12*nb03),
+                     (float *) ((char *)  dst->data + i10*nb1  + i11*nb2  + i12*nb3), nc);
     }
 }
 
@@ -11531,8 +11662,6 @@ static void ggml_compute_forward_get_rows_f32(
     const struct ggml_tensor * src0 = dst->src[0];
     const struct ggml_tensor * src1 = dst->src[1];
 
-    assert(params->ith == 0);
-
     if (params->type == GGML_TASK_TYPE_INIT || params->type == GGML_TASK_TYPE_FINALIZE) {
         return;
     }
@@ -11540,24 +11669,32 @@ static void ggml_compute_forward_get_rows_f32(
     GGML_TENSOR_BINARY_OP_LOCALS
 
     const int64_t nc = ne00;
-    const int64_t nr = ggml_nelements(src1); GGML_UNUSED(nr);
+    const int64_t nr = ggml_nelements(src1);
 
     assert(ne0  == nc);
     assert(ne02 == ne11);
     assert(nb00 == sizeof(float));
     assert(ggml_nrows(dst) == nr);
 
-    // TODO: multi-thread
-    for (int64_t i12 = 0; i12 < ne12; ++i12) {
-        for (int64_t i11 = 0; i11 < ne11; ++i11) {
-            for (int64_t i10 = 0; i10 < ne10; ++i10) {
-                const int64_t i01 = *(int32_t *) ((char *) src1->data + i10*nb10 + i11*nb11 + i12*nb12);
+    const int ith = params->ith;
+    const int nth = params->nth;
 
-                ggml_vec_cpy_f32(nc,
-                        (float *) ((char *)  dst->data + i10*nb1  + i11*nb2  + i12*nb3),
-                        (float *) ((char *) src0->data + i01*nb01 + i11*nb02 + i12*nb03));
-            }
-        }
+    // rows per thread
+    const int dr = (nr + nth - 1)/nth;
+
+    // row range for this thread
+    const int ir0 = dr*ith;
+    const int ir1 = MIN(ir0 + dr, nr);
+
+    for (int64_t i = ir0; i < ir1; ++i) {
+        const int64_t i12 = i/(ne11*ne10);
+        const int64_t i11 = (i - i12*ne11*ne10)/ne10;
+        const int64_t i10 = (i - i12*ne11*ne10 - i11*ne10);
+        const int64_t i01 = *(int32_t *) ((char *) src1->data + i10*nb10 + i11*nb11 + i12*nb12);
+
+        ggml_vec_cpy_f32(nc,
+                (float *) ((char *)  dst->data + i10*nb1  + i11*nb2  + i12*nb3),
+                (float *) ((char *) src0->data + i01*nb01 + i11*nb02 + i12*nb03));
     }
 }
 
@@ -12294,6 +12431,8 @@ static void ggml_compute_forward_alibi(
         case GGML_TYPE_I8:
         case GGML_TYPE_I16:
         case GGML_TYPE_I32:
+        case GGML_TYPE_I64:
+        case GGML_TYPE_F64:
         case GGML_TYPE_COUNT:
             {
                 GGML_ASSERT(false);
@@ -12380,6 +12519,8 @@ static void ggml_compute_forward_clamp(
         case GGML_TYPE_I8:
         case GGML_TYPE_I16:
         case GGML_TYPE_I32:
+        case GGML_TYPE_I64:
+        case GGML_TYPE_F64:
         case GGML_TYPE_COUNT:
             {
                 GGML_ASSERT(false);
@@ -14771,6 +14912,257 @@ static void ggml_compute_forward_flash_attn_back(
     }
 }
 
