vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

sync : ggml (backend v2) (#3912)

Browse source 
* sync : ggml (backend v2) (wip)

* sync : migrate examples and llama.cpp to dynamic graphs (wip)

* sync : update tests + fix max op params to 64

ggml-ci

* sync : ggml-cuda

ggml-ci

* llama : fix save/load state context size

ggml-ci

* sync : try to fix build on tvOS

* sync : pass custom graph sizes in training examples

* sync : update graph copies to new ggml API

* sync : update sync-ggml.sh with new files

* scripts : fix header in sync script

* train : fix context size calculations

* llama : increase inference graph size up to 4096 nodes

* train : allocate grads for backward graphs

* train : allocate grads for gb_tmp
This commit is contained in:
Georgi Gerganov 2023-11-13 14:16:23 +02:00 • committed by GitHub
parent bb50a792ec
commit 4760e7cc0b
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
22 changed files with 1994 additions and 864 deletions
Download patch file Download diff file Expand all files Collapse all files

25
ggml-metal.m
View file

@ -1,5 +1,6 @@
#import "ggml-metal.h"
#import "ggml-backend-impl.h"
#import "ggml.h"
#import <Foundation/Foundation.h>
@ -23,7 +24,7 @@
#define UNUSED(x) (void)(x)
#define GGML_MAX_CONCUR (2*GGML_MAX_NODES)
#define GGML_MAX_CONCUR (2*GGML_DEFAULT_GRAPH_SIZE)
struct ggml_metal_buffer {
const char * name;
@ -744,6 +745,20 @@ void ggml_metal_graph_compute(
struct ggml_tensor * src1 = gf->nodes[i]->src[1];
struct ggml_tensor * dst = gf->nodes[i];
switch (dst->op) {
case GGML_OP_NONE:
case GGML_OP_RESHAPE:
case GGML_OP_VIEW:
case GGML_OP_TRANSPOSE:
case GGML_OP_PERMUTE:
{
// noop -> next node
} continue;
default:
{
} break;
}
const int64_t ne00 = src0 ? src0->ne[0] : 0;
const int64_t ne01 = src0 ? src0->ne[1] : 0;
const int64_t ne02 = src0 ? src0->ne[2] : 0;
@ -797,14 +812,6 @@ void ggml_metal_graph_compute(
//}
switch (dst->op) {
case GGML_OP_NONE:
case GGML_OP_RESHAPE:
case GGML_OP_VIEW:
case GGML_OP_TRANSPOSE:
case GGML_OP_PERMUTE:
{
// noop
} break;
case GGML_OP_CONCAT:
{
const int64_t nb = ne00;