First rough draft of recoverable errors feature.
This commit is contained in:
KerfuffleV2
parent
3924088512
commit
1a6987ad92
2 changed files with 145 additions and 19 deletions
116
ggml.c
116
ggml.c
|
|
@ -4245,6 +4245,10 @@ struct ggml_context {
|
|||
|
||||
struct ggml_scratch scratch;
|
||||
struct ggml_scratch scratch_save;
|
||||
#ifdef GGML_RECOVERABLE_ERRORS
|
||||
enum ggml_errcode last_error_code;
|
||||
char last_error_msg[GGML_ERROR_BUFFER_LEN];
|
||||
#endif
|
||||
};
|
||||
|
||||
struct ggml_context_container {
|
||||
|
|
@ -4566,6 +4570,10 @@ struct ggml_context * ggml_init(struct ggml_init_params params) {
|
|||
/*.objects_end =*/ NULL,
|
||||
/*.scratch =*/ { 0, 0, NULL, },
|
||||
/*.scratch_save =*/ { 0, 0, NULL, },
|
||||
#ifdef GGML_RECOVERABLE_ERRORS
|
||||
/*.last_error_code =*/ GGML_ERRCODE_SUCCESS,
|
||||
/*.last_error_msg =*/ { 0 },
|
||||
#endif
|
||||
};
|
||||
|
||||
GGML_ASSERT(ctx->mem_buffer != NULL);
|
||||
|
|
@ -4620,6 +4628,24 @@ size_t ggml_set_scratch(struct ggml_context * ctx, struct ggml_scratch scratch)
|
|||
return result;
|
||||
}
|
||||
|
||||
#ifdef GGML_RECOVERABLE_ERRORS
|
||||
// enum ggml_errcode ggml_clear_error(struct ggml_context * ctx) {
|
||||
// enum ggml_errcode last_code = ctx->last_error_code;
|
||||
|
||||
// ctx->last_error_code = GGML_ERRCODE_SUCCESS;
|
||||
// ctx->last_error_msg[0] = 0;
|
||||
// return last_code;
|
||||
// }
|
||||
|
||||
enum ggml_errcode ggml_last_error_code(struct ggml_context * ctx) {
|
||||
return ctx->last_error_code;
|
||||
}
|
||||
|
||||
char * ggml_last_error_msg(struct ggml_context * ctx) {
|
||||
return ctx->last_error_code == GGML_ERRCODE_SUCCESS ? "Success" : &ctx->last_error_msg;
|
||||
}
|
||||
#endif
|
||||
|
||||
////////////////////////////////////////////////////////////////////////////////
|
||||
|
||||
struct ggml_tensor * ggml_new_tensor_impl(
|
||||
|
|
@ -4652,12 +4678,11 @@ struct ggml_tensor * ggml_new_tensor_impl(
|
|||
if (ctx->scratch.data == NULL || data != NULL) {
|
||||
size_needed += sizeof(struct ggml_tensor);
|
||||
|
||||
if (cur_end + size_needed + GGML_OBJECT_SIZE > ctx->mem_size) {
|
||||
GGML_PRINT("%s: not enough space in the context's memory pool (needed %zu, available %zu)\n",
|
||||
__func__, cur_end + size_needed + GGML_OBJECT_SIZE, ctx->mem_size);
|
||||
assert(false);
|
||||
return NULL;
|
||||
}
|
||||
GGML_RECOVERABLE_ASSERT(ctx, GGML_ERRCODE_CTX_MEMORY_LIMIT,
|
||||
cur_end + size_needed + GGML_OBJECT_SIZE <= ctx->mem_size,
|
||||
"%s: not enough space in the context's memory pool (needed %zu, available %zu)\n",
|
||||
__func__, cur_end + size_needed + GGML_OBJECT_SIZE, ctx->mem_size
|
||||
);
|
||||
|
||||
*obj_new = (struct ggml_object) {
|
||||
.offs = cur_end + GGML_OBJECT_SIZE,
|
||||
|
|
@ -4665,18 +4690,16 @@ struct ggml_tensor * ggml_new_tensor_impl(
|
|||
.next = NULL,
|
||||
};
|
||||
} else {
|
||||
if (ctx->scratch.offs + size_needed > ctx->scratch.size) {
|
||||
GGML_PRINT("%s: not enough space in the scratch memory\n", __func__);
|
||||
assert(false);
|
||||
return NULL;
|
||||
}
|
||||
GGML_RECOVERABLE_ASSERT(ctx, GGML_ERRCODE_SCRATCH_MEMORY_LIMIT,
|
||||