+// ggml_compute_forward_ssm_conv
+
+static void ggml_compute_forward_ssm_conv_f32(
+        const struct ggml_compute_params * params,
+        struct ggml_tensor * dst) {
+    if (params->type == GGML_TASK_TYPE_INIT || params->type == GGML_TASK_TYPE_FINALIZE) {
+        return;
+    }
+
+    const struct ggml_tensor * src0 = dst->src[0]; // conv_state
+    const struct ggml_tensor * src1 = dst->src[1]; // x
+    const struct ggml_tensor * src2 = dst->src[2]; // conv1d.weight
+    const struct ggml_tensor * src3 = dst->src[3]; // state_seq
+
+    const int ith = params->ith;
+    const int nth = params->nth;
+
+    const int nc   = src2->ne[0]; // d_conv
+    const int nr   = src0->ne[1]; // d_inner
+    const int n_t  = src1->ne[1]; // n_tokens
+    const int n_kv = src0->ne[2]; // max number of sequences in the batch
+
+    GGML_ASSERT((nr*n_t) + (nc*nr*n_kv) == ggml_nelements(dst));
+    GGML_ASSERT(src0->nb[0] == sizeof(float));
+    GGML_ASSERT(src1->nb[0] == sizeof(float));
+    GGML_ASSERT(src2->nb[0] == sizeof(float));
+    GGML_ASSERT(src3->nb[0] == sizeof(int32_t));
+    GGML_ASSERT(src0->nb[1] == src0->ne[0]*sizeof(float));
+    // for use with the destination state offset between sequences
+    GGML_ASSERT(src2->nb[2] == src2->ne[1]*src2->ne[0]*sizeof(float));
+
+    // rows per thread
+    const int dr = (nr + nth - 1)/nth;
+
+    // row range for this thread
+    const int ir0 = dr*ith;
+    const int ir1 = MIN(ir0 + dr, nr);
+    const int ir  = ir1 - ir0;
+
+    if (n_kv > 1) {
+        // multiple sequences means it's hard to know when it's the first time a state is read,
+        // so copy them all over to the destination, just to be sure.
+        for (int i3 = 0; i3 < n_kv; ++i3) {
+            float * s0 = (float *) ((char *) src0->data + ir0*(src0->nb[1]) + i3*(src0->nb[2]));
+            float * s  = (float *) ((char *)  dst->data + ir0*(src2->nb[1]) + i3*(src2->nb[2]) + nr*n_t*sizeof(float));
+            // can't use memcpy because of d_conv vs d_conv - 1
+            for (int i1 = 0; i1 < ir; ++i1) {
+                for (int i0 = 0; i0 < nc - 1; ++i0) {
+                    // copy s0 to last (d_conv - 1) columns of s
+                    s[1 + i0 + i1*nc] = s0[i0 + i1*(nc - 1)];
+                }
+            }
+        }
+    }
+
+    for (int i2 = 0; i2 < n_t; ++i2) {
+        int32_t * sq = (int32_t *) ((char *) src3->data +  i2*(src3->nb[1])); // {n_kv, n_tokens}
+        float *   x  = (float *)   ((char *)  dst->data + ir0*sizeof(float) + i2*(nr*sizeof(float))); // {d_inner, n_tokens}
+        float *   s  = (float *)   ((char *)  dst->data + ir0*(src2->nb[1]) + sq[0]*(src2->nb[2]) + nr*n_t*sizeof(float)); // {d_conv, d_inner, n_kv}
+        float *   s0; // {d_conv - 1, d_inner, n_kv}
+        float *   x0 = (float *)   ((char *) src1->data + ir0*(src1->nb[0]) + i2*(src1->nb[1])); // {d_inner, n_tokens}
+        float *   c  = (float *)   ((char *) src2->data + ir0*(src2->nb[1])); // {d_conv, d_inner}
+        int ne0s0;
+
+        GGML_ASSERT(0 <= sq[0] && sq[0] < n_kv);
+
+        // avoid needing to copy the state for the first token
+        if (i2 == 0) {
+            s0 = (float *) ((char *) src0->data + ir0*(src0->nb[1]) + sq[0]*(src0->nb[2])); // {d_conv - 1, d_inner, n_kv}
+            ne0s0 = src0->ne[0];
+        } else {
+            // the source is the last (d_conv - 1) columns of the destination
+            s0 = s + 1;
+            ne0s0 = nc;
+        }
+
+        // d_inner
+        for (int i1 = 0; i1 < ir; ++i1) {
+            // shift state left
+            for (int i0 = 0; i0 < nc - 1; ++i0) {
+                s[i0 + i1*nc] = s0[i0 + i1*ne0s0];
+            }
+            // insert x on the last column
+            s[(nc - 1) + i1*nc] = x0[i1];
+        }
+
+        // handle copies when there are multiple output states
+        for (int i3 = 1; i3 < n_kv; ++i3) {
+            int32_t seq = sq[i3];
+            if (0 <= seq && seq < n_kv) {
+                float * s1 = s + (seq - sq[0])*nc*nr;
+                memcpy(s1, s, nc*ir*sizeof(float));
+            } else {
+                // stop at negative or too big seq_ids
+                break;
+            }
+        }
+
+        // it seems a little faster when this is separate from the state shift
+        for (int i1 = 0; i1 < ir; ++i1) {
+            // rowwise dot product
+            float sumf = 0.0f;
+            for (int i0 = 0; i0 < nc; ++i0) {
+                int i = i0 + i1*nc;
+                sumf += s[i] * c[i];
+            }
+            x[i1] = sumf;
+        }
+    }
+}
+
+static void ggml_compute_forward_ssm_conv(
+        const struct ggml_compute_params * params,
+        struct ggml_tensor * dst) {
+    switch (dst->src[0]->type) {
+        case GGML_TYPE_F32:
+            {
+                ggml_compute_forward_ssm_conv_f32(params, dst);
+            } break;
+        default:
+            {
+                GGML_ASSERT(false);
+            } break;
+    }
+}
+
+// ggml_compute_forward_ssm_scan
+
+static void ggml_compute_forward_ssm_scan_f32(
+        const struct ggml_compute_params * params,
+        struct ggml_tensor * dst) {
+    if (params->type == GGML_TASK_TYPE_INIT || params->type == GGML_TASK_TYPE_FINALIZE) {
