}
func = ggml_cuda_mul;
break;
- case GGML_OP_GELU:
- if (!any_on_device) {
- return false;
- }
- func = ggml_cuda_gelu;
- break;
- case GGML_OP_SILU:
- if (!any_on_device) {
- return false;
- }
- func = ggml_cuda_silu;
- break;
+ case GGML_OP_UNARY:
+ switch (ggml_get_unary_op(tensor)) {
+ case GGML_UNARY_OP_GELU:
+ if (!any_on_device) {
+ return false;
+ }
+ func = ggml_cuda_gelu;
+ break;
+ case GGML_UNARY_OP_SILU:
+ if (!any_on_device) {
+ return false;
+ }
+ func = ggml_cuda_silu;
+ break;
+ default:
+ return false;
+ } break;
case GGML_OP_NORM:
if (!any_on_device) {
return false;
[encoder dispatchThreadgroups:MTLSizeMake(n, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
} break;
- case GGML_OP_SILU:
- {
- if (encoder == nil) {
- encoder = [command_buffer computeCommandEncoder];
- }
-
- [encoder setComputePipelineState:ctx->pipeline_silu];
- [encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
- [encoder setBuffer:id_dst offset:offs_dst atIndex:1];
-
- const int64_t n = ggml_nelements(dst);
-
- [encoder dispatchThreadgroups:MTLSizeMake(n, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
- } break;
- case GGML_OP_RELU:
- {
- if (encoder == nil) {
- encoder = [command_buffer computeCommandEncoder];
- }
-
- [encoder setComputePipelineState:ctx->pipeline_relu];
- [encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
- [encoder setBuffer:id_dst offset:offs_dst atIndex:1];
-
- const int64_t n = ggml_nelements(dst);
-
- [encoder dispatchThreadgroups:MTLSizeMake(n, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
+ case GGML_OP_UNARY:
+ switch (ggml_get_unary_op(gf->nodes[i])) {
+ case GGML_UNARY_OP_SILU:
+ {
+ if (encoder == nil) {
+ encoder = [command_buffer computeCommandEncoder];
+ }
+
+ [encoder setComputePipelineState:ctx->pipeline_silu];
+ [encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
+ [encoder setBuffer:id_dst offset:offs_dst atIndex:1];
+
+ const int64_t n = ggml_nelements(dst);
+
+ [encoder dispatchThreadgroups:MTLSizeMake(n, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
+ } break;
+ case GGML_UNARY_OP_RELU:
+ {
+ if (encoder == nil) {
+ encoder = [command_buffer computeCommandEncoder];
+ }
+
+ [encoder setComputePipelineState:ctx->pipeline_relu];
+ [encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
+ [encoder setBuffer:id_dst offset:offs_dst atIndex:1];
+
+ const int64_t n = ggml_nelements(dst);
+
+ [encoder dispatchThreadgroups:MTLSizeMake(n, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
+ } break;
+ case GGML_UNARY_OP_GELU:
+ {
+ if (encoder == nil) {
+ encoder = [command_buffer computeCommandEncoder];
+ }
+
+ [encoder setComputePipelineState:ctx->pipeline_gelu];
+ [encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
+ [encoder setBuffer:id_dst offset:offs_dst atIndex:1];
+
+ const int64_t n = ggml_nelements(dst);
+
+ [encoder dispatchThreadgroups:MTLSizeMake(n, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
+ } break;
+ default:
+ {
+ fprintf(stderr, "%s: node %3d, op = %8s not implemented\n", __func__, i, ggml_op_name(dst->op));
+ GGML_ASSERT(false);
+ }
} break;
- case GGML_OP_GELU:
- {
- if (encoder == nil) {
- encoder = [command_buffer computeCommandEncoder];
- }
-
- [encoder setComputePipelineState:ctx->pipeline_gelu];
- [encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
- [encoder setBuffer:id_dst offset:offs_dst atIndex:1];
-
- const int64_t n = ggml_nelements(dst);
-
- [encoder dispatchThreadgroups:MTLSizeMake(n, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
- } break;
case GGML_OP_SOFT_MAX:
{
if (encoder == nil) {
[encoder dispatchThreadgroups:MTLSizeMake(ne01, ne02, ne03) threadsPerThreadgroup:MTLSizeMake(nth, 1, 1)];
} break;
default:
- fprintf(stderr, "%s: node %3d, op = %8s not implemented\n", __func__, i, ggml_op_name(dst->op));
- GGML_ASSERT(false);
+ {
+ fprintf(stderr, "%s: node %3d, op = %8s not implemented\n", __func__, i, ggml_op_name(dst->op));
+ GGML_ASSERT(false);
+ }
}
}
//inline static void ggml_vec_scale_f32(const int n, float * y, const float v) { for (int i = 0; i < n; ++i) y[i] *= v; }
inline static void ggml_vec_scale_f32(const int n, float * y, const float v) {
-#if defined(GGML_SIMD)
+#if defined(GGML_USE_ACCELERATE)
+ vDSP_vsmul(y, 1, &v, y, 1, n);
+#elif defined(GGML_SIMD)
const int np = (n & ~(GGML_F32_STEP - 1));
GGML_F32_VEC vx = GGML_F32_VEC_SET1(v);
#endif
}
-inline static void ggml_vec_sum_ggf(const int n, ggml_float * s, const float * x) {
+inline static void ggml_vec_sum_f32_ggf(const int n, ggml_float * s, const float * x) {
ggml_float sum = 0.0;
for (int i = 0; i < n; ++i) {
sum += (ggml_float)x[i];
*s = sum;
}
+inline static void ggml_vec_sum_f16_ggf(const int n, float * s, const ggml_fp16_t * x) {
+ float sum = 0.0f;
+ for (int i = 0; i < n; ++i) {
+ sum += GGML_FP16_TO_FP32(x[i]);
+ }
+ *s = sum;
+}
+
inline static void ggml_vec_max_f32(const int n, float * s, const float * x) {
#ifndef GGML_USE_ACCELERATE
float max = -INFINITY;
"ARGMAX",
"REPEAT",
"REPEAT_BACK",
- "ABS",
- "SGN",
- "NEG",
- "STEP",
- "TANH",
- "ELU",
- "RELU",
- "GELU",
- "GELU_QUICK",
- "SILU",
"SILU_BACK",
"NORM",
"RMS_NORM",
"WIN_PART",
"WIN_UNPART",
+ "UNARY",
+
"MAP_UNARY",
"MAP_BINARY",
"CROSS_ENTROPY_LOSS_BACK",
};
-static_assert(GGML_OP_COUNT == 68, "GGML_OP_COUNT != 68");
+static_assert(GGML_OP_COUNT == 59, "GGML_OP_COUNT != 59");
static const char * GGML_OP_SYMBOL[GGML_OP_COUNT] = {
"none",
"argmax(x)",
"repeat(x)",
"repeat_back(x)",
- "abs(x)",
- "sgn(x)",
- "-x",
- "step(x)",
- "tanh(x)",
- "elu(x)",
- "relu(x)",
- "gelu(x)",
- "gelu_quick(x)",
- "silu(x)",
"silu_back(x)",
"norm(x)",
"rms_norm(x)",
"win_part(x)",
"win_unpart(x)",
+ "unary(x)",
+
"f(x)",
"f(x,y)",
"cross_entropy_loss_back(x,y)",
};
-static_assert(GGML_OP_COUNT == 68, "GGML_OP_COUNT != 68");
+static_assert(GGML_OP_COUNT == 59, "GGML_OP_COUNT != 59");
static_assert(GGML_OP_POOL_COUNT == 2, "GGML_OP_POOL_COUNT != 2");
return GGML_OP_NAME[op];
}
+const char * ggml_op_symbol(enum ggml_op op) {
+ return GGML_OP_SYMBOL[op];
+}
+
size_t ggml_element_size(const struct ggml_tensor * tensor) {
return GGML_TYPE_SIZE[tensor->type];
}
return result;
}
+bool ggml_get_no_alloc(struct ggml_context * ctx) {
+ return ctx->no_alloc;
+}
+
void ggml_set_no_alloc(struct ggml_context * ctx, bool no_alloc) {
ctx->no_alloc = no_alloc;
}
// this is an error prone process, but it is necessary to support inplace
// operators when using scratch buffers
// TODO: implement a better way
-void ggml_scratch_save(struct ggml_context * ctx) {
+static void ggml_scratch_save(struct ggml_context * ctx) {
// this is needed to allow opt tensors to store their data
// TODO: again, need to find a better way
ctx->no_alloc_save = ctx->no_alloc;
ctx->scratch.data = NULL;
}
-void ggml_scratch_load(struct ggml_context * ctx) {
+static void ggml_scratch_load(struct ggml_context * ctx) {
ctx->no_alloc = ctx->no_alloc_save;
ctx->scratch = ctx->scratch_save;
////////////////////////////////////////////////////////////////////////////////
-struct ggml_tensor * ggml_new_tensor_impl(
+static struct ggml_tensor * ggml_new_tensor_impl(
struct ggml_context * ctx,
enum ggml_type type,
int n_dims,
return result;
}
+static void ggml_set_op_params(struct ggml_tensor * tensor, const void * params, size_t params_size) {
+ assert(params_size <= GGML_MAX_OP_PARAMS);
+ memcpy(tensor->op_params, params, params_size);
+}
+
+static int32_t ggml_get_op_params_i32(const struct ggml_tensor * tensor, uint32_t i) {
+ assert(i < GGML_MAX_OP_PARAMS / sizeof(int32_t));
+ return ((const int32_t *)(tensor->op_params))[i];
+}
+
+static void ggml_set_op_params_i32(struct ggml_tensor * tensor, uint32_t i, int32_t value) {