ctx->scratch.offs + size_needed <= ctx->scratch.size,
|
||||
"%s: not enough space in the scratch memory\n", __func__
|
||||
);
|
||||
|
||||
if (cur_end + sizeof(struct ggml_tensor) + GGML_OBJECT_SIZE > ctx->mem_size) {
|
||||
GGML_PRINT("%s: not enough space in the context's memory pool (needed %zu, available %zu)\n",
|
||||
__func__, cur_end + sizeof(struct ggml_tensor) + GGML_OBJECT_SIZE, ctx->mem_size);
|
||||
assert(false);
|
||||
return NULL;
|
||||
}
|
||||
GGML_RECOVERABLE_ASSERT(ctx, GGML_ERRCODE_CTX_MEMORY_LIMIT,
|
||||
cur_end + sizeof(struct ggml_tensor) + GGML_OBJECT_SIZE <= ctx->mem_size,
|
||||
"%s: not enough space in the context's memory pool (needed %zu, available %zu)\n",
|
||||
__func__, cur_end + sizeof(struct ggml_tensor) + GGML_OBJECT_SIZE, ctx->mem_size
|
||||
);
|
||||
|
||||
data = (char * const) ctx->scratch.data + ctx->scratch.offs;
|
||||
|
||||
|
|
@ -4794,6 +4817,7 @@ struct ggml_tensor * ggml_new_i32(struct ggml_context * ctx, int32_t value) {
|
|||
ctx->scratch.data = NULL;
|
||||
|
||||
struct ggml_tensor * result = ggml_new_tensor_1d(ctx, GGML_TYPE_I32, 1);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
ctx->scratch = ctx->scratch_save;
|
||||
|
||||
|
|
@ -4807,6 +4831,7 @@ struct ggml_tensor * ggml_new_f32(struct ggml_context * ctx, float value) {
|
|||
ctx->scratch.data = NULL;
|
||||
|
||||
struct ggml_tensor * result = ggml_new_tensor_1d(ctx, GGML_TYPE_F32, 1);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
ctx->scratch = ctx->scratch_save;
|
||||
|
||||
|
|
@ -5090,6 +5115,7 @@ struct ggml_tensor * ggml_view_tensor(
|
|||
struct ggml_context * ctx,
|
||||
const struct ggml_tensor * src) {
|
||||
struct ggml_tensor * result = ggml_new_tensor_impl(ctx, src->type, src->n_dims, src->ne, src->data);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
result->nb[0] = src->nb[0];
|
||||
result->nb[1] = src->nb[1];
|
||||
|
|
@ -5114,6 +5140,7 @@ struct ggml_tensor * ggml_dup_impl(
|
|||
}
|
||||
|
||||
struct ggml_tensor * result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
result->op = GGML_OP_DUP;
|
||||
result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
|
||||
|
|
@ -5142,7 +5169,10 @@ struct ggml_tensor * ggml_add_impl(
|
|||
struct ggml_tensor * a,
|
||||
struct ggml_tensor * b,
|
||||
bool inplace) {
|
||||
GGML_ASSERT(ggml_are_same_shape(a, b));
|
||||
GGML_RECOVERABLE_ASSERT(ctx, GGML_ERRCODE_BAD_SHAPE,
|
||||
ggml_are_same_shape(a, b),
|
||||
"Shape mismatch in ggml_add"
|
||||
);
|
||||
|
||||
bool is_node = false;
|
||||
|
||||
|
|
@ -5151,6 +5181,7 @@ struct ggml_tensor * ggml_add_impl(
|
|||
}
|
||||
|
||||
struct ggml_tensor * result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
result->op = GGML_OP_ADD;
|
||||
result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
|
||||
|
|
@ -5190,6 +5221,7 @@ struct ggml_tensor * ggml_sub_impl(
|
|||
}
|
||||
|
||||
struct ggml_tensor * result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
result->op = GGML_OP_SUB;
|
||||
result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
|
||||
|
|
@ -5233,6 +5265,7 @@ struct ggml_tensor * ggml_mul_impl(
|
|||
}
|
||||
|
||||