+        return;
+    }
+
+    const struct ggml_tensor * src0 = dst->src[0]; // s
+    const struct ggml_tensor * src1 = dst->src[1]; // x
+    const struct ggml_tensor * src2 = dst->src[2]; // dt
+    const struct ggml_tensor * src3 = dst->src[3]; // A
+    const struct ggml_tensor * src4 = dst->src[4]; // B
+    const struct ggml_tensor * src5 = dst->src[5]; // C
+    const struct ggml_tensor * src6 = dst->src[6]; // sq
+
+    const int ith = params->ith;
+    const int nth = params->nth;
+
+    const int64_t nc   = src0->ne[0]; // d_state
+    const int64_t nr   = src0->ne[1]; // d_inner
+    const int64_t n_t  = src1->ne[1]; // number of tokens in the batch
+    const int64_t n_kv = src0->ne[2]; // max number of sequences in the batch
+
+    GGML_ASSERT(ggml_nelements(src1) + ggml_nelements(src0) == ggml_nelements(dst));
+    GGML_ASSERT(src0->nb[0] == sizeof(float));
+    GGML_ASSERT(src1->nb[0] == sizeof(float));
+    GGML_ASSERT(src2->nb[0] == sizeof(float));
+    GGML_ASSERT(src3->nb[0] == sizeof(float));
+    GGML_ASSERT(src4->nb[0] == sizeof(float));
+    GGML_ASSERT(src5->nb[0] == sizeof(float));
+    // required for the dot product between s and C, and when copying the states
+    GGML_ASSERT(src0->nb[1] == src0->ne[0]*sizeof(float));
+    // required for per-sequence offsets for states
+    GGML_ASSERT(src0->nb[2] == src0->ne[0]*src0->ne[1]*sizeof(float));
+    // required to get correct offset for state destination (i.e. src1->nb[2])
+    GGML_ASSERT(src1->nb[2] == src1->ne[0]*src1->ne[1]*sizeof(float));
+
+    // rows per thread
+    const int dr = (nr + nth - 1)/nth;
+
+    // row range for this thread
+    const int ir0 = dr*ith;
+    const int ir1 = MIN(ir0 + dr, nr);
+    const int ir  = ir1 - ir0;
+
+    if (n_kv > 1) {
+        // it's hard to know if the source states have already been copied
+        // when there are multiple, so copy them already.
+        for (int i3 = 0; i3 < n_kv; ++i3) {
+            float * s0 = (float *) ((char *) src0->data + ir0*(src0->nb[1]) + i3*(src0->nb[2]));
+            float * s  = (float *) ((char *)  dst->data + ir0*(src0->nb[1]) + i3*(src0->nb[2]) + src1->nb[2]);
+            memcpy(s, s0, nc*ir*sizeof(float));
+        }
+    }
+
+    for (int i2 = 0; i2 < n_t; ++i2) {
+        int32_t * sq = (int32_t *) ((char *) src6->data +  i2*(src6->nb[1])); // {n_kv, n_tokens}
+        float *   y  = (float *)   ((char *)  dst->data + ir0*(src1->nb[0]) +    i2*(src1->nb[1])); // {d_inner, n_tokens}
+        float *   s  = (float *)   ((char *)  dst->data + ir0*(src0->nb[1]) + sq[0]*(src0->nb[2]) + src1->nb[2]); // {d_state, d_inner, n_kv}
+        float *   s0;
+        float *   x  = (float *)   ((char *) src1->data + ir0*(src1->nb[0]) + i2*(src1->nb[1])); // {d_inner, n_tokens}
+        float *   dt = (float *)   ((char *) src2->data + ir0*(src2->nb[0]) + i2*(src2->nb[1])); // {d_inner, n_tokens}
+        float *   A  = (float *)   ((char *) src3->data + ir0*(src3->nb[1])); // {d_state, d_inner}
+        float *   B  = (float *)   ((char *) src4->data +  i2*(src4->nb[1])); // {d_state, n_tokens}
+        float *   C  = (float *)   ((char *) src5->data +  i2*(src5->nb[1])); // {d_state, n_tokens}
+
+        GGML_ASSERT(0 <= sq[0] && sq[0] < n_kv);
+
+        // avoid needing to copy the state for the first token
+        if (i2 == 0) {
+            s0 = (float *) ((char *) src0->data + ir0*(src0->nb[1]) + sq[0]*(src0->nb[2])); // {d_state, d_inner, n_kv}
+        } else {
+            // otherwise the source is the same as the destination
+            s0 = s;
+        }
+
+        // d_inner
+        for (int i1 = 0; i1 < ir; ++i1) {
+            // ref: https://github.com/state-spaces/mamba/blob/34076d664838588a3c97727b263478ab9f621a07/mamba_ssm/ops/triton/selective_state_update.py#L78
+            float dt_soft_plus = dt[i1] <= 20.0f ? log1pf(expf(dt[i1])) : dt[i1];
+            float x_dt = x[i1] * dt_soft_plus;
+            float sumf = 0.0f;
+            // d_state
+            for (int i0 = 0; i0 < nc; ++i0) {
+                int i = i0 + i1*nc;
+                // state = prev_state * dA + dB * x
+                float state = (s0[i] * expf(dt_soft_plus * A[i])) + (B[i0] * x_dt);
+                // y = rowwise_dotprod(state, C)
+                sumf += state * C[i0];
+                s[i] = state;
+            }
+            y[i1] = sumf;
+        }
+
+        // handle copies when there are multiple output states
+        for (int i3 = 1; i3 < n_kv; ++i3) {
+            int32_t seq = sq[i3];
+            if (0 <= seq && seq < n_kv) {
+                float * s1 = s + (seq - sq[0])*nc*nr;
+                memcpy(s1, s, nc*ir*sizeof(float));
+            } else {
+                // stop at negative or too big seq_ids
+                break;
+            }
+        }
+    }
+}
+
+static void ggml_compute_forward_ssm_scan(
+        const struct ggml_compute_params * params,
+        struct ggml_tensor * dst) {
+    switch (dst->src[0]->type) {
+        case GGML_TYPE_F32:
+            {
+                ggml_compute_forward_ssm_scan_f32(params, dst);
+            } break;
+        default:
+            {
+                GGML_ASSERT(false);
+            } break;
+    }
+}
+
 // ggml_compute_forward_win_part
 
