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

Removing MIRROR_MODE code for this PR

Browse source
This commit is contained in:
root 2024-02-07 21:46:19 +00:00
parent 3eccea1b63
commit 61c37ba93c
3 changed files with 0 additions and 17 deletions
Download patch file Download diff file Expand all files Collapse all files

6
common/common.cpp
View file

@ -662,9 +662,6 @@ bool gpt_params_parse_ex(int argc, char ** argv, gpt_params & params) {
/**/ if (value == "interleave" || value == "" ) { params.numa = GGML_NUMA_STRATEGY_INTERLEAVE; } /**/ if (value == "interleave" || value == "" ) { params.numa = GGML_NUMA_STRATEGY_INTERLEAVE; }
else if (value == "isolate") { params.numa = GGML_NUMA_STRATEGY_ISOLATE; } else if (value == "isolate") { params.numa = GGML_NUMA_STRATEGY_ISOLATE; }
else if (value == "numactl") { params.numa = GGML_NUMA_STRATEGY_NUMACTL; } else if (value == "numactl") { params.numa = GGML_NUMA_STRATEGY_NUMACTL; }
#ifdef GGUF_NUMA_MIRROR
else if (value == "mirror") { params.numa = GGML_NUMA_STRATEGY_MIRROR; }
#endif
else { invalid_param = true; break; } else { invalid_param = true; break; }
} }
} else if (arg == "--verbose-prompt") { } else if (arg == "--verbose-prompt") {
@ -1011,9 +1008,6 @@ void gpt_print_usage(int /*argc*/, char ** argv, const gpt_params & params) {
printf(" - interleave: (default) spread execution evenly over all nodes\n"); printf(" - interleave: (default) spread execution evenly over all nodes\n");
printf(" - isolate: only spawn threads on CPUs on the node that execution started on\n"); printf(" - isolate: only spawn threads on CPUs on the node that execution started on\n");
printf(" - numactl: use the CPU map provided my numactl\n"); printf(" - numactl: use the CPU map provided my numactl\n");
#ifdef GGML_NUMA_MIRROR
printf(" - mirror: NOT YET IMPLEMENTED - attempt to mirror GGUF data buffer on each node's local memory to increase throughput.\n");
#endif
printf(" if run without this previously, it is recommended to drop the system page cache before using this\n"); printf(" if run without this previously, it is recommended to drop the system page cache before using this\n");
printf(" see https://github.com/ggerganov/llama.cpp/issues/1437\n"); printf(" see https://github.com/ggerganov/llama.cpp/issues/1437\n");
if (llama_supports_gpu_offload()) { if (llama_supports_gpu_offload()) {

3
examples/server/server.cpp
View file

@ -1825,9 +1825,6 @@ static void server_print_usage(const char *argv0, const gpt_params &params,
printf(" - interleave: (default) spread execution evenly over all nodes\n"); printf(" - interleave: (default) spread execution evenly over all nodes\n");
printf(" - isolate: only spawn threads on CPUs on the node that execution started on\n"); printf(" - isolate: only spawn threads on CPUs on the node that execution started on\n");
printf(" - numactl: use the CPU map provided my numactl\n"); printf(" - numactl: use the CPU map provided my numactl\n");
#ifdef GGML_NUMA_MIRROR
printf(" - mirror: NOT YET IMPLEMENTED - attempt to mirror GGUF data buffer on each node's local memory to increase throughput.\n");
#endif
if (llama_supports_gpu_offload()) { if (llama_supports_gpu_offload()) {
printf(" -ngl N, --n-gpu-layers N\n"); printf(" -ngl N, --n-gpu-layers N\n");
printf(" number of layers to store in VRAM\n"); printf(" number of layers to store in VRAM\n");

8
ggml.c
View file

@ -24,10 +24,6 @@
#include <stdarg.h> #include <stdarg.h>
#include <signal.h> #include <signal.h>
#ifdef GGML_NUMA_MIRROR
#include <numa.h>
#endif
#ifdef GGML_USE_METAL #ifdef GGML_USE_METAL
#include <unistd.h> #include <unistd.h>
#endif #endif
@ -16635,10 +16631,6 @@ static void set_numa_thread_affinity(int thread_n, int n_threads) {
fprintf(stderr, "warning: pthread_setaffinity_np() failed: %s\n",strerror(rv)); fprintf(stderr, "warning: pthread_setaffinity_np() failed: %s\n",strerror(rv));
} }
return; return;
#ifdef GGML_NUMA_MIRROR
case GGML_NUMA_STRATEGY_MIRROR:
printf("Mirror Mode Enabled");
#endif
default: default:
return; return;
} }