+ assert(i < GGML_MAX_OP_PARAMS / sizeof(int32_t));
+ ((int32_t *)(tensor->op_params))[i] = value;
+}
+
struct ggml_tensor * ggml_new_tensor(
struct ggml_context * ctx,
enum ggml_type type,
return (float *)(tensor->data);
}
+enum ggml_unary_op ggml_get_unary_op(const struct ggml_tensor * tensor) {
+ GGML_ASSERT(tensor->op == GGML_OP_UNARY);
+ return (enum ggml_unary_op) ggml_get_op_params_i32(tensor, 0);
+}
+
+static void ggml_set_unary_op(struct ggml_tensor * tensor, enum ggml_unary_op op) {
+ GGML_ASSERT(tensor->op = GGML_OP_UNARY);
+ ggml_set_op_params_i32(tensor, 0, (int32_t) op);
+}
+
const char * ggml_get_name(const struct ggml_tensor * tensor) {
return tensor->name;
}
return tensor;
}
-static void ggml_set_op_params(struct ggml_tensor * tensor, const void * params, size_t params_size) {
- assert(params_size <= GGML_MAX_OP_PARAMS);
- memcpy(tensor->op_params, params, params_size);
-}
-
struct ggml_tensor * ggml_view_tensor(
struct ggml_context * ctx,
const struct ggml_tensor * src) {
// ggml_dup
-struct ggml_tensor * ggml_dup_impl(
+static struct ggml_tensor * ggml_dup_impl(
struct ggml_context * ctx,
struct ggml_tensor * a,
bool inplace) {
// ggml_add
-struct ggml_tensor * ggml_add_impl(
+static struct ggml_tensor * ggml_add_impl(
struct ggml_context * ctx,
struct ggml_tensor * a,
struct ggml_tensor * b,
// ggml_add1
-struct ggml_tensor * ggml_add1_impl(
+static struct ggml_tensor * ggml_add1_impl(
struct ggml_context * ctx,
struct ggml_tensor * a,
struct ggml_tensor * b,
// ggml_acc
-struct ggml_tensor * ggml_acc_impl(
+static struct ggml_tensor * ggml_acc_impl(
struct ggml_context * ctx,
struct ggml_tensor * a,
struct ggml_tensor * b,
// ggml_sub
-struct ggml_tensor * ggml_sub_impl(
+static struct ggml_tensor * ggml_sub_impl(
struct ggml_context * ctx,
struct ggml_tensor * a,
struct ggml_tensor * b,
// ggml_mul
-struct ggml_tensor * ggml_mul_impl(
+static struct ggml_tensor * ggml_mul_impl(
struct ggml_context * ctx,
struct ggml_tensor * a,
struct ggml_tensor * b,
// ggml_div
-struct ggml_tensor * ggml_div_impl(
+static struct ggml_tensor * ggml_div_impl(
struct ggml_context * ctx,
struct ggml_tensor * a,
struct ggml_tensor * b,
// ggml_sqr
-struct ggml_tensor * ggml_sqr_impl(
+static struct ggml_tensor * ggml_sqr_impl(
struct ggml_context * ctx,
struct ggml_tensor * a,
bool inplace) {
// ggml_sqrt
-struct ggml_tensor * ggml_sqrt_impl(
+static struct ggml_tensor * ggml_sqrt_impl(
struct ggml_context * ctx,
struct ggml_tensor * a,
bool inplace) {
// ggml_log
-struct ggml_tensor * ggml_log_impl(
+static struct ggml_tensor * ggml_log_impl(
struct ggml_context * ctx,
struct ggml_tensor * a,
bool inplace) {
// ggml_abs
-struct ggml_tensor * ggml_abs_impl(
- struct ggml_context * ctx,
- struct ggml_tensor * a,
- bool inplace) {
- bool is_node = false;
-
- if (!inplace && (a->grad)) {
- is_node = true;
- }
-
- struct ggml_tensor * result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
-
- result->op = GGML_OP_ABS;
- result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
- result->src[0] = a;
-
- return result;
-}
-
struct ggml_tensor * ggml_abs(
struct ggml_context * ctx,
struct ggml_tensor * a) {
- return ggml_abs_impl(ctx, a, false);
+ return ggml_unary(ctx, a, GGML_UNARY_OP_ABS);
}
struct ggml_tensor * ggml_abs_inplace(
struct ggml_context * ctx,
struct ggml_tensor * a) {
- return ggml_abs_impl(ctx, a, true);
+ return ggml_unary_inplace(ctx, a, GGML_UNARY_OP_ABS);
}
-
// ggml_sgn
-struct ggml_tensor * ggml_sgn_impl(
- struct ggml_context * ctx,
- struct ggml_tensor * a,
- bool inplace) {
- bool is_node = false;
-
- if (!inplace && (a->grad)) {
- is_node = true;
- }
-
- struct ggml_tensor * result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
-
- result->op = GGML_OP_SGN;
- result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
- result->src[0] = a;
-
- return result;
-}
-
struct ggml_tensor * ggml_sgn(
struct ggml_context * ctx,
struct ggml_tensor * a) {
- return ggml_sgn_impl(ctx, a, false);
+ return ggml_unary(ctx, a, GGML_UNARY_OP_SGN);
}
struct ggml_tensor * ggml_sgn_inplace(
struct ggml_context * ctx,
struct ggml_tensor * a) {
- return ggml_sgn_impl(ctx, a, true);
+ return ggml_unary_inplace(ctx, a, GGML_UNARY_OP_SGN);
}
// ggml_neg
-struct ggml_tensor * ggml_neg_impl(
- struct ggml_context * ctx,
- struct ggml_tensor * a,
- bool inplace) {
- bool is_node = false;
-
- if (!inplace && (a->grad)) {
- is_node = true;
- }
-
- struct ggml_tensor * result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
-
- result->op = GGML_OP_NEG;
- result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
- result->src[0] = a;
-
- return result;
-}
-
struct ggml_tensor * ggml_neg(
struct ggml_context * ctx,
struct ggml_tensor * a) {
- return ggml_neg_impl(ctx, a, false);
+ return ggml_unary(ctx, a, GGML_UNARY_OP_NEG);
}
struct ggml_tensor * ggml_neg_inplace(
struct ggml_context * ctx,
struct ggml_tensor * a) {
- return ggml_neg_impl(ctx, a, true);
+ return ggml_unary_inplace(ctx, a, GGML_UNARY_OP_NEG);
}
// ggml_step
-struct ggml_tensor * ggml_step_impl(
- struct ggml_context * ctx,
- struct ggml_tensor * a,
- bool inplace) {
- bool is_node = false;
-
- if (!inplace && (a->grad)) {
- is_node = true;
- }
-
- struct ggml_tensor * result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
-
- result->op = GGML_OP_STEP;
- result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
- result->src[0] = a;
-
- return result;
-}
-
struct ggml_tensor * ggml_step(
struct ggml_context * ctx,
struct ggml_tensor * a) {
- return ggml_step_impl(ctx, a, false);
+ return ggml_unary(ctx, a, GGML_UNARY_OP_STEP);
}
struct ggml_tensor * ggml_step_inplace(
struct ggml_context * ctx,
struct ggml_tensor * a) {
- return ggml_step_impl(ctx, a, true);
+ return ggml_unary_inplace(ctx, a, GGML_UNARY_OP_STEP);
}
// ggml_tanh
-struct ggml_tensor * ggml_tanh_impl(
- struct ggml_context * ctx,
- struct ggml_tensor * a,
- bool inplace) {
- bool is_node = false;
-
- if (!inplace && (a->grad)) {
- is_node = true;
- }
-
- struct ggml_tensor * result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
-
- result->op = GGML_OP_TANH;
- result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
- result->src[0] = a;
-
- return result;
-}
-
struct ggml_tensor * ggml_tanh(
struct ggml_context * ctx,
struct ggml_tensor * a) {
- return ggml_tanh_impl(ctx, a, false);
+ return ggml_unary(ctx, a, GGML_UNARY_OP_TANH);
}
struct ggml_tensor * ggml_tanh_inplace(
struct ggml_context * ctx,
struct ggml_tensor * a) {
- return ggml_tanh_impl(ctx, a, true);
+ return ggml_unary_inplace(ctx, a, GGML_UNARY_OP_TANH);
}
// ggml_elu
-struct ggml_tensor * ggml_elu_impl(
- struct ggml_context * ctx,
- struct ggml_tensor * a,
- bool inplace) {
- bool is_node = false;
-
- if (!inplace && (a->grad)) {
- is_node = true;
- }
-
- struct ggml_tensor * result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
-
- result->op = GGML_OP_ELU;
- result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
- result->src[0] = a;
-
- return result;
-}
-
struct ggml_tensor * ggml_elu(
struct ggml_context * ctx,
struct ggml_tensor * a) {
- return ggml_elu_impl(ctx, a, false);
+ return ggml_unary(ctx, a, GGML_UNARY_OP_ELU);
}
struct ggml_tensor * ggml_elu_inplace(
struct ggml_context * ctx,
struct ggml_tensor * a) {
- return ggml_elu_impl(ctx, a, true);
+ return ggml_unary_inplace(ctx, a, GGML_UNARY_OP_ELU);
}
// ggml_relu
-struct ggml_tensor * ggml_relu_impl(
- struct ggml_context * ctx,
- struct ggml_tensor * a,
- bool inplace) {
- bool is_node = false;
-
- if (!inplace && (a->grad)) {
- is_node = true;
- }
-