 static void ggml_compute_forward_win_part_f32(
@@ -15830,6 +16222,14 @@ static void ggml_compute_forward(struct ggml_compute_params * params, struct ggm
                 bool masked = t != 0;
                 ggml_compute_forward_flash_attn_back(params, masked, tensor);
             } break;
+        case GGML_OP_SSM_CONV:
+            {
+                ggml_compute_forward_ssm_conv(params, tensor);
+            } break;
+        case GGML_OP_SSM_SCAN:
+            {
+                ggml_compute_forward_ssm_scan(params, tensor);
+            } break;
         case GGML_OP_WIN_PART:
             {
                 ggml_compute_forward_win_part(params, tensor);
@@ -16884,6 +17284,11 @@ static void ggml_compute_backward(struct ggml_context * ctx, struct ggml_tensor
             {
                 GGML_ASSERT(false); // not supported
             } break;
+        case GGML_OP_SSM_CONV:
+        case GGML_OP_SSM_SCAN:
+            {
+                GGML_ASSERT(false); // TODO: not implemented
+            } break;
         case GGML_OP_WIN_PART:
         case GGML_OP_WIN_UNPART:
         case GGML_OP_UNARY:
@@ -17426,7 +17831,7 @@ static void ggml_graph_compute_perf_stats_node(struct ggml_tensor * node, const
     node->perf_time_us += time_us_cur;
 }
 
