Updated llama.cpp, generated bindings, and removed MacOS/CLBlast refe…

…rence from the documentation.
nightscape · Jun 18, 2023 · abcf77b · abcf77b
1 parent b401251
commit abcf77b
Show file tree

Hide file tree

Showing 4 changed files with 8 additions and 4 deletions.
diff --git a/crates/ggml/sys/llama-cpp b/crates/ggml/sys/llama-cpp
diff --git a/crates/ggml/sys/src/lib.rs b/crates/ggml/sys/src/lib.rs
@@ -840,10 +840,13 @@ extern "C" {
     pub fn ggml_set_no_alloc(ctx: *mut ggml_context, no_alloc: bool);
 }
 extern "C" {
-    pub fn ggml_get_mem_buffer(ctx: *mut ggml_context) -> *mut ::std::os::raw::c_void;
+    pub fn ggml_get_mem_buffer(ctx: *const ggml_context) -> *mut ::std::os::raw::c_void;
 }
 extern "C" {
-    pub fn ggml_get_mem_size(ctx: *mut ggml_context) -> usize;
+    pub fn ggml_get_mem_size(ctx: *const ggml_context) -> usize;
+}
+extern "C" {
+    pub fn ggml_get_max_tensor_size(ctx: *const ggml_context) -> usize;
 }
 extern "C" {
     pub fn ggml_new_tensor(

diff --git a/crates/ggml/sys/src/metal.rs b/crates/ggml/sys/src/metal.rs
@@ -28,6 +28,7 @@ extern "C" {
         name: *const ::std::os::raw::c_char,
         data: *mut ::std::os::raw::c_void,
         size: usize,
+        max_size: usize,
     ) -> bool;
 }
 extern "C" {

diff --git a/doc/CONTRIBUTING.md b/doc/CONTRIBUTING.md
@@ -32,7 +32,7 @@ cargo run --release --package generate-ggml-bindings
 
 ## Acceleration Support for Building
 
-The `ggml-sys` crate includes various acceleration backends, selectable via `--features` flags. The availability of supported backends varies by platform, and `ggml-sys` can only be built with a single active acceleration backend at a time. If cublas and clblast are both specified, cublas is prioritized and clblast is ignored. On MacOS clblast is prioritized over metal.
+The `ggml-sys` crate includes various acceleration backends, selectable via `--features` flags. The availability of supported backends varies by platform, and `ggml-sys` can only be built with a single active acceleration backend at a time. If cublas and clblast are both specified, cublas is prioritized and clblast is ignored. 
 
 | Platform/OS | `cublas`           | `clblast`          | `metal`            |
 | ----------- | ------------------ | ------------------ | ------------------ |
+2 −0		.flake8
+2 −0		.gitignore
+15 −0		.pre-commit-config.yaml
+54 −10		CMakeLists.txt
+13 −7		Makefile
+41 −1		README.md
+18 −8		convert.py
+1 −0		examples/CMakeLists.txt
+5 −1		examples/baby-llama/baby-llama.cpp
+7 −3		examples/benchmark/benchmark-matmult.cpp
+5 −4		examples/common.cpp
+4 −0		examples/embedding/embedding.cpp
+4 −3		examples/jeopardy/graph.py
+7 −1		examples/main/main.cpp
+4 −2		examples/metal/metal.cpp
+4 −0		examples/perplexity/perplexity.cpp
+4 −0		examples/quantize-stats/quantize-stats.cpp
+1 −1		examples/save-load-state/save-load-state.cpp
+4 −0		examples/server/CMakeLists.txt
+91 −227		examples/server/README.md
+89 −0		examples/server/chat.mjs
+77 −0		examples/server/chat.sh
+855 −726		examples/server/server.cpp
+7 −0		examples/simple/CMakeLists.txt
+177 −0		examples/simple/simple.cpp
+1 −1		examples/train-text-from-scratch/README.md
+10 −8		examples/train-text-from-scratch/train-text-from-scratch.cpp
+13 −0		flake.nix
+402 −255		ggml-cuda.cu
+4 −1		ggml-metal.h
+157 −19		ggml-metal.m
+149 −0		ggml-metal.metal
+493 −4		ggml-opencl.cpp
+28 −2		ggml.c
+3 −2		ggml.h
+40 −12		llama.cpp
+4 −0		pocs/vdot/vdot.cpp
+3 −1		scripts/verify-checksum-models.py
+9 −6		tests/test-quantize-fns.cpp
+4 −0		tests/test-quantize-perf.cpp
+16 −16		tests/test-sampling.cpp
+1 −1		tests/test-tokenizer-0.cpp