struct ggml_tensor * result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
result->op = GGML_OP_MUL;
|
||||
result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
|
||||
|
|
@ -5276,6 +5309,7 @@ struct ggml_tensor * ggml_div_impl(
|
|||
}
|
||||
|
||||
struct ggml_tensor * result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
result->op = GGML_OP_DIV;
|
||||
result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
|
||||
|
|
@ -5312,6 +5346,7 @@ struct ggml_tensor * ggml_sqr_impl(
|
|||
}
|
||||
|
||||
struct ggml_tensor * result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
result->op = GGML_OP_SQR;
|
||||
result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
|
||||
|
|
@ -5346,6 +5381,7 @@ struct ggml_tensor * ggml_sqrt_impl(
|
|||
}
|
||||
|
||||
struct ggml_tensor * result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
result->op = GGML_OP_SQRT;
|
||||
result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
|
||||
|
|
@ -5379,6 +5415,7 @@ struct ggml_tensor * ggml_sum(
|
|||
}
|
||||
|
||||
struct ggml_tensor * result = ggml_new_tensor_1d(ctx, a->type, 1);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
result->op = GGML_OP_SUM;
|
||||
result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
|
||||
|
|
@ -5402,6 +5439,7 @@ struct ggml_tensor * ggml_mean(
|
|||
|
||||
int64_t ne[GGML_MAX_DIMS] = { 1, a->ne[1], a->ne[2], a->ne[3] };
|
||||
struct ggml_tensor * result = ggml_new_tensor(ctx, GGML_TYPE_F32, a->n_dims, ne);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
result->op = GGML_OP_MEAN;
|
||||
result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
|
||||
|
|
@ -5430,6 +5468,7 @@ struct ggml_tensor * ggml_repeat(
|
|||
}
|
||||
|
||||
struct ggml_tensor * result = ggml_new_tensor(ctx, a->type, b->n_dims, b->ne);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
result->op = GGML_OP_REPEAT;
|
||||
result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
|
||||
|
|
@ -5452,6 +5491,7 @@ struct ggml_tensor * ggml_abs_impl(
|
|||
}
|
||||
|
||||
struct ggml_tensor * result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
result->op = GGML_OP_ABS;
|
||||
result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
|
||||
|
|
@ -5487,6 +5527,7 @@ struct ggml_tensor * ggml_sgn_impl(
|
|||
}
|
||||
|
||||
struct ggml_tensor * result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
result->op = GGML_OP_SGN;
|
||||
result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
|
||||
|
|
@ -5521,6 +5562,7 @@ struct ggml_tensor * ggml_neg_impl(
|
|||
}
|
||||
|
||||
struct ggml_tensor * result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
result->op = GGML_OP_NEG;
|
||||
result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
|
||||
|
|
@ -5555,6 +5597,7 @@ struct ggml_tensor * ggml_step_impl(
|
|||
}
|
||||
|
||||
struct ggml_tensor * result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
result->op = GGML_OP_STEP;
|
||||
result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
|
||||
|
|
@ -5589,6 +5632,7 @@ struct ggml_tensor * ggml_relu_impl(
|
|||
}
|
||||
|
||||
struct ggml_tensor * result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