-static int ggml_get_n_tasks(struct ggml_tensor * node, int n_threads) {
+static int ggml_get_n_tasks(struct ggml_tensor * node, int n_threads, int n_cur_threads) {
     int n_tasks = 0;
 
     switch (node->op) {
@@ -17507,6 +17912,12 @@ static int ggml_get_n_tasks(struct ggml_tensor * node, int n_threads) {
             {
                 n_tasks = n_threads;
             } break;
+        case GGML_OP_GET_ROWS:
+            {
+                // FIXME: the cost of launching additional threads decreases performance with GPU offloading
+                //n_tasks = MIN(n_threads, ggml_nelements(node->src[1]));
+                n_tasks = MIN(n_cur_threads, ggml_nelements(node->src[1]));
+            } break;
         case GGML_OP_SCALE:
         case GGML_OP_SET:
         case GGML_OP_CONT:
@@ -17514,7 +17925,6 @@ static int ggml_get_n_tasks(struct ggml_tensor * node, int n_threads) {
         case GGML_OP_VIEW:
         case GGML_OP_PERMUTE:
         case GGML_OP_TRANSPOSE:
-        case GGML_OP_GET_ROWS:
         case GGML_OP_GET_ROWS_BACK:
         case GGML_OP_DIAG:
             {
@@ -17590,6 +18000,11 @@ static int ggml_get_n_tasks(struct ggml_tensor * node, int n_threads) {
             {
                 n_tasks = n_threads;
             } break;
+        case GGML_OP_SSM_CONV:
+        case GGML_OP_SSM_SCAN:
+            {
+                n_tasks = n_threads;
+            } break;
         case GGML_OP_WIN_PART:
         case GGML_OP_WIN_UNPART:
         case GGML_OP_GET_REL_POS:
@@ -17727,7 +18142,7 @@ static thread_ret_t ggml_graph_compute_thread(void * data) {
                 /* FINALIZE */
                 struct ggml_tensor * node = cgraph->nodes[node_n];
                 if (GGML_OP_HAS_FINALIZE[node->op]) {
-                    params.nth = ggml_get_n_tasks(node, n_threads);
+                    params.nth = ggml_get_n_tasks(node, n_threads, state->shared->n_threads);
                     ggml_compute_forward(&params, node);
                 }
                 ggml_graph_compute_perf_stats_node(node, state->shared);
@@ -17737,7 +18152,7 @@ static thread_ret_t ggml_graph_compute_thread(void * data) {
             while (++node_n < cgraph->n_nodes) {
                 GGML_PRINT_DEBUG_5("%s: %d/%d\n", __func__, node_n, cgraph->n_nodes);
                 struct ggml_tensor * node = cgraph->nodes[node_n];
-                const int n_tasks = ggml_get_n_tasks(node, n_threads);
+                const int n_tasks = ggml_get_n_tasks(node, n_threads, state->shared->n_threads);
 
