update

768ab4df · xuxzh1 · 44f79c7e · 768ab4df · 768ab4df · 768ab4df
Commit 768ab4df authored Nov 13, 2024 by xuxzh1 🎱
20 changed files
--- a/llm/llama.cpp/ggml/src/ggml-impl.h
+++ b/llm/llama.cpp/ggml/src/ggml-impl.h
--- a/llm/llama.cpp/ggml/src/ggml-kompute.cpp
+++ b/llm/llama.cpp/ggml/src/ggml-kompute.cpp
--- a/llm/llama.cpp/ggml/src/ggml-metal.m
+++ b/llm/llama.cpp/ggml/src/ggml-metal.m
@@ -1545,27 +1545,23 @@ static enum ggml_status ggml_metal_graph_compute(
                        // to the matrix-vector kernel
                        int ne11_mm_min = 1;

-#if 0
                        // the numbers below are measured on M2 Ultra for 7B and 13B models
                        // these numbers do not translate to other devices or model sizes
                        // TODO: need to find a better approach
-                        if ([ctx->device.name isEqualToString:@"Apple M2 Ultra"]) {
-                            switch (src0t) {
-                                case GGML_TYPE_F16:  ne11_mm_min = 2;  break;
-                                case GGML_TYPE_Q8_0: ne11_mm_min = 7;  break;
-                                case GGML_TYPE_Q2_K: ne11_mm_min = 15; break;
-                                case GGML_TYPE_Q3_K: ne11_mm_min = 7;  break;
-                                case GGML_TYPE_Q4_0:
-                                case GGML_TYPE_Q4_1: ne11_mm_min = 15; break;
-                                case GGML_TYPE_Q4_K: ne11_mm_min = 11; break;
-                                case GGML_TYPE_Q5_0:                          // not tested yet
-                                case GGML_TYPE_Q5_1: ne11_mm_min = 13; break; // not tested yet
-                                case GGML_TYPE_Q5_K: ne11_mm_min = 7;  break;
-                                case GGML_TYPE_Q6_K: ne11_mm_min = 7;  break;
-                                default:             ne11_mm_min = 1;  break;
-                            }
+                        switch (src0t) {
+                            case GGML_TYPE_F16:  ne11_mm_min = 2;  break;
+                            case GGML_TYPE_Q8_0: ne11_mm_min = 7;  break;
+                            case GGML_TYPE_Q2_K: ne11_mm_min = 15; break;
+                            case GGML_TYPE_Q3_K: ne11_mm_min = 7;  break;
+                            case GGML_TYPE_Q4_0:
+                            case GGML_TYPE_Q4_1: ne11_mm_min = 15; break;
+                            case GGML_TYPE_Q4_K: ne11_mm_min = 11; break;
+                            case GGML_TYPE_Q5_0:                          // not tested yet
+                            case GGML_TYPE_Q5_1: ne11_mm_min = 13; break; // not tested yet
+                            case GGML_TYPE_Q5_K: ne11_mm_min = 7;  break;
+                            case GGML_TYPE_Q6_K: ne11_mm_min = 7;  break;
+                            default:             ne11_mm_min = 1;  break;
                        }
-#endif

                        // for now the matrix-matrix multiplication kernel only works on A14+/M1+ SoCs
                        // AMD GPU and older A-chips will reuse matrix-vector multiplication kernel

--- a/llm/llama.cpp/ggml/src/ggml-quants.c
+++ b/llm/llama.cpp/ggml/src/ggml-quants.c
--- a/llm/llama.cpp/ggml/src/ggml-quants.h
+++ b/llm/llama.cpp/ggml/src/ggml-quants.h
--- a/llm/llama.cpp/ggml/src/ggml-rpc.cpp
+++ b/llm/llama.cpp/ggml/src/ggml-rpc.cpp
--- a/llm/llama.cpp/ggml/src/ggml-sycl.cpp
+++ b/llm/llama.cpp/ggml/src/ggml-sycl.cpp
--- a/llm/llama.cpp/ggml/src/ggml-sycl/backend.hpp
+++ b/llm/llama.cpp/ggml/src/ggml-sycl/backend.hpp
--- a/llm/llama.cpp/ggml/src/ggml-sycl/common.cpp
+++ b/llm/llama.cpp/ggml/src/ggml-sycl/common.cpp
--- a/llm/llama.cpp/ggml/src/ggml-sycl/common.hpp
+++ b/llm/llama.cpp/ggml/src/ggml-sycl/common.hpp
--- a/llm/llama.cpp/ggml/src/ggml-sycl/concat.cpp
+++ b/llm/llama.cpp/ggml/src/ggml-sycl/concat.cpp
--- a/llm/llama.cpp/ggml/src/ggml-sycl/concat.hpp
+++ b/llm/llama.cpp/ggml/src/ggml-sycl/concat.hpp
--- a/llm/llama.cpp/ggml/src/ggml-sycl/conv.cpp
+++ b/llm/llama.cpp/ggml/src/ggml-sycl/conv.cpp
--- a/llm/llama.cpp/ggml/src/ggml-sycl/conv.hpp
+++ b/llm/llama.cpp/ggml/src/ggml-sycl/conv.hpp
--- a/llm/llama.cpp/ggml/src/ggml-sycl/convert.cpp
+++ b/llm/llama.cpp/ggml/src/ggml-sycl/convert.cpp
--- a/llm/llama.cpp/ggml/src/ggml-sycl/convert.hpp
+++ b/llm/llama.cpp/ggml/src/ggml-sycl/convert.hpp
--- a/llm/llama.cpp/ggml/src/ggml-sycl/dequantize.hpp
+++ b/llm/llama.cpp/ggml/src/ggml-sycl/dequantize.hpp
--- a/llm/llama.cpp/ggml/src/ggml-sycl/dmmv.cpp
+++ b/llm/llama.cpp/ggml/src/ggml-sycl/dmmv.cpp
--- a/llm/llama.cpp/ggml/src/ggml-sycl/dmmv.hpp
+++ b/llm/llama.cpp/ggml/src/ggml-sycl/dmmv.hpp
--- a/llm/llama.cpp/ggml/src/ggml-sycl/dpct/helper.hpp
+++ b/llm/llama.cpp/ggml/src/ggml-sycl/dpct/helper.hpp