- struct ggml_tensor * result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
-
- result->op = GGML_OP_RELU;
- result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
- result->src[0] = a;
-
- return result;
-}
-
struct ggml_tensor * ggml_relu(
struct ggml_context * ctx,
struct ggml_tensor * a) {
- return ggml_relu_impl(ctx, a, false);
+ return ggml_unary(ctx, a, GGML_UNARY_OP_RELU);
}
struct ggml_tensor * ggml_relu_inplace(
struct ggml_context * ctx,
struct ggml_tensor * a) {
- return ggml_relu_impl(ctx, a, true);
+ return ggml_unary_inplace(ctx, a, GGML_UNARY_OP_RELU);
}
// ggml_gelu
-struct ggml_tensor * ggml_gelu_impl(
- struct ggml_context * ctx,
- struct ggml_tensor * a,
- bool inplace) {
- bool is_node = false;
-
- if (!inplace && (a->grad)) {
- is_node = true;
- }
-
- struct ggml_tensor * result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
-
- result->op = GGML_OP_GELU;
- result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
- result->src[0] = a;
-
- return result;
-}
-
struct ggml_tensor * ggml_gelu(
struct ggml_context * ctx,
struct ggml_tensor * a) {
- return ggml_gelu_impl(ctx, a, false);
+ return ggml_unary(ctx, a, GGML_UNARY_OP_GELU);
}
struct ggml_tensor * ggml_gelu_inplace(
struct ggml_context * ctx,
struct ggml_tensor * a) {
- return ggml_gelu_impl(ctx, a, true);
+ return ggml_unary_inplace(ctx, a, GGML_UNARY_OP_GELU);
}
// ggml_gelu_quick
-struct ggml_tensor * ggml_gelu_quick_impl(
- struct ggml_context * ctx,
- struct ggml_tensor * a,
- bool inplace) {
- bool is_node = false;
-
- if (!inplace && (a->grad)) {
- is_node = true;
- }
-
- struct ggml_tensor * result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
-
- result->op = GGML_OP_GELU_QUICK;
- result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
- result->src[0] = a;
-
- return result;
-}
-
struct ggml_tensor * ggml_gelu_quick(
struct ggml_context * ctx,
struct ggml_tensor * a) {
- return ggml_gelu_quick_impl(ctx, a, false);
+ return ggml_unary(ctx, a, GGML_UNARY_OP_GELU_QUICK);
}
struct ggml_tensor * ggml_gelu_quick_inplace(
struct ggml_context * ctx,
struct ggml_tensor * a) {
- return ggml_gelu_quick_impl(ctx, a, true);
+ return ggml_unary_inplace(ctx, a, GGML_UNARY_OP_GELU_QUICK);
}
// ggml_silu
-struct ggml_tensor * ggml_silu_impl(
- struct ggml_context * ctx,
- struct ggml_tensor * a,
- bool inplace) {
- bool is_node = false;
-
- if (!inplace && (a->grad)) {
- is_node = true;
- }
-
- struct ggml_tensor * result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
-
- result->op = GGML_OP_SILU;
- result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
- result->src[0] = a;
-
- return result;
-}
-
struct ggml_tensor * ggml_silu(
struct ggml_context * ctx,
struct ggml_tensor * a) {
- return ggml_silu_impl(ctx, a, false);
+ return ggml_unary(ctx, a, GGML_UNARY_OP_SILU);
}
struct ggml_tensor * ggml_silu_inplace(
struct ggml_context * ctx,
struct ggml_tensor * a) {
- return ggml_silu_impl(ctx, a, true);
+ return ggml_unary_inplace(ctx, a, GGML_UNARY_OP_SILU);
}
// ggml_silu_back
// ggml_norm
-struct ggml_tensor * ggml_norm_impl(
+static struct ggml_tensor * ggml_norm_impl(
struct ggml_context * ctx,
struct ggml_tensor * a,
bool inplace) {
return ggml_norm_impl(ctx, a, true);
}
-struct ggml_tensor * ggml_rms_norm_impl(
+static struct ggml_tensor * ggml_rms_norm_impl(
struct ggml_context * ctx,
struct ggml_tensor * a,
bool inplace) {
// ggml_scale
-struct ggml_tensor * ggml_scale_impl(
+static struct ggml_tensor * ggml_scale_impl(
struct ggml_context * ctx,
struct ggml_tensor * a,
struct ggml_tensor * b,
// ggml_set
-struct ggml_tensor * ggml_set_impl(
+static struct ggml_tensor * ggml_set_impl(
struct ggml_context * ctx,
struct ggml_tensor * a,
struct ggml_tensor * b,
// ggml_cpy
-struct ggml_tensor * ggml_cpy_impl(
+static struct ggml_tensor * ggml_cpy_impl(
struct ggml_context * ctx,
struct ggml_tensor * a,
struct ggml_tensor * b,
// ggml_cont
-struct ggml_tensor * ggml_cont_impl(
+static struct ggml_tensor * ggml_cont_impl(
struct ggml_context * ctx,
struct ggml_tensor * a,
bool inplace) {
// ggml_diag_mask_inf
-struct ggml_tensor * ggml_diag_mask_inf_impl(
+static struct ggml_tensor * ggml_diag_mask_inf_impl(
struct ggml_context * ctx,
struct ggml_tensor * a,
int n_past,
// ggml_diag_mask_zero
-struct ggml_tensor * ggml_diag_mask_zero_impl(
+static struct ggml_tensor * ggml_diag_mask_zero_impl(
struct ggml_context * ctx,
struct ggml_tensor * a,
int n_past,
// ggml_soft_max
-struct ggml_tensor * ggml_soft_max_impl(
+static struct ggml_tensor * ggml_soft_max_impl(
struct ggml_context * ctx,
struct ggml_tensor * a,
bool inplace) {
// ggml_soft_max_back
-struct ggml_tensor * ggml_soft_max_back_impl(
+static struct ggml_tensor * ggml_soft_max_back_impl(
struct ggml_context * ctx,
struct ggml_tensor * a,
struct ggml_tensor * b,
// ggml_rope
-struct ggml_tensor * ggml_rope_impl(
+static struct ggml_tensor * ggml_rope_impl(
struct ggml_context * ctx,
struct ggml_tensor * a,
int n_past,
return result;
}
+// gmml_unary
+
+static struct ggml_tensor * ggml_unary_impl(
+ struct ggml_context * ctx,
+ struct ggml_tensor * a,
+ enum ggml_unary_op op,
+ bool inplace) {
+ bool is_node = false;
+
+ if (!inplace && (a->grad)) {
+ is_node = true;
+ }
+
+ struct ggml_tensor * result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
+
+ ggml_set_unary_op(result, op);
+
+ result->op = GGML_OP_UNARY;
+ result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
+ result->src[0] = a;
+
+ return result;
+}
+
+struct ggml_tensor * ggml_unary(
+ struct ggml_context * ctx,
+ struct ggml_tensor * a,
+ enum ggml_unary_op op) {
+ return ggml_unary_impl(ctx, a, op, false);
+}
+
+struct ggml_tensor * ggml_unary_inplace(
+ struct ggml_context * ctx,
+ struct ggml_tensor * a,
+ enum ggml_unary_op op) {
+ return ggml_unary_impl(ctx, a, op, true);
+}
+
// ggml_map_unary
-struct ggml_tensor * ggml_map_unary_impl_f32(
+static struct ggml_tensor * ggml_map_unary_impl_f32(
struct ggml_context * ctx,
struct ggml_tensor * a,
const ggml_unary_op_f32_t fun,
// ggml_map_binary
-struct ggml_tensor * ggml_map_binary_impl_f32(
+static struct ggml_tensor * ggml_map_binary_impl_f32(
struct ggml_context * ctx,
struct ggml_tensor * a,
struct ggml_tensor * b,
// ggml_map_custom1
-struct ggml_tensor * ggml_map_custom1_impl_f32(
+static struct ggml_tensor * ggml_map_custom1_impl_f32(
struct ggml_context * ctx,
struct ggml_tensor * a,
const ggml_custom1_op_f32_t fun,
// ggml_map_custom2
-struct ggml_tensor * ggml_map_custom2_impl_f32(
+static struct ggml_tensor * ggml_map_custom2_impl_f32(
struct ggml_context * ctx,
struct ggml_tensor * a,
struct ggml_tensor * b,
// ggml_map_custom3
-struct ggml_tensor * ggml_map_custom3_impl_f32(
+static struct ggml_tensor * ggml_map_custom3_impl_f32(
struct ggml_context * ctx,
struct ggml_tensor * a,
struct ggml_tensor * b,
for (int64_t i03 = 0; i03 < ne03; i03++) {
for (int64_t i02 = 0; i02 < ne02; i02++) {
for (int64_t i01 = 0; i01 < ne01; i01++) {
- ggml_vec_sum_ggf(ne00,
+ ggml_vec_sum_f32_ggf(ne00,
&row_sum,
(float *) ((char *) src0->data + i01*nb01 + i02*nb02 + i03*nb03));
sum += row_sum;
((float *) dst->data)[0] = sum;
}
+static void ggml_compute_forward_sum_f16(
+ const struct ggml_compute_params * params,
+ const struct ggml_tensor * src0,
+ struct ggml_tensor * dst) {
+ assert(params->ith == 0);
+ assert(ggml_is_scalar(dst));
+
+ if (params->type == GGML_TASK_INIT || params->type == GGML_TASK_FINALIZE) {
+ return;
+ }
+
+ assert(src0->nb[0] == sizeof(ggml_fp16_t));
+
+ GGML_TENSOR_LOCALS(int64_t, ne0, src0, ne);
+ GGML_TENSOR_LOCALS(size_t, nb0, src0, nb);
+
+ float sum = 0;
+ float row_sum = 0;
+