                 state->shared->perf_node_start_cycles  = ggml_perf_cycles();
                 state->shared->perf_node_start_time_us = ggml_perf_time_us();
@@ -17785,7 +18200,7 @@ static thread_ret_t ggml_graph_compute_thread(void * data) {
 
         /* INIT & COMPUTE */
         struct ggml_tensor * node = cgraph->nodes[node_n];
-        const int n_tasks = ggml_get_n_tasks(node, n_threads);
+        const int n_tasks = ggml_get_n_tasks(node, n_threads, state->shared->n_threads);
 
         struct ggml_compute_params params = {
             /*.type  =*/ GGML_TASK_TYPE_INIT,
@@ -17850,7 +18265,7 @@ struct ggml_cplan ggml_graph_plan(const struct ggml_cgraph * cgraph, int n_threa
     for (int i = 0; i < cgraph->n_nodes; i++) {
         struct ggml_tensor * node = cgraph->nodes[i];
 
-        const int n_tasks = ggml_get_n_tasks(node, n_threads);
+        const int n_tasks = ggml_get_n_tasks(node, n_threads, 1);
 
         max_tasks = MAX(max_tasks, n_tasks);
 
@@ -19784,133 +20199,6 @@ void ggml_quantize_free(void) {
     ggml_critical_section_end();
 }
 