result->op = GGML_OP_RELU;
|
||||
result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
|
||||
|
|
@ -5623,6 +5667,7 @@ struct ggml_tensor * ggml_gelu_impl(
|
|||
}
|
||||
|
||||
struct ggml_tensor * result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
result->op = GGML_OP_GELU;
|
||||
result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
|
||||
|
|
@ -5657,6 +5702,7 @@ struct ggml_tensor * ggml_silu_impl(
|
|||
}
|
||||
|
||||
struct ggml_tensor * result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
result->op = GGML_OP_SILU;
|
||||
result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
|
||||
|
|
@ -5692,6 +5738,7 @@ struct ggml_tensor * ggml_norm_impl(
|
|||
}
|
||||
|
||||
struct ggml_tensor * result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
result->op = GGML_OP_NORM;
|
||||
result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
|
||||
|
|
@ -5725,6 +5772,7 @@ struct ggml_tensor * ggml_rms_norm_impl(
|
|||
}
|
||||
|
||||
struct ggml_tensor * result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
result->op = GGML_OP_RMS_NORM;
|
||||
result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
|
||||
|
|
@ -5763,6 +5811,7 @@ struct ggml_tensor * ggml_mul_mat(
|
|||
|
||||
const int64_t ne[4] = { a->ne[1], b->ne[1], a->ne[2], b->ne[3] };
|
||||
struct ggml_tensor * result = ggml_new_tensor(ctx, GGML_TYPE_F32, MIN(a->n_dims, b->n_dims), ne);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
result->op = GGML_OP_MUL_MAT;
|
||||
result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
|
||||
|
|
@ -5792,6 +5841,7 @@ struct ggml_tensor * ggml_scale_impl(
|
|||
// TODO: when implement backward, fix this:
|
||||
//struct ggml_tensor * result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
|
||||
struct ggml_tensor * result = ggml_view_tensor(ctx, a);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
result->op = GGML_OP_SCALE;
|
||||
result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
|
||||
|
|
@ -5833,6 +5883,7 @@ struct ggml_tensor * ggml_cpy_impl(
|
|||
|
||||
// make a view of the destination
|
||||
struct ggml_tensor * result = ggml_view_tensor(ctx, b);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
result->op = GGML_OP_CPY;
|
||||
result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
|
||||
|
|
@ -5870,6 +5921,7 @@ struct ggml_tensor * ggml_cont_impl(
|
|||
}
|
||||
|
||||
struct ggml_tensor * result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
result->op = GGML_OP_CONT;
|
||||
result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
|
||||
|
|
@ -5909,6 +5961,7 @@ struct ggml_tensor * ggml_reshape(
|
|||
}
|
||||
|
||||
struct ggml_tensor * result = ggml_new_tensor_impl(ctx, a->type, b->n_dims, b->ne, a->data);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
result->op = GGML_OP_RESHAPE;
|
||||
result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
|
||||
|
|
@ -5935,6 +5988,7 @@ struct ggml_tensor * ggml_reshape_2d(
|
|||
|
||||
const int64_t ne[2] = { ne0, ne1 };
|
||||
struct ggml_tensor * result = ggml_new_tensor_impl(ctx, a->type, 2, ne, a->data);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