+ for (int64_t i03 = 0; i03 < ne03; i03++) {
+ for (int64_t i02 = 0; i02 < ne02; i02++) {
+ for (int64_t i01 = 0; i01 < ne01; i01++) {
+ ggml_vec_sum_f16_ggf(ne00,
+ &row_sum,
+ (ggml_fp16_t *) ((char *) src0->data + i01 * nb01 + i02 * nb02 + i03 * nb03));
+ sum += row_sum;
+ }
+ }
+ }
+ ((ggml_fp16_t *) dst->data)[0] = GGML_FP32_TO_FP16(sum);
+}
+
static void ggml_compute_forward_sum(
const struct ggml_compute_params * params,
const struct ggml_tensor * src0,
{
ggml_compute_forward_sum_f32(params, src0, dst);
} break;
+ case GGML_TYPE_F16:
+ {
+ ggml_compute_forward_sum_f16(params, src0, dst);
+ } break;
default:
{
GGML_ASSERT(false);
}
}
-
// ggml_compute_forward_silu_back
static void ggml_compute_forward_silu_back_f32(
}
}
+//gmml_compute_forward_unary
+
+static void ggml_compute_forward_unary(
+ const struct ggml_compute_params * params,
+ const struct ggml_tensor * src0,
+ struct ggml_tensor * dst) {
+ const enum ggml_unary_op op = ggml_get_unary_op(dst);
+
+ switch (op) {
+ case GGML_UNARY_OP_ABS:
+ {
+ ggml_compute_forward_abs(params, src0, dst);
+ } break;
+ case GGML_UNARY_OP_SGN:
+ {
+ ggml_compute_forward_sgn(params, src0, dst);
+ } break;
+ case GGML_UNARY_OP_NEG:
+ {
+ ggml_compute_forward_neg(params, src0, dst);
+ } break;
+ case GGML_UNARY_OP_STEP:
+ {
+ ggml_compute_forward_step(params, src0, dst);
+ } break;
+ case GGML_UNARY_OP_TANH:
+ {
+ ggml_compute_forward_tanh(params, src0, dst);
+ } break;
+ case GGML_UNARY_OP_ELU:
+ {
+ ggml_compute_forward_elu(params, src0, dst);
+ } break;
+ case GGML_UNARY_OP_RELU:
+ {
+ ggml_compute_forward_relu(params, src0, dst);
+ } break;
+ case GGML_UNARY_OP_GELU:
+ {
+ ggml_compute_forward_gelu(params, src0, dst);
+ } break;
+ case GGML_UNARY_OP_GELU_QUICK:
+ {
+ ggml_compute_forward_gelu_quick(params, src0, dst);
+ } break;
+ case GGML_UNARY_OP_SILU:
+ {
+ ggml_compute_forward_silu(params, src0, dst);
+ } break;
+ default:
+ {
+ GGML_ASSERT(false);
+ } break;
+ }
+}
+
// ggml_compute_forward_map_unary
static void ggml_compute_forward_map_unary_f32(
{
ggml_compute_forward_repeat_back(params, tensor->src[0], tensor);
} break;
- case GGML_OP_ABS:
- {
- ggml_compute_forward_abs(params, tensor->src[0], tensor);
- } break;
- case GGML_OP_SGN:
- {
- ggml_compute_forward_sgn(params, tensor->src[0], tensor);
- } break;
- case GGML_OP_NEG:
- {
- ggml_compute_forward_neg(params, tensor->src[0], tensor);
- } break;
- case GGML_OP_STEP:
- {
- ggml_compute_forward_step(params, tensor->src[0], tensor);
- } break;
- case GGML_OP_TANH:
- {
- ggml_compute_forward_tanh(params, tensor->src[0], tensor);
- } break;
- case GGML_OP_ELU:
- {
- ggml_compute_forward_elu(params, tensor->src[0], tensor);
- } break;
- case GGML_OP_RELU:
- {
- ggml_compute_forward_relu(params, tensor->src[0], tensor);
- } break;
- case GGML_OP_GELU:
- {
- ggml_compute_forward_gelu(params, tensor->src[0], tensor);
- } break;
- case GGML_OP_GELU_QUICK:
- {
- ggml_compute_forward_gelu_quick(params, tensor->src[0], tensor);
- } break;
- case GGML_OP_SILU:
- {
- ggml_compute_forward_silu(params, tensor->src[0], tensor);
- } break;
case GGML_OP_SILU_BACK:
{
ggml_compute_forward_silu_back(params, tensor->src[0], tensor->src[1], tensor);
{
ggml_compute_forward_win_unpart(params, tensor->src[0], tensor);
} break;
+ case GGML_OP_UNARY:
+ {
+ ggml_compute_forward_unary(params, tensor->src[0], tensor);
+ } break;
case GGML_OP_MAP_UNARY:
{
ggml_unary_op_f32_t fun;
inplace);
}
} break;
- case GGML_OP_ABS:
- {
- if (src0->grad) {
- src0->grad =
- ggml_add_impl(ctx,
- src0->grad,
- ggml_mul(ctx,
- ggml_sgn(ctx, src0),
- tensor->grad),
- inplace);
- }
- } break;
- case GGML_OP_SGN:
- {
- if (src0->grad) {
- // noop
- }
- } break;
- case GGML_OP_NEG:
- {
- if (src0->grad) {
- src0->grad = ggml_sub_impl(ctx, src0->grad, tensor->grad, inplace);
- }
- } break;
- case GGML_OP_STEP:
- {
- if (src0->grad) {
- // noop
- }
- } break;
- case GGML_OP_TANH:
- {
- GGML_ASSERT(false); // TODO: not implemented
- } break;
- case GGML_OP_ELU:
- {
- GGML_ASSERT(false); // TODO: not implemented
- } break;
- case GGML_OP_RELU:
- {
- if (src0->grad) {
- src0->grad = ggml_sub_impl(ctx,
- src0->grad,
- ggml_mul(ctx,
- ggml_step(ctx, src0),
- tensor->grad),
- inplace);
- }
- } break;
- case GGML_OP_GELU:
- {
- GGML_ASSERT(false); // TODO: not implemented
- } break;
- case GGML_OP_GELU_QUICK:
- {
- GGML_ASSERT(false); // TODO: not implemented
- } break;
- case GGML_OP_SILU:
- {
- // necessary for llama
- if (src0->grad) {
- src0->grad = ggml_add_impl(ctx,
- src0->grad,
- ggml_silu_back(ctx, src0, tensor->grad),
- inplace);
- }
- } break;
case GGML_OP_SILU_BACK:
{
GGML_ASSERT(false); // TODO: not implemented
ggml_diag_mask_zero_impl(ctx, tensor->grad, n_past, false),
inplace);
}
- if (src1->grad) {
- // noop
- }
} break;
case GGML_OP_DIAG_MASK_ZERO:
{
ggml_diag_mask_zero_impl(ctx, tensor->grad, n_past, false),
inplace);
}
- if (src1->grad) {
- // noop
- }
} break;
case GGML_OP_SOFT_MAX:
{
n_ctx),
inplace);
}
- if (src1->grad) {
- // noop
- }
} break;
case GGML_OP_ROPE_BACK:
{
n_ctx),
inplace);
}
- if (src1->grad) {
- // noop
- }
} break;
case GGML_OP_ALIBI:
{
} break;
case GGML_OP_WIN_PART:
case GGML_OP_WIN_UNPART:
+ case GGML_OP_UNARY:
+ {
+ switch (ggml_get_unary_op(tensor)) {
+ case GGML_UNARY_OP_ABS:
+ {
+ if (src0->grad) {
+ src0->grad =
+ ggml_add_impl(ctx,
+ src0->grad,
+ ggml_mul(ctx,
+ ggml_sgn(ctx, src0),
+ tensor->grad),
+ inplace);
+ }
+ } break;
+ case GGML_UNARY_OP_SGN:
+ {
+ if (src0->grad) {
+ // noop
+ }
+ } break;
+ case GGML_UNARY_OP_NEG:
+ {
+ if (src0->grad) {
+ src0->grad = ggml_sub_impl(ctx, src0->grad, tensor->grad, inplace);
+ }
+ } break;
+ case GGML_UNARY_OP_STEP:
+ {
+ if (src0->grad) {
+ // noop
+ }
+ } break;
+ case GGML_UNARY_OP_TANH:
+ {
+ GGML_ASSERT(false); // TODO: not implemented
+ } break;
+ case GGML_UNARY_OP_ELU:
+ {
+ GGML_ASSERT(false); // TODO: not implemented
+ } break;
+ case GGML_UNARY_OP_RELU:
+ {
+ if (src0->grad) {
+ src0->grad = ggml_add_impl(ctx,
+ src0->grad,
+ ggml_mul(ctx,
+ ggml_step(ctx, src0),
+ tensor->grad),
+ inplace);
+ }
+ } break;
+ case GGML_UNARY_OP_GELU:
+ {
+ GGML_ASSERT(false); // TODO: not implemented
+ } break;
+ case GGML_UNARY_OP_GELU_QUICK:
+ {
+ GGML_ASSERT(false); // TODO: not implemented
+ } break;
+ case GGML_UNARY_OP_SILU:
+ {
+ // necessary for llama
+ if (src0->grad) {
+ src0->grad = ggml_add_impl(ctx,
+ src0->grad,
+ ggml_silu_back(ctx, src0, tensor->grad),
+ inplace);
+ }
+ } break;
+ default:
+ GGML_ASSERT(false);
+ }
+ } break;
case GGML_OP_MAP_UNARY:
case GGML_OP_MAP_BINARY:
case GGML_OP_MAP_CUSTOM1:
// Android's libc implementation "bionic" does not support setting affinity
#if defined(__linux__) && !defined(__BIONIC__)
-void set_numa_thread_affinity(int thread_n, int n_threads) {
+static void set_numa_thread_affinity(int thread_n, int n_threads) {
if (!ggml_is_numa()) {
return;
}
CPU_FREE(cpus);
}
-void clear_numa_thread_affinity(void) {
+static void clear_numa_thread_affinity(void) {
if (!ggml_is_numa()) {
return;
}
#else
// TODO: Windows etc.
// (the linux implementation may also work on BSD, someone should test)
-void set_numa_thread_affinity(int thread_n, int n_threads) { UNUSED(thread_n); UNUSED(n_threads); }
-void clear_numa_thread_affinity(void) {}
+static void set_numa_thread_affinity(int thread_n, int n_threads) { UNUSED(thread_n); UNUSED(n_threads); }
+static void clear_numa_thread_affinity(void) {}
#endif
struct ggml_compute_state_shared {
case GGML_OP_ARGMAX:
case GGML_OP_REPEAT:
case GGML_OP_REPEAT_BACK:
- case GGML_OP_ABS:
- case GGML_OP_SGN:
- case GGML_OP_NEG:
- case GGML_OP_STEP:
- case GGML_OP_TANH:
- case GGML_OP_ELU:
- case GGML_OP_RELU:
