baalajimaestro/ollama
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

fix ggml arm64 cuda build (#520)

Browse source
This commit is contained in:
Bruce MacDonald 2023-09-12 17:06:48 -04:00 • committed by GitHub
parent f221637053
commit f59c4d03f7
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
2 changed files with 35 additions and 2 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

5
llm/llama.cpp/generate_linux.go
View file

@ -6,9 +6,10 @@ package llm
//go:generate -command git-apply git -C ggml apply
//go:generate git-apply ../ggml_patch/0001-add-detokenize-endpoint.patch
//go:generate git-apply ../ggml_patch/0002-34B-model-support.patch
//go:generate cmake -S ggml -B ggml/build/gpu -DLLAMA_CUBLAS=on -DLLAMA_ACCELERATE=on -DLLAMA_K_QUANTS=on
//go:generate git-apply ../ggml_patch/0005-ggml-support-CUDA-s-half-type-for-aarch64-1455-2670.patch
//go:generate cmake -S ggml -B ggml/build/gpu -DLLAMA_CUBLAS=on -DLLAMA_ACCELERATE=on -DLLAMA_K_QUANTS=on
//go:generate cmake --build ggml/build/gpu --target server --config Release
//go:generate git submodule update --force gguf
//go:generate cmake -S gguf -B gguf/build/gpu -DLLAMA_CUBLAS=on -DLLAMA_ACCELERATE=on -DLLAMA_K_QUANTS=on
//go:generate cmake -S gguf -B gguf/build/gpu -DLLAMA_CUBLAS=on -DLLAMA_ACCELERATE=on -DLLAMA_K_QUANTS=on
//go:generate cmake --build gguf/build/gpu --target server --config Release

32
llm/llama.cpp/ggml_patch/0005-ggml-support-CUDA-s-half-type-for-aarch64-1455-2670.patch Normal file
View file

@ -0,0 +1,32 @@
From 1e3bc523d8053a77df3ac7126a84d0297ee97ef6 Mon Sep 17 00:00:00 2001
From: Kylin <56434533+KyL0N@users.noreply.github.com>
Date: Tue, 22 Aug 2023 15:14:23 +0800
Subject: [PATCH] ggml : support CUDA's half type for aarch64(#1455) (#2670)
* ggml: support CUDA's half type for aarch64(#1455)
support CUDA's half type for aarch64 in ggml_fp16_t definition
* ggml: use __CUDACC__ to recognise nvcc compiler
---
ggml.h | 5 +++--
1 file changed, 3 insertions(+), 2 deletions(-)
diff --git a/ggml.h b/ggml.h
index 544ad2d..0ec7ec5 100644
--- a/ggml.h
+++ b/ggml.h
@@ -259,8 +259,9 @@
extern "C" {
#endif
-#ifdef __ARM_NEON
- // we use the built-in 16-bit float type
+#if defined(__ARM_NEON) && defined(__CUDACC__)
+ typedef half ggml_fp16_t;
+#elif defined(__ARM_NEON)
typedef __fp16 ggml_fp16_t;
#else
typedef uint16_t ggml_fp16_t;
--
2.39.2 (Apple Git-143)