-size_t ggml_quantize_q4_0(const float * src, void * dst, int n, int k, int64_t * hist) {
-    assert(k % QK4_0 == 0);
-    const int nb = k / QK4_0;
-
-    for (int b = 0; b < n; b += k) {
-        block_q4_0 * restrict y = (block_q4_0 *) dst + b/QK4_0;
-
-        quantize_row_q4_0_reference(src + b, y, k);
-
-        for (int i = 0; i < nb; i++) {
-            for (int j = 0; j < QK4_0; j += 2) {
-                const uint8_t vi0 = y[i].qs[j/2] & 0x0F;
-                const uint8_t vi1 = y[i].qs[j/2] >> 4;
-
-                hist[vi0]++;
-                hist[vi1]++;
-            }
-        }
-    }
-
-    return (n/QK4_0*sizeof(block_q4_0));
-}
-
-size_t ggml_quantize_q4_1(const float * src, void * dst, int n, int k, int64_t * hist) {
-    assert(k % QK4_1 == 0);
-    const int nb = k / QK4_1;
-
-    for (int b = 0; b < n; b += k) {
-        block_q4_1 * restrict y = (block_q4_1 *) dst + b/QK4_1;
-
-        quantize_row_q4_1_reference(src + b, y, k);
-
-        for (int i = 0; i < nb; i++) {
-            for (int j = 0; j < QK4_1; j += 2) {
-                const uint8_t vi0 = y[i].qs[j/2] & 0x0F;
-                const uint8_t vi1 = y[i].qs[j/2] >> 4;
-
-                hist[vi0]++;
-                hist[vi1]++;
-            }
-        }
-    }
-
-    return (n/QK4_1*sizeof(block_q4_1));
-}
-
-size_t ggml_quantize_q5_0(const float * src, void * dst, int n, int k, int64_t * hist) {
-    assert(k % QK5_0 == 0);
-    const int nb = k / QK5_0;
-
-    for (int b = 0; b < n; b += k) {
-        block_q5_0 * restrict y = (block_q5_0 *)dst + b/QK5_0;
-
-        quantize_row_q5_0_reference(src + b, y, k);
-
-        for (int i = 0; i < nb; i++) {
-            uint32_t qh;
-            memcpy(&qh, &y[i].qh, sizeof(qh));
-
-            for (int j = 0; j < QK5_0; j += 2) {
-                const uint8_t vh0 = ((qh & (1u << (j/2 + 0 ))) >> (j/2 + 0 )) << 4;
-                const uint8_t vh1 = ((qh & (1u << (j/2 + 16))) >> (j/2 + 12));
-
-                // cast to 16 bins
-                const uint8_t vi0 = ((y[i].qs[j/2] & 0x0F) | vh0) / 2;
-                const uint8_t vi1 = ((y[i].qs[j/2] >>   4) | vh1) / 2;
-
-                hist[vi0]++;
-                hist[vi1]++;
-            }
-        }
-    }
-
-    return (n/QK5_0*sizeof(block_q5_0));
-}
-
-size_t ggml_quantize_q5_1(const float * src, void * dst, int n, int k, int64_t * hist) {
-    assert(k % QK5_1 == 0);
-    const int nb = k / QK5_1;
-
-    for (int b = 0; b < n; b += k) {
-        block_q5_1 * restrict y = (block_q5_1 *)dst + b/QK5_1;
-
-        quantize_row_q5_1_reference(src + b, y, k);
-
-        for (int i = 0; i < nb; i++) {
-            uint32_t qh;
-            memcpy(&qh, &y[i].qh, sizeof(qh));
-
-            for (int j = 0; j < QK5_1; j += 2) {
-                const uint8_t vh0 = ((qh & (1u << (j/2 + 0 ))) >> (j/2 + 0 )) << 4;
-                const uint8_t vh1 = ((qh & (1u << (j/2 + 16))) >> (j/2 + 12));
-
-                // cast to 16 bins
-                const uint8_t vi0 = ((y[i].qs[j/2] & 0x0F) | vh0) / 2;
-                const uint8_t vi1 = ((y[i].qs[j/2] >>   4) | vh1) / 2;
-
-                hist[vi0]++;
-                hist[vi1]++;
-            }
-        }
-    }
-
-    return (n/QK5_1*sizeof(block_q5_1));
-}
-
-size_t ggml_quantize_q8_0(const float * src, void * dst, int n, int k, int64_t * hist) {
-    assert(k % QK8_0 == 0);
-    const int nb = k / QK8_0;
-
-    for (int b = 0; b < n; b += k) {
-        block_q8_0 * restrict y = (block_q8_0 *)dst + b/QK8_0;
-
-        quantize_row_q8_0_reference(src + b, y, k);
-
-        for (int i = 0; i < nb; i++) {
-            for (int j = 0; j < QK8_0; ++j) {
-                const int8_t vi = y[i].qs[j];
-
-                hist[vi/16 + 8]++;
-            }
-        }
-    }
-
-    return (n/QK8_0*sizeof(block_q8_0));
-}
-
 bool ggml_quantize_requires_imatrix(enum ggml_type type) {
     return
         type == GGML_TYPE_IQ2_XXS ||
@@ -19918,177 +20206,52 @@ bool ggml_quantize_requires_imatrix(enum ggml_type type) {
         type == GGML_TYPE_IQ1_S;
 }
 
-size_t ggml_quantize_chunk(enum ggml_type type, const float * src, void * dst, int start,
-        int nrows, int n_per_row, int64_t * hist, const float * imatrix) {
+size_t ggml_quantize_chunk(
+        enum ggml_type   type,
+           const float * src,
+                  void * dst,
+                   int   start,
+                   int   nrows,
+                   int   n_per_row,
+           const float * imatrix) {
+    const int n = nrows * n_per_row;
+
+    if (ggml_quantize_requires_imatrix(type)) {
+        GGML_ASSERT(imatrix != NULL);
+    }
+
+    GGML_ASSERT(start % type_traits[type].blck_size == 0);
+    GGML_ASSERT(start % n_per_row == 0);
+
     ggml_quantize_init(type); // this is noop if already initialized
+
+    const size_t start_row = start / n_per_row;
+    const size_t row_size  = ggml_row_size(type, n_per_row);
+
     size_t result = 0;
-    int n = nrows * n_per_row;
+
     switch (type) {
-        case GGML_TYPE_Q4_0:
-            {
-                GGML_ASSERT(start % QK4_0 == 0);
-                GGML_ASSERT(start % n_per_row == 0);
-                size_t start_row = start / n_per_row;
-                size_t row_size = ggml_row_size(type, n_per_row);
-                result = quantize_q4_0(src + start, (char *)dst + start_row * row_size, nrows, n_per_row, hist, imatrix);
-                GGML_ASSERT(result == row_size * nrows);
-            } break;
-        case GGML_TYPE_Q4_1:
-            {
-                GGML_ASSERT(start % QK4_1 == 0);
-                GGML_ASSERT(start % n_per_row == 0);
-                size_t start_row = start / n_per_row;
-                size_t row_size = ggml_row_size(type, n_per_row);
-                result = quantize_q4_1(src + start, (char *)dst + start_row * row_size, nrows, n_per_row, hist, imatrix);
-                GGML_ASSERT(result == row_size * nrows);
-            } break;
-        case GGML_TYPE_Q5_0:
-            {
-                GGML_ASSERT(start % QK5_0 == 0);
-                GGML_ASSERT(start % n_per_row == 0);
-                size_t start_row = start / n_per_row;
-                size_t row_size = ggml_row_size(type, n_per_row);
-                result = quantize_q5_0(src + start, (char *)dst + start_row * row_size, nrows, n_per_row, hist, imatrix);
-                GGML_ASSERT(result == row_size * nrows);
-            } break;
-        case GGML_TYPE_Q5_1:
-            {
-                GGML_ASSERT(start % QK5_1 == 0);
-                GGML_ASSERT(start % n_per_row == 0);
-                size_t start_row = start / n_per_row;
-                size_t row_size = ggml_row_size(type, n_per_row);
-                result = quantize_q5_1(src + start, (char *)dst + start_row * row_size, nrows, n_per_row, hist, imatrix);
-                GGML_ASSERT(result == row_size * nrows);
-            } break;
-        case GGML_TYPE_Q8_0:
-            {
-                GGML_ASSERT(start % QK8_0 == 0);
-                block_q8_0 * block = (block_q8_0*)dst + start / QK8_0;
-                result = ggml_quantize_q8_0(src + start, block, n, n, hist);
-            } break;
-        case GGML_TYPE_Q2_K:
-            {
-                GGML_ASSERT(start % QK_K == 0);
-                GGML_ASSERT(start % n_per_row == 0);
-                size_t start_row = start / n_per_row;
-                size_t row_size = ggml_row_size(type, n_per_row);
-                result = quantize_q2_K(src + start, (char *)dst + start_row * row_size, nrows, n_per_row, hist, imatrix);
-                GGML_ASSERT(result == row_size * nrows);
-            } break;
-        case GGML_TYPE_Q3_K:
-            {
-                GGML_ASSERT(start % QK_K == 0);
-                GGML_ASSERT(start % n_per_row == 0);
-                size_t start_row = start / n_per_row;
-                size_t row_size = ggml_row_size(type, n_per_row);
-                result = quantize_q3_K(src + start, (char *)dst + start_row * row_size, nrows, n_per_row, hist, imatrix);
-                GGML_ASSERT(result == row_size * nrows);
-            } break;
-        case GGML_TYPE_Q4_K:
-            {
-                GGML_ASSERT(start % QK_K == 0);
-                GGML_ASSERT(start % n_per_row == 0);
-                size_t start_row = start / n_per_row;
-                size_t row_size = ggml_row_size(type, n_per_row);
-                result = quantize_q4_K(src + start, (char *)dst + start_row * row_size, nrows, n_per_row, hist, imatrix);
-                GGML_ASSERT(result == row_size * nrows);
-            } break;
-        case GGML_TYPE_Q5_K:
-            {
-                GGML_ASSERT(start % QK_K == 0);
-                GGML_ASSERT(start % n_per_row == 0);
-                size_t start_row = start / n_per_row;
-                size_t row_size = ggml_row_size(type, n_per_row);
-                result = quantize_q5_K(src + start, (char *)dst + start_row * row_size, nrows, n_per_row, hist, imatrix);
-                GGML_ASSERT(result == row_size * nrows);
-            } break;
-        case GGML_TYPE_Q6_K:
-            {
-                GGML_ASSERT(start % QK_K == 0);
-                GGML_ASSERT(start % n_per_row == 0);
-                size_t start_row = start / n_per_row;
-                size_t row_size = ggml_row_size(type, n_per_row);
-                result = quantize_q6_K(src + start, (char *)dst + start_row * row_size, nrows, n_per_row, hist, imatrix);
-                GGML_ASSERT(result == row_size * nrows);
-            } break;
-        case GGML_TYPE_IQ2_XXS:
-            {
-                GGML_ASSERT(start % QK_K == 0);
-                GGML_ASSERT(start % n_per_row == 0);
-                GGML_ASSERT(imatrix);
-                size_t start_row = start / n_per_row;
-                size_t row_size = ggml_row_size(type, n_per_row);
-                result = quantize_iq2_xxs(src + start, (char *)dst + start_row * row_size, nrows, n_per_row, hist, imatrix);
-                GGML_ASSERT(result == row_size * nrows);
-            } break;
-        case GGML_TYPE_IQ2_XS:
-            {
-                GGML_ASSERT(start % QK_K == 0);
-                GGML_ASSERT(start % n_per_row == 0);
-                GGML_ASSERT(imatrix);
-                size_t start_row = start / n_per_row;
-                size_t row_size = ggml_row_size(type, n_per_row);
-                result = quantize_iq2_xs(src + start, (char *)dst + start_row * row_size, nrows, n_per_row, hist, imatrix);
-                GGML_ASSERT(result == row_size * nrows);
-            } break;
-        case GGML_TYPE_IQ3_XXS:
-            {
-                GGML_ASSERT(start % QK_K == 0);
-                GGML_ASSERT(start % n_per_row == 0);
-                size_t start_row = start / n_per_row;
-                size_t row_size = ggml_row_size(type, n_per_row);
-                result = quantize_iq3_xxs(src + start, (char *)dst + start_row * row_size, nrows, n_per_row, hist, imatrix);
-                GGML_ASSERT(result == row_size * nrows);
-            } break;
-        case GGML_TYPE_IQ3_S:
-            {
-                GGML_ASSERT(start % QK_K == 0);
-                GGML_ASSERT(start % n_per_row == 0);
-                size_t start_row = start / n_per_row;
-                size_t row_size = ggml_row_size(type, n_per_row);
-                result = quantize_iq3_s(src + start, (char *)dst + start_row * row_size, nrows, n_per_row, hist, imatrix);
-                GGML_ASSERT(result == row_size * nrows);
-            } break;
-        case GGML_TYPE_IQ2_S:
-            {
-                GGML_ASSERT(start % QK_K == 0);
-                GGML_ASSERT(start % n_per_row == 0);
-                size_t start_row = start / n_per_row;
-                size_t row_size = ggml_row_size(type, n_per_row);
-                result = quantize_iq2_s(src + start, (char *)dst + start_row * row_size, nrows, n_per_row, hist, imatrix);
-                GGML_ASSERT(result == row_size * nrows);
-            } break;
-        case GGML_TYPE_IQ1_S:
-            {
-                GGML_ASSERT(start % QK_K == 0);
-                GGML_ASSERT(start % n_per_row == 0);
-                size_t start_row = start / n_per_row;
-                size_t row_size = ggml_row_size(type, n_per_row);
-                result = quantize_iq1_s(src + start, (char *)dst + start_row * row_size, nrows, n_per_row, hist, imatrix);
-                GGML_ASSERT(result == row_size * nrows);
-            } break;
-        case GGML_TYPE_IQ4_NL:
+        case GGML_TYPE_Q4_0:    result = quantize_q4_0(src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
+        case GGML_TYPE_Q4_1:    result = quantize_q4_1(src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
+        case GGML_TYPE_Q5_0:    result = quantize_q5_0(src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
+        case GGML_TYPE_Q5_1:    result = quantize_q5_1(src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
+        case GGML_TYPE_Q8_0:    result = quantize_q8_0(src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
+        case GGML_TYPE_Q2_K:    result = quantize_q2_K(src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
+        case GGML_TYPE_Q3_K:    result = quantize_q3_K(src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
+        case GGML_TYPE_Q4_K:    result = quantize_q4_K(src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
+        case GGML_TYPE_Q5_K:    result = quantize_q5_K(src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
+        case GGML_TYPE_Q6_K:    result = quantize_q6_K(src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
+        case GGML_TYPE_IQ2_XXS: result = quantize_iq2_xxs(src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
+        case GGML_TYPE_IQ2_XS:  result = quantize_iq2_xs (src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
+        case GGML_TYPE_IQ3_XXS: result = quantize_iq3_xxs(src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
+        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_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:
-#endif
-            {
-                GGML_ASSERT(start % QK4_NL == 0);
-                GGML_ASSERT(start % n_per_row == 0);
-                size_t start_row = start / n_per_row;
-                size_t row_size = ggml_row_size(type, n_per_row);
-                result = quantize_iq4_nl(src + start, (char *)dst + start_row * row_size, nrows, n_per_row, hist, imatrix);
-                GGML_ASSERT(result == row_size * nrows);
-            } break;
-#if QK_K != 64
-        case GGML_TYPE_IQ4_XS:
-            {
-                GGML_ASSERT(start % QK_K == 0);
-                GGML_ASSERT(start % n_per_row == 0);
-                size_t start_row = start / n_per_row;
-                size_t row_size = ggml_row_size(type, n_per_row);
-                result = quantize_iq4_xs(src + start, (char *)dst + start_row * row_size, nrows, n_per_row, hist, imatrix);
-                GGML_ASSERT(result == row_size * nrows);
-            } break;
+        case GGML_TYPE_IQ4_XS:  result = quantize_iq4_nl (src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
+#else
+        case GGML_TYPE_IQ4_XS:  result = quantize_iq4_xs (src + start, (char *) dst + start_row * row_size, nrows, n_per_row, imatrix); break;
 #endif
         case GGML_TYPE_F16:
             {
@@ -20105,6 +20268,9 @@ size_t ggml_quantize_chunk(enum ggml_type type, const float * src, void * dst, i
         default:
             assert(false);
     }
+
+    GGML_ASSERT(result == nrows * row_size);
+
     return result;
 }