- {
+ {
n_tasks = 1;
} break;
- case GGML_OP_MUL:
- case GGML_OP_GELU:
- case GGML_OP_GELU_QUICK:
- case GGML_OP_SILU:
+
+ case GGML_OP_UNARY:
+ {
+ switch (ggml_get_unary_op(node)) {
+ case GGML_UNARY_OP_ABS:
+ case GGML_UNARY_OP_SGN:
+ case GGML_UNARY_OP_NEG:
+ case GGML_UNARY_OP_STEP:
+ case GGML_UNARY_OP_TANH:
+ case GGML_UNARY_OP_ELU:
+ case GGML_UNARY_OP_RELU:
+ {
+ n_tasks = 1;
+ } break;
+
+ case GGML_UNARY_OP_GELU:
+ case GGML_UNARY_OP_GELU_QUICK:
+ case GGML_UNARY_OP_SILU:
+ {
+ n_tasks = n_threads;
+ } break;
+ }
+ } break;
case GGML_OP_SILU_BACK:
+ case GGML_OP_MUL:
case GGML_OP_NORM:
case GGML_OP_RMS_NORM:
case GGML_OP_RMS_NORM_BACK:
fwrite(&nb, sizeof(uint64_t), 1, fout);
}
- fwrite(tensor->name, sizeof(char), GGML_MAX_NAME, fout);
+ fwrite(tensor->name, sizeof(char), GGML_MAX_NAME, fout);
+ fwrite(tensor->op_params, sizeof(char), GGML_MAX_OP_PARAMS, fout);
// dump the data
// TODO: pad this to 32 byte boundary
fwrite(&nb, sizeof(uint64_t), 1, fout);
}
- fwrite(tensor->name, sizeof(char), GGML_MAX_NAME, fout);
+ fwrite(tensor->name, sizeof(char), GGML_MAX_NAME, fout);
+ fwrite(tensor->op_params, sizeof(char), GGML_MAX_OP_PARAMS, fout);
// output the op arguments
{
tensor->op = (enum ggml_op) op;
- memcpy(tensor->name, ptr, GGML_MAX_NAME); ptr += GGML_MAX_NAME;
+ memcpy(tensor->name, ptr, GGML_MAX_NAME); ptr += GGML_MAX_NAME;
+ memcpy(tensor->op_params, ptr, GGML_MAX_OP_PARAMS); ptr += GGML_MAX_OP_PARAMS;
tensor->data = (void *) ptr;
nb[j] = nb_cur;
}
- const char * ptr_name = ptr; ptr += GGML_MAX_NAME;
+ const char * ptr_name = ptr; ptr += GGML_MAX_NAME;
+ const char * ptr_op_params = ptr; ptr += GGML_MAX_OP_PARAMS;
const int32_t * ptr_arg_idx = (const int32_t *) ptr; ptr += GGML_MAX_SRC*sizeof(int32_t);
{
tensor = ggml_view_4d(*ctx_eval, args[0], ne[0], ne[1], ne[2], ne[3], 0, 0, 0, 0);
- uint64_t offs;
- memcpy(&offs, tensor->op_params, sizeof(offs));
+ size_t offs;
+ memcpy(&offs, ptr_op_params, sizeof(offs));
tensor->data = ((char *) tensor->data) + offs;
} break;
} break;
}
- memcpy(tensor->name, ptr_name, GGML_MAX_NAME);
+ memcpy(tensor->name, ptr_name, GGML_MAX_NAME);
+ memcpy(tensor->op_params, ptr_op_params, GGML_MAX_OP_PARAMS);
for (int j = 0; j < GGML_MAX_DIMS; ++j) {
tensor->nb[j] = nb[j];
GGML_PRINT(" - %3d: [ %5" PRId64 ", %5" PRId64 ", %5" PRId64 "] %16s %s (%3d) cpu = %7.3f / %7.3f ms, wall = %7.3f / %7.3f ms\n",
i,
node->ne[0], node->ne[1], node->ne[2],
- GGML_OP_NAME[node->op], node->is_param ? "x" : node->grad ? "g" : " ", node->perf_runs,
+ ggml_op_name(node->op), node->is_param ? "x" : node->grad ? "g" : " ", node->perf_runs,
(double) node->perf_cycles / (double) ggml_cycles_per_ms(),
(double) node->perf_cycles / (double) ggml_cycles_per_ms() / (double) node->perf_runs,
(double) node->perf_time_us / 1000.0,
GGML_PRINT(" - %3d: [ %5" PRId64 ", %5" PRId64 "] %8s\n",
i,
node->ne[0], node->ne[1],
- GGML_OP_NAME[node->op]);
+ ggml_op_name(node->op));
}
for (int i = 0; i < GGML_OP_COUNT; i++) {
continue;
}
- GGML_PRINT("perf_total_per_op_us[%16s] = %7.3f ms\n", GGML_OP_NAME[i], (double) perf_total_per_op_us[i] / 1000.0);
+ GGML_PRINT("perf_total_per_op_us[%16s] = %7.3f ms\n", ggml_op_name(i), (double) perf_total_per_op_us[i] / 1000.0);
}
GGML_PRINT("========================================\n");
}
if (node->n_dims == 2) {
- fprintf(fp, "%d [%" PRId64 ", %" PRId64 "] | <x>%s", i, node->ne[0], node->ne[1], GGML_OP_SYMBOL[node->op]);
+ fprintf(fp, "%d [%" PRId64 ", %" PRId64 "] | <x>%s", i, node->ne[0], node->ne[1], ggml_op_symbol(node->op));
} else {
- fprintf(fp, "%d [%" PRId64 ", %" PRId64 ", %" PRId64 "] | <x>%s", i, node->ne[0], node->ne[1], node->ne[2], GGML_OP_SYMBOL[node->op]);
+ fprintf(fp, "%d [%" PRId64 ", %" PRId64 ", %" PRId64 "] | <x>%s", i, node->ne[0], node->ne[1], node->ne[2], ggml_op_symbol(node->op));
}
if (node->grad) {
- fprintf(fp, " | <g>%s\"; ]\n", GGML_OP_SYMBOL[node->grad->op]);
+ fprintf(fp, " | <g>%s\"; ]\n", ggml_op_symbol(node->grad->op));
} else {
fprintf(fp, "\"; ]\n");
}
GGML_OP_ARGMAX,
GGML_OP_REPEAT,
GGML_OP_REPEAT_BACK,
- GGML_OP_ABS,
- GGML_OP_SGN,
- GGML_OP_NEG,
- GGML_OP_STEP,
- GGML_OP_TANH,
- GGML_OP_ELU,
- GGML_OP_RELU,
- GGML_OP_GELU,
- GGML_OP_GELU_QUICK,
- GGML_OP_SILU,
GGML_OP_SILU_BACK,
GGML_OP_NORM, // normalize
GGML_OP_RMS_NORM,
GGML_OP_WIN_PART,
GGML_OP_WIN_UNPART,
+ GGML_OP_UNARY,
+
GGML_OP_MAP_UNARY,
GGML_OP_MAP_BINARY,
GGML_OP_COUNT,
};
+ enum ggml_unary_op {
+ GGML_UNARY_OP_ABS,
+ GGML_UNARY_OP_SGN,
+ GGML_UNARY_OP_NEG,
+ GGML_UNARY_OP_STEP,
+ GGML_UNARY_OP_TANH,
+ GGML_UNARY_OP_ELU,
+ GGML_UNARY_OP_RELU,
+ GGML_UNARY_OP_GELU,
+ GGML_UNARY_OP_GELU_QUICK,
+ GGML_UNARY_OP_SILU,
+ };
// ggml object
struct ggml_object {
GGML_API const char * ggml_type_name(enum ggml_type type);
GGML_API const char * ggml_op_name (enum ggml_op op);
+ GGML_API const char * ggml_op_symbol(enum ggml_op op);
GGML_API size_t ggml_element_size(const struct ggml_tensor * tensor);
GGML_API size_t ggml_used_mem(const struct ggml_context * ctx);
GGML_API size_t ggml_set_scratch (struct ggml_context * ctx, struct ggml_scratch scratch);
+ GGML_API bool ggml_get_no_alloc(struct ggml_context * ctx);
GGML_API void ggml_set_no_alloc(struct ggml_context * ctx, bool no_alloc);
GGML_API void * ggml_get_mem_buffer (const struct ggml_context * ctx);
GGML_API void * ggml_get_data (const struct ggml_tensor * tensor);
GGML_API float * ggml_get_data_f32(const struct ggml_tensor * tensor);
- GGML_API const char * ggml_get_name(const struct ggml_tensor * tensor);
- GGML_API struct ggml_tensor * ggml_set_name(struct ggml_tensor * tensor, const char * name);
- GGML_API struct ggml_tensor * ggml_format_name(struct ggml_tensor * tensor, const char * fmt, ...);
+ GGML_API enum ggml_unary_op ggml_get_unary_op(const struct ggml_tensor * tensor);
+
+ GGML_API const char * ggml_get_name (const struct ggml_tensor * tensor);
+ GGML_API struct ggml_tensor * ggml_set_name ( struct ggml_tensor * tensor, const char * name);
+ GGML_API struct ggml_tensor * ggml_format_name( struct ggml_tensor * tensor, const char * fmt, ...);
//
// operations on tensors with backpropagation
struct ggml_context * ctx,
struct ggml_tensor * a);
+ // in-place, returns view(a)
+ GGML_API struct ggml_tensor * ggml_dup_inplace(
+ struct ggml_context * ctx,
+ struct ggml_tensor * a);
+
GGML_API struct ggml_tensor * ggml_add(
struct ggml_context * ctx,
struct ggml_tensor * a,
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);
+
// return view(a), b specifies the new shape
// TODO: when we start computing gradient, make a copy instead of view
GGML_API struct ggml_tensor * ggml_reshape(
typedef void (*ggml_custom2_op_f32_t)(struct ggml_tensor *, const struct ggml_tensor *, const struct ggml_tensor *);
typedef void (*ggml_custom3_op_f32_t)(struct ggml_tensor *, const struct ggml_tensor *, const struct ggml_tensor *, const struct ggml_tensor *);
+ GGML_API struct ggml_tensor * ggml_unary(
+ struct ggml_context * ctx,
+ struct ggml_tensor * a,
+ enum ggml_unary_op op);
+
+ GGML_API struct ggml_tensor * ggml_unary_inplace(
+ struct ggml_context * ctx,
+ struct ggml_tensor * a,
+ enum ggml_unary_op op);
+
GGML_API struct ggml_tensor * ggml_map_unary_f32(
struct ggml_context * ctx,
struct ggml_tensor * a,
}
}
-struct ggml_tensor * get_random_tensor(
+struct ggml_tensor * get_random_tensor_f32(
struct ggml_context * ctx0,
int ndims,
int64_t ne[],
return result;
}
-struct ggml_tensor * get_random_tensor_int(
+struct ggml_tensor * get_random_tensor_f16(
+ struct ggml_context * ctx0,
+ int ndims,
+ int64_t ne[],
+ float fmin,
+ float fmax) {
+ struct ggml_tensor * result = ggml_new_tensor(ctx0, GGML_TYPE_F16, ndims, ne);
+
+ switch (ndims) {
+ case 1:
+ for (int i0 = 0; i0 < ne[0]; i0++) {
+ ((ggml_fp16_t *)result->data)[i0] = ggml_fp32_to_fp16(frand()*(fmax - fmin) + fmin);
+ }
+ break;
+ case 2:
+ for (int i1 = 0; i1 < ne[1]; i1++) {
+ for (int i0 = 0; i0 < ne[0]; i0++) {
+ ((ggml_fp16_t *)result->data)[i1*ne[0] + i0] = ggml_fp32_to_fp16(frand()*(fmax - fmin) + fmin);
+ }
+ }
+ break;
+ case 3:
+ for (int i2 = 0; i2 < ne[2]; i2++) {
+ for (int i1 = 0; i1 < ne[1]; i1++) {
+ for (int i0 = 0; i0 < ne[0]; i0++) {
+ ((ggml_fp16_t *)result->data)[i2*ne[1]*ne[0] + i1*ne[0] + i0] = ggml_fp32_to_fp16(frand()*(fmax - fmin) + fmin);
+ }
+ }
+ }
+ break;
+ case 4:
+ for (int i3 = 0; i3 < ne[3]; i3++) {
+ for (int i2 = 0; i2 < ne[2]; i2++) {
+ for (int i1 = 0; i1 < ne[1]; i1++) {
+ for (int i0 = 0; i0 < ne[0]; i0++) {
+ ((ggml_fp16_t *)result->data)[i3*ne[2]*ne[1]*ne[0] + i2*ne[1]*ne[0] + i1*ne[0] + i0] = ggml_fp32_to_fp16(frand()*(fmax - fmin) + fmin);
+ }
+ }
+ }
+ }
+ break;
+ default:
+ assert(false);
+ };
+
+ return result;
+}
+
+struct ggml_tensor * get_random_tensor_i32(
struct ggml_context * ctx0,
int ndims,
int64_t ne[],
return result;
}
-float get_element(const struct ggml_tensor * t, int idx) {
- if (t->type == GGML_TYPE_F32) {
- return ((float *)t->data)[idx];
- }
-
- if (t->type == GGML_TYPE_I32) {
- return ((int32_t *)t->data)[idx];
- }
-
- assert(false);
- return INFINITY;
-}
-
-void set_element(struct ggml_tensor * t, int idx, float value) {
- ((float *)t->data)[idx] = value;
-}
-
void print_elements(const char* label, const struct ggml_tensor * t) {
if (!t) {
printf("%s: %s = null\n", __func__, label);
printf("%s: %s = [", __func__, label);
for (int k = 0; k < nelements; ++k) {
if (k > 0) { printf(", "); }
- printf("%.5f", get_element(t, k));
+ printf("%.5f", ggml_get_f32_1d(t, k));
}
printf("] shape: [");
for (int k = 0; k < t->n_dims; ++k) {
const int nelements = ggml_nelements(x[i]);
for (int k = 0; k < nelements; ++k) {
// compute gradient using finite differences
- const float x0 = get_element(x[i], k);
+ const float x0 = ggml_get_f32_1d(x[i], k);
const float xm = x0 - eps;
const float xp = x0 + eps;
- set_element(x[i], k, xp);
+ ggml_set_f32_1d(x[i], k, xp);
ggml_graph_compute_with_ctx(ctx0, &gf, n_threads);
const float f0 = ggml_get_f32_1d(f, 0);
- set_element(x[i], k, xm);
+ ggml_set_f32_1d(x[i], k, xm);
ggml_graph_compute_with_ctx(ctx0, &gf, n_threads);
const float f1 = ggml_get_f32_1d(f, 0);
const float g0 = (f0 - f1)/(2.0f*eps);
- set_element(x[i], k, x0);
+ ggml_set_f32_1d(x[i], k, x0);
// compute gradient using backward graph
ggml_graph_reset (&gf);
ggml_graph_compute_with_ctx(ctx0, &gb, n_threads);
- const float g1 = get_element(x[i]->grad, k);
+ const float g1 = ggml_get_f32_1d(x[i]->grad, k);
const float error_abs = fabsf(g0 - g1);
const float error_rel = g0 != 0 ? fabsf(g0 - g1)/fabsf(g0) : 0;
struct ggml_tensor * x[MAX_NARGS];
- // add
+ // add f32
{
const int nargs = 2;
for (int ndims = 1; ndims <= 4; ++ndims) {
for (int i = 0; i < nargs; ++i) {
- x[i] = get_random_tensor(ctx0, ndims, ne, -1.0f, 1.0f);
+ x[i] = get_random_tensor_f32(ctx0, ndims, ne, -1.0f, 1.0f);
ggml_set_param(ctx0, x[i]);
}
struct ggml_tensor * f = ggml_sum(ctx0, ggml_add(ctx0, x[0], x[1]));
- check_gradient("add", ctx0, x, f, ndims, nargs, 1e-3f, 2e-3f, 2e-3f);
+ check_gradient("add f32", ctx0, x, f, ndims, nargs, 1e-3f, 2e-3f, 2e-3f);
+ }
+ }
+
+ // add f16
+ {
+ const int nargs = 2;
+
+ for (int ndims = 1; ndims <= 4; ++ndims) {
+ for (int i = 0; i < nargs; ++i) {
+ x[i] = get_random_tensor_f16(ctx0, ndims, ne, -1.0f, 1.0f);
+ ggml_set_param(ctx0, x[i]);
+ }
+
+ struct ggml_tensor * f = ggml_sum(ctx0, ggml_add(ctx0, x[0], x[1]));
+
+ check_gradient("add f16", ctx0, x, f, ndims, nargs, 1e-1f, 2e-1f, 2e-1f);
}
}
for (int ndims = 1; ndims <= 4; ++ndims) {
for (int i = 0; i < nargs; ++i) {
- x[i] = get_random_tensor(ctx0, ndims, ne, -1.0f, 1.0f);
+ x[i] = get_random_tensor_f32(ctx0, ndims, ne, -1.0f, 1.0f);
ggml_set_param(ctx0, x[i]);
}
for (int ndims = 1; ndims <= 4; ++ndims) {
for (int i = 0; i < nargs; ++i) {
- x[i] = get_random_tensor(ctx0, ndims, ne, -1.0f, 1.0f);
+ x[i] = get_random_tensor_f32(ctx0, ndims, ne, -1.0f, 1.0f);
ggml_set_param(ctx0, x[i]);
}
for (int ndims = 1; ndims <= 4; ++ndims) {
for (int i = 0; i < nargs; ++i) {
- x[i] = get_random_tensor(ctx0, ndims, ne, 0.5f, 1.0f);
+ x[i] = get_random_tensor_f32(ctx0, ndims, ne, 0.5f, 1.0f);
ggml_set_param(ctx0, x[i]);
}
for (int ndims = 1; ndims <= 2; ++ndims) {
for (int i = 0; i < nargs; ++i) {
- x[i] = get_random_tensor(ctx0, ndims, ne, -1.0f, 1.0f);
+ x[i] = get_random_tensor_f32(ctx0, ndims, ne, -1.0f, 1.0f);
ggml_set_param(ctx0, x[i]);
}
for (int ndims = 1; ndims <= 2; ++ndims) {
for (int i = 0; i < nargs; ++i) {
- x[i] = get_random_tensor(ctx0, ndims, ne, 2.0f*1e-3f, 1.0f);
+ x[i] = get_random_tensor_f32(ctx0, ndims, ne, 2.0f*1e-3f, 1.0f);
ggml_set_param(ctx0, x[i]);
}
for (int ndims = 1; ndims <= 2; ++ndims) {
for (int i = 0; i < nargs; ++i) {
- x[i] = get_random_tensor(ctx0, ndims, ne, 2.0f*1e-3f, 1.0f);
+ x[i] = get_random_tensor_f32(ctx0, ndims, ne, 2.0f*1e-3f, 1.0f);
ggml_set_param(ctx0, x[i]);
}
for (int ndims = 1; ndims <= 2; ++ndims) {
for (int i = 0; i < nargs; ++i) {
- x[i] = get_random_tensor(ctx0, ndims, ne, -1.0f, 1.0f);
+ x[i] = get_random_tensor_f32(ctx0, ndims, ne, -1.0f, 1.0f);
ggml_set_param(ctx0, x[i]);
}
for (int ndims = 1; ndims <= 4; ++ndims) {
for (int i = 0; i < nargs; ++i) {
- x[i] = get_random_tensor(ctx0, ndims, ne, -1.0f, 1.0f);
+ x[i] = get_random_tensor_f32(ctx0, ndims, ne, -1.0f, 1.0f);
ggml_set_param(ctx0, x[i]);
}
}
}
+ // mean, not yet fully implemented
+ if(0)
+ {
+ const int nargs = 1;
+
+ for (int ndims = 1; ndims <= 4; ++ndims) {
+ for (int i = 0; i < nargs; ++i) {
+ x[i] = get_random_tensor_f32(ctx0, ndims, ne, -1.0f, 1.0f);
+ ggml_set_param(ctx0, x[i]);
+ }
+
+ struct ggml_tensor * f = ggml_sum(ctx0, ggml_mean(ctx0, x[0]));
+
+ check_gradient("mean", ctx0, x, f, ndims, nargs, 1e-3f, 1e-3f, 1e-3f);
+ }
+ }
+
+ // argmax
+ if (0)
+ {
+ const int nargs = 1;
+
+ for (int ndims = 1; ndims <= 4; ++ndims) {
+ for (int i = 0; i < nargs; ++i) {
+ x[i] = get_random_tensor_f32(ctx0, ndims, ne, -1.0f, 1.0f);
+ ggml_set_param(ctx0, x[i]);
+ }
+
+ struct ggml_tensor * f = ggml_sum(ctx0, ggml_argmax(ctx0, x[0]));
+
+ check_gradient("argmax", ctx0, x, f, ndims, nargs, 1e-3f, 1e-3f, 1e-3f);
+ }
+ }
+
// repeat
{
int64_t ne2[4];
const int nargs = 1;
for (int ndims = 1; ndims <= 2; ++ndims) {
- x[0] = get_random_tensor(ctx0, ndims, ne, -1.0f, 1.0f);
- x[1] = get_random_tensor(ctx0, ndims, ne2, -1.0f, 1.0f);
+ x[0] = get_random_tensor_f32(ctx0, ndims, ne, -1.0f, 1.0f);
+ x[1] = get_random_tensor_f32(ctx0, ndims, ne2, -1.0f, 1.0f);
ggml_set_param(ctx0, x[0]);
struct ggml_tensor * f = ggml_sum(ctx0, ggml_sqr(ctx0, ggml_sub(ctx0, x[1], ggml_repeat(ctx0, x[0], x[1]))));
check_gradient("repeat", ctx0, x, f, ndims, nargs, 1e-3f, 1e-2f, INFINITY);
}
+ }
+
+ // repeat back
+ {
+ int64_t ne2[4];
+ get_random_dims(ne2, 4);
+
+ ne2[0] = ne[0] * ne2[0];
+ ne2[1] = ne[1] * ne2[1];
+ ne2[2] = 1;
+ ne2[3] = 1;
+
+ const int nargs = 1;
+ for (int ndims = 1; ndims <= 2; ++ndims) {
+ x[0] = get_random_tensor_f32(ctx0, ndims, ne, -1.0f, 1.0f);
+ x[1] = get_random_tensor_f32(ctx0, ndims, ne2, -1.0f, 1.0f);
+ ggml_set_param(ctx0, x[0]);
+
+ struct ggml_tensor * f = ggml_sum(ctx0, ggml_sqr(ctx0, ggml_sub(ctx0, x[0], ggml_repeat_back(ctx0, x[1], x[0]))));
+ check_gradient("repeat back", ctx0, x, f, ndims, nargs, 1e-3f, 1e-2f, INFINITY);
+ }
}
// abs (finite differences do not work)
// for (int ndims = 1; ndims <= 2; ++ndims) {
// for (int i = 0; i < nargs; ++i) {