result->op = GGML_OP_RESHAPE;
|
||||
result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
|
||||
|
|
@ -5962,6 +6016,7 @@ struct ggml_tensor * ggml_reshape_3d(
|
|||
|
||||
const int64_t ne[3] = { ne0, ne1, ne2 };
|
||||
struct ggml_tensor * result = ggml_new_tensor_impl(ctx, a->type, 3, ne, a->data);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
result->op = GGML_OP_RESHAPE;
|
||||
result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
|
||||
|
|
@ -5983,6 +6038,7 @@ struct ggml_tensor * ggml_view_1d(
|
|||
}
|
||||
|
||||
struct ggml_tensor * result = ggml_new_tensor_impl(ctx, a->type, 1, &ne0, (char *) a->data + offset);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
result->op = GGML_OP_VIEW;
|
||||
result->grad = NULL;
|
||||
|
|
@ -6008,6 +6064,7 @@ struct ggml_tensor * ggml_view_2d(
|
|||
const int64_t ne[GGML_MAX_DIMS] = { ne0, ne1, 1, 1 };
|
||||
|
||||
struct ggml_tensor * result = ggml_new_tensor_impl(ctx, a->type, 2, ne, (char *) a->data + offset);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
result->nb[1] = nb1;
|
||||
result->nb[2] = result->nb[1]*ne1;
|
||||
|
|
@ -6039,6 +6096,7 @@ struct ggml_tensor * ggml_view_3d(
|
|||
const int64_t ne[GGML_MAX_DIMS] = { ne0, ne1, ne2, 1 };
|
||||
|
||||
struct ggml_tensor * result = ggml_new_tensor_impl(ctx, a->type, 3, ne, (char *) a->data + offset);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
result->nb[1] = nb1;
|
||||
result->nb[2] = nb2;
|
||||
|
|
@ -6081,6 +6139,7 @@ struct ggml_tensor * ggml_permute(
|
|||
}
|
||||
|
||||
struct ggml_tensor * result = ggml_view_tensor(ctx, a);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
int ne[GGML_MAX_DIMS];
|
||||
int nb[GGML_MAX_DIMS];
|
||||
|
|
@ -6159,6 +6218,7 @@ struct ggml_tensor * ggml_get_rows(
|
|||
// TODO: implement non F32 return
|
||||
//struct ggml_tensor * result = ggml_new_tensor_2d(ctx, a->type, a->ne[0], b->ne[0]);
|
||||
struct ggml_tensor * result = ggml_new_tensor_2d(ctx, GGML_TYPE_F32, a->ne[0], b->ne[0]);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
result->op = GGML_OP_GET_ROWS;
|
||||
result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
|
||||
|
|
@ -6184,7 +6244,9 @@ struct ggml_tensor * ggml_diag_mask_inf(
|
|||
// TODO: when implement backward, fix this:
|
||||
//struct ggml_tensor * result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
|
||||
struct ggml_tensor * result = ggml_view_tensor(ctx, a);
|
||||
if (result == NULL) return NULL;
|
||||
struct ggml_tensor * b = ggml_new_i32(ctx, n_past);
|
||||
GGML_ASSERT(b != NULL);
|
||||
ggml_set_name(b, "n_past");
|
||||
|
||||
result->op = GGML_OP_DIAG_MASK_INF;
|
||||
|
|
@ -6210,6 +6272,7 @@ struct ggml_tensor * ggml_soft_max(
|
|||
// TODO: when implement backward, fix this:
|
||||
//struct ggml_tensor * result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
|
||||
struct ggml_tensor * result = ggml_view_tensor(ctx, a);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
result->op = GGML_OP_SOFT_MAX;
|
||||
result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
|
||||
|
|
@ -6238,8 +6301,9 @@ struct ggml_tensor * ggml_rope(
|
|||
// TODO: when implement backward, fix this:
|
||||
//struct ggml_tensor * result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
|
||||
struct ggml_tensor * result = ggml_view_tensor(ctx, a);
|
||||
|
||||
if (result == NULL) return NULL;
|
||||
struct ggml_tensor * b = ggml_new_tensor_1d(ctx, GGML_TYPE_I32, 3);
|
||||
GGML_ASSERT(b != NULL);
|
||||
((int32_t *) b->data)[0] = n_past;
|
||||
((int32_t *) b->data)[1] = n_dims;
|
||||
((int32_t *) b->data)[2] = mode;
|
||||
|
|
@ -6271,8 +6335,10 @@ struct ggml_tensor * ggml_alibi(
|
|||
// TODO: when implement backward, fix this:
|
||||
//struct ggml_tensor * result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
|
||||
struct ggml_tensor * result = ggml_view_tensor(ctx, a);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
struct ggml_tensor * b = ggml_new_tensor_1d(ctx, GGML_TYPE_I32, 2);
|
||||
GGML_ASSERT(b != NULL);
|
||||
((int32_t *) b->data)[0] = n_past;
|
||||
((int32_t *) b->data)[1] = n_head;
|
||||
|
||||
|
|
@ -6302,6 +6368,7 @@ struct ggml_tensor * ggml_conv_1d_1s(
|
|||
|
||||
const int64_t ne[4] = { b->ne[0], a->ne[2], 1, 1, };
|
||||
struct ggml_tensor * result = ggml_new_tensor(ctx, GGML_TYPE_F32, 2, ne);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
result->op = GGML_OP_CONV_1D_1S;
|
||||
result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
|
||||
|
|
@ -6329,6 +6396,7 @@ struct ggml_tensor * ggml_conv_1d_2s(
|
|||
|
||||
const int64_t ne[4] = { b->ne[0]/2, a->ne[2], 1, 1, };
|
||||
struct ggml_tensor * result = ggml_new_tensor(ctx, GGML_TYPE_F32, 2, ne);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
result->op = GGML_OP_CONV_1D_2S;
|
||||
result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
|
||||
|
|
@ -6358,6 +6426,7 @@ struct ggml_tensor * ggml_flash_attn(
|
|||
|
||||
//struct ggml_tensor * result = ggml_dup_tensor(ctx, q);
|
||||
struct ggml_tensor * result = ggml_new_tensor(ctx, GGML_TYPE_F32, 4, q->ne);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
result->op = GGML_OP_FLASH_ATTN;
|
||||
result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
|
||||
|
|
@ -6365,6 +6434,7 @@ struct ggml_tensor * ggml_flash_attn(
|
|||
result->src1 = k;
|
||||
result->opt[0] = v;
|
||||
result->opt[1] = ggml_new_i32(ctx, masked ? 1 : 0);
|
||||
GGML_ASSERT(result->opt[1] != NULL);
|
||||
|
||||
return result;
|
||||
}
|
||||
|
|
@ -6390,6 +6460,7 @@ struct ggml_tensor * ggml_flash_ff(
|
|||
|
||||
//struct ggml_tensor * result = ggml_dup_tensor(ctx, a);
|
||||
struct ggml_tensor * result = ggml_new_tensor(ctx, GGML_TYPE_F32, 4, a->ne);
|
||||
if (result == NULL) return NULL;
|
||||
|
||||
result->op = GGML_OP_FLASH_FF;
|
||||
result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
|
||||
|
|
@ -6416,8 +6487,10 @@ struct ggml_tensor * ggml_map_unary_impl_f32(
|
|||
}
|
||||
|
||||
struct ggml_tensor * addr_tensor = ggml_new_tensor_1d(ctx, GGML_TYPE_I32, sizeof(void *) / sizeof(int32_t));
|
||||
if (addr_tensor == NULL) return NULL;
|
||||
*((void (**)(void))addr_tensor->data) = (void (*)(void))fun;
|
||||
struct ggml_tensor *result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
|
||||
GGML_ASSERT(result != NULL);
|
||||
|