- // x[i] = get_random_tensor(ctx0, ndims, ne, -1.0f, 1.0f);
+ // x[i] = get_random_tensor_f32(ctx0, ndims, ne, -1.0f, 1.0f);
// ggml_set_param(ctx0, x[i]);
// }
// }
//}
+ // sgn
+ {
+ const int nargs = 1;
+
+ for (int ndims = 1; ndims <= 4; ++ndims) {
+ for (int i = 0; i < nargs; ++i) {
+ x[i] = get_random_tensor_f32(ctx0, ndims, ne, -1.0f, 1.0f);
+ ggml_set_param(ctx0, x[i]);
+ }
+
+ struct ggml_tensor* f = ggml_sum(ctx0, ggml_sgn(ctx0, x[0]));
+
+ check_gradient("sgn", ctx0, x, f, ndims, nargs, 1e-3f, 1e-3f, 1e-3f);
+ }
+ }
+
+ // neg
+ {
+ const int nargs = 1;
+
+ for (int ndims = 1; ndims <= 4; ++ndims) {
+ for (int i = 0; i < nargs; ++i) {
+ x[i] = get_random_tensor_f32(ctx0, ndims, ne, -1.0f, 1.0f);
+ ggml_set_param(ctx0, x[i]);
+ }
+
+ struct ggml_tensor* f = ggml_sum(ctx0, ggml_neg(ctx0, x[0]));
+
+ check_gradient("neg", ctx0, x, f, ndims, nargs, 1e-3f, 1e-3f, 1e-3f);
+ }
+ }
+
+ // step
+ {
+ const int nargs = 1;
+
+ for (int ndims = 1; ndims <= 4; ++ndims) {
+ for (int i = 0; i < nargs; ++i) {
+ x[i] = get_random_tensor_f32(ctx0, ndims, ne, -1.0f, 1.0f);
+ ggml_set_param(ctx0, x[i]);
+ }
+
+ struct ggml_tensor* f = ggml_sum(ctx0, ggml_step(ctx0, x[0]));
+
+ check_gradient("step", ctx0, x, f, ndims, nargs, 1e-3f, 1e-3f, 1e-3f);
+ }
+ }
+
+ // tanh, not yet fully implemented
+ if(0)
+ {
+ const int nargs = 1;
+
+ for (int ndims = 1; ndims <= 4; ++ndims) {
+ for (int i = 0; i < nargs; ++i) {
+ x[i] = get_random_tensor_f32(ctx0, ndims, ne, -1.0f, 1.0f);
+ ggml_set_param(ctx0, x[i]);
+ }
+
+ struct ggml_tensor* f = ggml_sum(ctx0, ggml_tanh(ctx0, x[0]));
+
+ check_gradient("tanh", ctx0, x, f, ndims, nargs, 1e-3f, 1e-3f, 1e-3f);
+ }
+ }
+
// mul_mat
{
const int nargs = 2;
for (int ndims = 2; ndims <= 2; ++ndims) {
- x[0] = get_random_tensor(ctx0, ndims, ne, -1.0f, 1.0f);
+ x[0] = get_random_tensor_f32(ctx0, ndims, ne, -1.0f, 1.0f);
{
int64_t ne2[4];
get_random_dims(ne2, 4);
ne2[0] = ne[0];
- x[1] = get_random_tensor(ctx0, ndims, ne2, -1.0f, 1.0f);
+ x[1] = get_random_tensor_f32(ctx0, ndims, ne2, -1.0f, 1.0f);
}
ggml_set_param(ctx0, x[0]);
}
}
+ // elu, not yet fully implemented
+ if(0)
+ {
+ const int nargs = 1;
+
+ for (int ndims = 1; ndims <= 4; ++ndims) {
+ for (int i = 0; i < nargs; ++i) {
+ x[i] = get_random_tensor_f32(ctx0, ndims, ne, -1.0f, 1.0f);
+ ggml_set_param(ctx0, x[i]);
+ }
+
+ struct ggml_tensor* f = ggml_sum(ctx0, ggml_elu(ctx0, x[0]));
+
+ check_gradient("elu", ctx0, x, f, ndims, nargs, 1e-3f, 1e-3f, 1e-3f);
+ }
+ }
+
+ // relu
+ {
+ const int nargs = 1;
+
+ for (int ndims = 1; ndims <= 4; ++ndims) {
+ for (int i = 0; i < nargs; ++i) {
+ x[i] = get_random_tensor_f32(ctx0, ndims, ne, -1.0f, 1.0f);
+ ggml_set_param(ctx0, x[i]);
+ }
+
+ struct ggml_tensor* f = ggml_sum(ctx0, ggml_relu(ctx0, x[0]));
+
+ check_gradient("relu", ctx0, x, f, ndims, nargs, 1e-3f, 1e-3f, INFINITY);
+ }
+ }
+
+ // gelu, not yet fully implemented
+ if(0)
+ {
+ const int nargs = 1;
+
+ for (int ndims = 1; ndims <= 4; ++ndims) {
+ for (int i = 0; i < nargs; ++i) {
+ x[i] = get_random_tensor_f32(ctx0, ndims, ne, -1.0f, 1.0f);
+ ggml_set_param(ctx0, x[i]);
+ }
+
+ struct ggml_tensor* f = ggml_sum(ctx0, ggml_gelu(ctx0, x[0]));
+
+ check_gradient("gelu", ctx0, x, f, ndims, nargs, 1e-3f, 1e-3f, 1e-3f);
+ }
+ }
+
// silu
{
const int nargs = 1;
for (int ndims = 1; ndims <= 2; ++ndims) {
for (int i = 0; i < nargs; ++i) {
- x[i] = get_random_tensor(ctx0, ndims, ne, -1.0f, 1.0f);
+ x[i] = get_random_tensor_f32(ctx0, ndims, ne, -1.0f, 1.0f);
ggml_set_param(ctx0, x[i]);
}
for (int ndims = 1; ndims <= 2; ++ndims) {
for (int i = 0; i < nargs; ++i) {
- x[i] = get_random_tensor(ctx0, ndims, ne, -1.0f, 1.0f);
+ x[i] = get_random_tensor_f32(ctx0, ndims, ne, -1.0f, 1.0f);
ggml_set_param(ctx0, x[i]);
}
ne2[0] = 1;
for (int ndims = 1; ndims <= 2; ++ndims) {
- x[1] = get_random_tensor(ctx0, 1, ne2, -1.0f, 1.0f);
- x[0] = get_random_tensor(ctx0, ndims, ne, -1.0f, 1.0f);
+ x[1] = get_random_tensor_f32(ctx0, 1, ne2, -1.0f, 1.0f);
+ x[0] = get_random_tensor_f32(ctx0, ndims, ne, -1.0f, 1.0f);
ggml_set_param(ctx0, x[0]);
ggml_set_param(ctx0, x[1]);
}
}
- // cpy
+ // cpy f32
{
const int nargs = 2;
for (int ndims = 1; ndims <= 2; ++ndims) {
for (int i = 0; i < nargs; ++i) {
- x[i] = get_random_tensor(ctx0, ndims, ne, -1.0f, 1.0f);
+ x[i] = get_random_tensor_f32(ctx0, ndims, ne, -1.0f, 1.0f);
ggml_set_param(ctx0, x[i]);
}
// x[1] is overwritten by x[0], so the gradients don't propagate to x[1]
struct ggml_tensor * f = ggml_sum(ctx0, ggml_cpy(ctx0, x[0], x[1]));
- check_gradient("cpy", ctx0, x, f, ndims, nargs, 1e-3f, 1e-3f, INFINITY);
+ check_gradient("cpy f32", ctx0, x, f, ndims, nargs, 1e-3f, 1e-3f, INFINITY);
+ }
+ }
+
+ // cpy f16
+ {
+ const int nargs = 2;
+
+ for (int ndims = 1; ndims <= 2; ++ndims) {
+ for (int i = 0; i < nargs; ++i) {
+ x[i] = get_random_tensor_f16(ctx0, ndims, ne, -1.0f, 1.0f);
+ ggml_set_param(ctx0, x[i]);
+ }
+ // x[1] is overwritten by x[0], so the gradients don't propagate to x[1]
+
+ struct ggml_tensor * f = ggml_sum(ctx0, ggml_cpy(ctx0, x[0], x[1]));
+
+ check_gradient("cpy f16", ctx0, x, f, ndims, nargs, 1e-1f, 1e-1f, INFINITY);
}
}
for (int i = 0; i < ndims; ++i) {
ne2[0] *= ne[i];
}
- x[0] = get_random_tensor(ctx0, 1, ne2, -1.0f, 1.0f);
- x[1] = get_random_tensor(ctx0, ndims, ne, -1.0f, 1.0f);
+ x[0] = get_random_tensor_f32(ctx0, 1, ne2, -1.0f, 1.0f);
+ x[1] = get_random_tensor_f32(ctx0, ndims, ne, -1.0f, 1.0f);
ggml_set_param(ctx0, x[0]);
for (int i = 0; i < ndims; ++i) {
ne2[0] *= ne[i];
}
- x[0] = get_random_tensor(ctx0, ndims, ne, -1.0f, 1.0f);
- x[1] = get_random_tensor(ctx0, 1, ne2, -1.0f, 1.0f);
+ x[0] = get_random_tensor_f32(ctx0, ndims, ne, -1.0f, 1.0f);
+ x[1] = get_random_tensor_f32(ctx0, 1, ne2, -1.0f, 1.0f);
ggml_set_param(ctx0, x[0]);
const int nargs = 2;
for (int ndims = 1; ndims <= 4; ++ndims) {
- x[0] = get_random_tensor(ctx0, ndims, ne, -1.0f, 1.0f);
+ x[0] = get_random_tensor_f32(ctx0, ndims, ne, -1.0f, 1.0f);
ggml_set_param(ctx0, x[0]);
get_random_dims(ne2, 1);
get_random_dims(ne2, 1);
}
- x[1] = get_random_tensor(ctx0, 1, ne2, -1.0f, 1.0f);
+ x[1] = get_random_tensor_f32(ctx0, 1, ne2, -1.0f, 1.0f);
ggml_set_param(ctx0, x[1]);
const int max_offset = MAX(0, ggml_nelements(x[0]) - ggml_nelements(x[1]));
const int nargs = 2;
for (int ndims = 2; ndims <= 4; ++ndims) {
- x[0] = get_random_tensor(ctx0, ndims, ne, -1.0f, 1.0f);
+ x[0] = get_random_tensor_f32(ctx0, ndims, ne, -1.0f, 1.0f);
ggml_set_param(ctx0, x[0]);
get_random_dims(ne2, 2);
get_random_dims(ne2, 2);
}
- x[1] = get_random_tensor(ctx0, 2, ne2, -1.0f, 1.0f);
+ x[1] = get_random_tensor_f32(ctx0, 2, ne2, -1.0f, 1.0f);
ggml_set_param(ctx0, x[1]);
max_offsets[0] = MAX(0, x[0]->ne[0] - x[1]->ne[0]);
const int nargs = 2;
for (int ndims = 3; ndims <= 4; ++ndims) {
- x[0] = get_random_tensor(ctx0, ndims, ne, -1.0f, 1.0f);
+ x[0] = get_random_tensor_f32(ctx0, ndims, ne, -1.0f, 1.0f);
ggml_set_param(ctx0, x[0]);
get_random_dims(ne2, 3);
get_random_dims(ne2, 3);
}
- x[1] = get_random_tensor(ctx0, 3, ne2, -1.0f, 1.0f);
+ x[1] = get_random_tensor_f32(ctx0, 3, ne2, -1.0f, 1.0f);
ggml_set_param(ctx0, x[1]);
max_offsets[0] = MAX(0, x[0]->ne[0] - x[1]->ne[0]);
const int nargs = 2;
for (int ndims = 4; ndims <= 4; ++ndims) {
- x[0] = get_random_tensor(ctx0, ndims, ne, -1.0f, 1.0f);
+ x[0] = get_random_tensor_f32(ctx0, ndims, ne, -1.0f, 1.0f);
ggml_set_param(ctx0, x[0]);
get_random_dims(ne2, 4);
get_random_dims(ne2, 4);
}
- x[1] = get_random_tensor(ctx0, 4, ne2, -1.0f, 1.0f);
+ x[1] = get_random_tensor_f32(ctx0, 4, ne2, -1.0f, 1.0f);
ggml_set_param(ctx0, x[1]);
max_offsets[0] = MAX(0, x[0]->ne[0] - x[1]->ne[0]);
const int nargs = 2;
for (int ndims = 1; ndims <= 4; ++ndims) {
- x[0] = get_random_tensor(ctx0, ndims, ne, -1.0f, 1.0f);
+ x[0] = get_random_tensor_f32(ctx0, ndims, ne, -1.0f, 1.0f);
ggml_set_param(ctx0, x[0]);
get_random_dims(ne2, 1);
get_random_dims(ne2, 1);
}
- x[1] = get_random_tensor(ctx0, 1, ne2, -1.0f, 1.0f);
+ x[1] = get_random_tensor_f32(ctx0, 1, ne2, -1.0f, 1.0f);
ggml_set_param(ctx0, x[1]);
const int max_offset = MAX(0, ggml_nelements(x[0]) - ggml_nelements(x[1]));
const int nargs = 1;
for (int ndims = 2; ndims <= 4; ++ndims) {
- x[0] = get_random_tensor(ctx0, ndims, ne, -1.0f, 1.0f);
+ x[0] = get_random_tensor_f32(ctx0, ndims, ne, -1.0f, 1.0f);
ggml_set_param(ctx0, x[0]);
get_random_dims(ne2, 2);
get_random_dims(ne2, 2);
}
- x[1] = get_random_tensor(ctx0, 2, ne2, -1.0f, 1.0f);
+ x[1] = get_random_tensor_f32(ctx0, 2, ne2, -1.0f, 1.0f);
ggml_set_param(ctx0, x[1]);
max_offsets[0] = MAX(0, x[0]->ne[0] - x[1]->ne[0]);
const int nargs = 1;
for (int ndims = 1; ndims <= 4; ++ndims) {
- x[0] = get_random_tensor(ctx0, ndims, ne, -1.0f, 1.0f);
+ x[0] = get_random_tensor_f32(ctx0, ndims, ne, -1.0f, 1.0f);
ggml_set_param(ctx0, x[0]);
const int nargs = 1;
for (int ndims = 1; ndims <= 4; ++ndims) {
- x[0] = get_random_tensor(ctx0, ndims, ne, -1.0f, 1.0f);
+ x[0] = get_random_tensor_f32(ctx0, ndims, ne, -1.0f, 1.0f);
get_random_dims(ne2, 2);
while (ne2[0]*ne2[1] > ggml_nelements(x[0])) {
const int nargs = 1;
for (int ndims = 1; ndims <= 4; ++ndims) {
- x[0] = get_random_tensor(ctx0, ndims, ne, -1.0f, 1.0f);
+ x[0] = get_random_tensor_f32(ctx0, ndims, ne, -1.0f, 1.0f);
get_random_dims(ne2, 3);
while (ne2[0]*ne2[1]*ne2[2] > ggml_nelements(x[0])) {
for (int i=ndims; i<4; ++i) {
ne2[i] = 1;
}
- x[0] = get_random_tensor(ctx0, 4, ne2, -1.0f, 1.0f);
+ x[0] = get_random_tensor_f32(ctx0, 4, ne2, -1.0f, 1.0f);
ggml_set_param(ctx0, x[0]);
for (int i=ndims; i<4; ++i) {
ne2[i] = 1;
}
- x[0] = get_random_tensor(ctx0, 4, ne2, -1.0f, 1.0f);
+ x[0] = get_random_tensor_f32(ctx0, 4, ne2, -1.0f, 1.0f);
ggml_set_param(ctx0, x[0]);
int64_t ne3[4] = {1+irand(ne[1]), 1, 1, 1};
const int nargs = 1;
const int ndims = 2;
- x[0] = get_random_tensor(ctx0, ndims, ne2, -1.0f, 1.0f);
- x[1] = get_random_tensor_int(ctx0, 1, ne3, 0, ne2[1]);
+ x[0] = get_random_tensor_f32(ctx0, ndims, ne2, -1.0f, 1.0f);
+ x[1] = get_random_tensor_i32(ctx0, 1, ne3, 0, ne2[1]);
ggml_set_param(ctx0, x[0]);
const int nargs = 1;
const int ndims = 2;
- x[0] = get_random_tensor(ctx0, ndims, ne, -1.0f, 1.0f);
+ x[0] = get_random_tensor_f32(ctx0, ndims, ne, -1.0f, 1.0f);
ggml_set_param(ctx0, x[0]);
int n_past = irand(ne[0]);
const int nargs = 1;
const int ndims = 2;
- x[0] = get_random_tensor(ctx0, ndims, ne, -1.0f, 1.0f);
+ x[0] = get_random_tensor_f32(ctx0, ndims, ne, -1.0f, 1.0f);
ggml_set_param(ctx0, x[0]);
int n_past = irand(ne[0]);
get_random_dims(ne2, 4);
for (int ndims = 1; ndims <= 3; ++ndims) {
- x[0] = get_random_tensor(ctx0, ndims, ne2, -1.0f, 1.0f);
+ x[0] = get_random_tensor_f32(ctx0, ndims, ne2, -1.0f, 1.0f);
ggml_set_param(ctx0, x[0]);
struct ggml_tensor * f = ggml_sum(ctx0, ggml_soft_max(ctx0, x[0]));
get_random_dims(ne2, 4);
for (int ndims = 1; ndims <= 3; ++ndims) {
- x[0] = get_random_tensor(ctx0, ndims, ne2, -1.0f, 1.0f);
- x[1] = get_random_tensor(ctx0, ndims, ne2, 0.0f, 1.0f);
+ x[0] = get_random_tensor_f32(ctx0, ndims, ne2, -1.0f, 1.0f);
+ x[1] = get_random_tensor_f32(ctx0, ndims, ne2, 0.0f, 1.0f);
ggml_set_param(ctx0, x[0]);
struct ggml_tensor * f = ggml_sum(ctx0, ggml_cross_entropy_loss(ctx0, x[0], x[1]));
}
}
- // rope
+ // rope f32
{
const int nargs = 1;
for (int ndims = 3; ndims <= 4; ++ndims) {
for (int mode = 0; mode < 4; ++mode) {
for (int n_past = 1; n_past < ne2[2]; ++n_past) {
- x[0] = get_random_tensor(ctx0, ndims, ne2, -1.0f, 1.0f);
+ x[0] = get_random_tensor_f32(ctx0, ndims, ne2, -1.0f, 1.0f);
ggml_set_param(ctx0, x[0]);
struct ggml_tensor * f = ggml_sum(ctx0, ggml_rope(ctx0, x[0], n_past, n_rot, mode, 0));
- GGML_PRINT_DEBUG("rope: n_past: %d n_rot: %d mode: %d\n", n_past, n_rot, mode);
- check_gradient("rope", ctx0, x, f, ndims, nargs, 1e-2f, 1e-3f, INFINITY);
+ GGML_PRINT_DEBUG("rope f32: n_past: %d n_rot: %d mode: %d\n", n_past, n_rot, mode);
+ check_gradient("rope f32", ctx0, x, f, ndims, nargs, 1e-2f, 1e-3f, INFINITY);
+ }
+ }
+ }
+ }
+
+ // rope f16
+ {
+ const int nargs = 1;
+
+ int64_t ne2[4];
+ get_random_dims(ne2, 4);
+ ne2[0] += ne2[0] % 2;
+ int n_rot = ne2[0];
+
+ for (int ndims = 3; ndims <= 4; ++ndims) {
+ for (int mode = 0; mode < 4; ++mode) {
+ for (int n_past = 1; n_past < ne2[2]; ++n_past) {
+ x[0] = get_random_tensor_f16(ctx0, ndims, ne2, -1.0f, 1.0f);
+
+ ggml_set_param(ctx0, x[0]);
+
+ const bool skip_past = (mode & 1);
+ if (skip_past) {
+ // we have no past, so this would have to work on uninitialized memory.
+ // we only test the gradients here;
+ // skip_past should have no influence on gradient computation.
+ // so when other modes work, we assume that this does as well.
+ continue;
+ }
+
+ struct ggml_tensor * f = ggml_sum(ctx0, ggml_rope(ctx0, x[0], n_past, n_rot, mode, 0));
+
+ GGML_PRINT_DEBUG("rope f16: n_past: %d n_rot: %d mode: %d\n", n_past, n_rot, mode);
+ check_gradient("rope f16", ctx0, x, f, ndims, nargs, 1e-1f, 1e-1f, INFINITY);
+ }
+ }
+ }
+ }
+
+ // flash_attn f32
+ {
+ const int nargs = 3;
+
+ int64_t ne2[4];
+
+ get_random_dims(ne2, 4);
+ int64_t D = ne2[0];
+ int64_t N = ne2[1];
+ int64_t M = ne2[2] + N;
+ int64_t B = ne2[3];
+
+ for (int masked = 0; masked <= 1; ++masked) {
+ for (int ndims = 2; ndims <= 4; ++ndims) {
+ int64_t neq[4] = { D, N, B, ne[3] };
+ int64_t nek[4] = { D, M, B, ne[3] };
+ int64_t nev[4] = { M, D, B, ne[3] };
+ if (ndims == 2) {
+ neq[2] = 1; neq[3] = 1;
+ nek[2] = 1; nek[3] = 1;
+ nev[2] = 1; nev[3] = 1;
+ } else if (ndims == 3) {
+ neq[3] = 1;
+ nek[3] = 1;
+ nev[3] = 1;
}
+ x[0] = get_random_tensor_f32(ctx0, ndims, neq, -0.1250f, 0.1250f);
+ x[1] = get_random_tensor_f32(ctx0, ndims, nek, -0.1250f, 0.1250f);
+ x[2] = get_random_tensor_f32(ctx0, ndims, nev, -0.1250f, 0.1250f);
+ ggml_set_param(ctx0, x[0]);
+ ggml_set_param(ctx0, x[1]);
+ ggml_set_param(ctx0, x[2]);
+
+ struct ggml_tensor * f = ggml_sum(ctx0, ggml_flash_attn(ctx0, x[0], x[1], x[2], (masked == 0)));
+
+ check_gradient("flash_attn f32", ctx0, x, f, ndims, nargs, 1.5e-4f, INFINITY, 3.5f);
}
}
}
- // flash_attn
+ // flash_attn f16, not yet fully implemented
+ if(0)
{
const int nargs = 3;
nek[3] = 1;
nev[3] = 1;
}
- x[0] = get_random_tensor(ctx0, ndims, neq, -0.1250f, 0.1250f);
- x[1] = get_random_tensor(ctx0, ndims, nek, -0.1250f, 0.1250f);
- x[2] = get_random_tensor(ctx0, ndims, nev, -0.1250f, 0.1250f);
+ x[0] = get_random_tensor_f16(ctx0, ndims, neq, -0.1250f, 0.1250f);
+ x[1] = get_random_tensor_f16(ctx0, ndims, nek, -0.1250f, 0.1250f);
+ x[2] = get_random_tensor_f16(ctx0, ndims, nev, -0.1250f, 0.1250f);
ggml_set_param(ctx0, x[0]);
ggml_set_param(ctx0, x[1]);
ggml_set_param(ctx0, x[2]);
struct ggml_tensor * f = ggml_sum(ctx0, ggml_flash_attn(ctx0, x[0], x[1], x[2], (masked == 0)));
- check_gradient("flash_attn", ctx0, x, f, ndims, nargs, 1.5e-4f, INFINITY, 3.5f);
+ check_gradient("flash_attn f16", ctx0, x, f, ndims, nargs, 1.5e-4f, INFINITY, 3.5f);
}
}
}
};
struct ggml_context * ctx = ggml_init(params);
- int64_t ne1[4] = {4, 1024, 1, 1};
- int64_t ne2[4] = {4, 2048, 1, 1};;
- int64_t ne3[4] = {1024, 2048, 1, 1};
+ int64_t ne1[4] = {4, 128, 1, 1};
+ int64_t ne2[4] = {4, 256, 1, 1};;
+ int64_t ne3[4] = {128, 256, 1, 1};
struct ggml_tensor * a = get_random_tensor(ctx, 2, ne1, -1, +1);
struct ggml_tensor * b = get_random_tensor(ctx, 2, ne2, -1, +1);
overview / pkg / ggml / sources / llama.cpp / commitdiff