||||
result->op = GGML_OP_MAP_UNARY;
|
||||
result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
|
||||
|
|
@ -6458,8 +6531,10 @@ struct ggml_tensor * ggml_map_binary_impl_f32(
|
|||
}
|
||||
|
||||
struct ggml_tensor * addr_tensor = ggml_new_tensor_1d(ctx, GGML_TYPE_I32, sizeof(void *) / sizeof(int32_t));
|
||||
if (addr_tensor == NULL) return NULL;
|
||||
*((void (**)(void))addr_tensor->data) = (void (*)(void))fun;
|
||||
struct ggml_tensor *result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
|
||||
GGML_ASSERT(result != NULL);
|
||||
|
||||
result->op = GGML_OP_MAP_BINARY;
|
||||
result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
|
||||
|
|
@ -11741,6 +11816,9 @@ static thread_ret_t ggml_graph_compute_thread(void * data) {
|
|||
}
|
||||
|
||||
void ggml_graph_compute(struct ggml_context * ctx, struct ggml_cgraph * cgraph) {
|
||||
#ifdef GGML_RECOVERABLE_ERRORS
|
||||
GGML_ASSERT(ctx->last_error_code == GGML_ERRCODE_SUCCESS);
|
||||
#endif
|
||||
const int n_threads = cgraph->n_threads;
|
||||
|
||||
struct ggml_compute_state_shared state_shared = {
|
||||
|
|
|
|||
48
ggml.h
48
ggml.h
|
|
@ -205,6 +205,45 @@
|
|||
} \
|
||||
} while (0)
|
||||
|
||||
#define GGML_RECOVERABLE_ERRORS
|
||||
#ifndef GGML_RECOVERABLE_ERRORS
|
||||
#define GGML_RECOVERABLE_ASSERT(_ctx, _code, x, ...) \
|
||||
do { \
|
||||
if (!(x)) { \
|
||||
printf(__VA_ARGS__); \
|
||||
fprintf(stderr, "GGML_ASSERT: %s:%d: %s\n", __FILE__, __LINE__, #x); \
|
||||
abort(); \
|
||||
} \
|
||||
} while (0)
|
||||
|
||||
#else
|
||||
#define GGML_ERROR_BUFFER_LEN 256
|
||||
#define GGML_RECOVERABLE_ASSERT(ctx, errcode, x, ...) \
|
||||
do { \
|
||||
if ((ctx)->last_error_code != GGML_ERRCODE_SUCCESS) { \
|
||||
fprintf(stderr, "GGML_RECOVERABLE_ASSERT: Cannot continue due previous error condition: %s\n", (ctx)->last_error_msg); \
|
||||
GGML_ASSERT(false); \
|
||||
} else if (!(x)) { \
|
||||
snprintf((ctx)->last_error_msg, GGML_ERROR_BUFFER_LEN - 1, __VA_ARGS__); \
|
||||
(ctx)->last_error_msg[GGML_ERROR_BUFFER_LEN - 1] = 0; \
|
||||
(ctx)->last_error_code = errcode; \
|
||||
return NULL; \
|
||||
} \
|
||||
} while (0)
|
||||
|
||||
|
||||
#endif
|
||||
|
||||
enum ggml_errcode {
|
||||
GGML_ERRCODE_SUCCESS = 0,
|
||||
GGML_ERRCODE_OTHER,
|
||||
GGML_ERRCODE_CTX_MEMORY_LIMIT,
|
||||
GGML_ERRCODE_SCRATCH_MEMORY_LIMIT,
|
||||
GGML_ERRCODE_BAD_TYPE,
|
||||
GGML_ERRCODE_BAD_DIMENSIONS,
|
||||
GGML_ERRCODE_BAD_SHAPE,
|
||||
};
|
||||
|
||||
#ifdef __cplusplus
|
||||
extern "C" {
|
||||
#endif
|
||||
|
|
@ -425,6 +464,15 @@ extern "C" {
|
|||
|
||||
GGML_API size_t ggml_set_scratch(struct ggml_context * ctx, struct ggml_scratch scratch);
|
||||
|
||||
#ifdef GGML_RECOVERABLE_ERRORS
|
||||
// GGML_API enum ggml_errcode ggml_clear_error(
|
||||
// struct ggml_context * ctx);
|
||||
GGML_API enum ggml_errcode ggml_last_error_code(
|
||||
struct ggml_context * ctx);
|
||||
GGML_API char * ggml_last_error_msg(
|
||||
struct ggml_context * ctx);
|
||||
#endif
|
||||
|
||||
GGML_API struct ggml_tensor * ggml_new_tensor(
|
||||
struct ggml_context * ctx,
|
||||
enum ggml_type type,
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue