Mirror/LocalAI
1
0
Fork 0
mirror of https://github.com/mudler/LocalAI.git synced 2024-06-07 19:40:48 +00:00
Code Issues Packages Projects Releases Wiki Activity

feat(llama.cpp): enable ROCm/HIPBLAS support (#1100)

Browse Source 
**Description**

This PR fixes lack of HIPBLAS support in LocalAI.

**Notes for Reviewers**
This PR builds on https://github.com/go-skynet/go-llama.cpp/pull/235 to
enable ROCm/HIPBLAS support for gguf models running under llama.cpp
backend (not the stable ggml one). It can be enabled by using
BUILD_TYPE=hipblas. This was tested on a gfx1100 card, but should work
for gfx900,gfx1030 and other cards. Card support can be set with
AMDGPU_TARGETS environment variable.

**[Signed
commits](../CONTRIBUTING.md#signing-off-on-commits-developer-certificate-of-origin)**
- [x] Yes, I signed my commits.
 
<!--
Thank you for contributing to LocalAI! 

Contributing Conventions
-------------------------

The draft above helps to give a quick overview of your PR.

Remember to remove this comment and to at least:

1. Include descriptive PR titles with [<component-name>] prepended. We
use [conventional
commits](https://www.conventionalcommits.org/en/v1.0.0/).
2. Build and test your changes before submitting a PR (`make build`). 
3. Sign your commits
4. **Tag maintainer:** for a quicker response, tag the relevant
maintainer (see below).
5. **X/Twitter handle:** we announce bigger features on X/Twitter. If
your PR gets announced, and you'd like a mention, we'll gladly shout you
out!

By following the community's contribution conventions upfront, the
review process will
be accelerated and your PR merged more quickly.

If no one reviews your PR within a few days, please @-mention @mudler.
-->

---------

Signed-off-by: 65a <65a@63bit.net>
This commit is contained in:
65a 2023-09-28 12:42:20 -07:00 committed by GitHub
parent 274ace2898
commit 55e38fea0e
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
3 changed files with 18 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
Makefile
View File

@ -4,7 +4,7 @@ GOVET=$(GOCMD) vet
BINARY_NAME=local-ai
# llama.cpp versions
GOLLAMA_VERSION?=d9f6176409de0a2b5ce798de502545c6721e346e
GOLLAMA_VERSION?=d84f03c81a5e336c0146ec9c2ef1c96cb4fe7f09
GOLLAMA_STABLE_VERSION?=50cee7712066d9e38306eccadcfbb44ea87df4b7
@ -38,6 +38,7 @@ STABLEDIFFUSION_VERSION?=d89260f598afb809279bc72aa0107b4292587632
GOGGLLM_VERSION?=862477d16eefb0805261c19c9b0d053e3b2b684b
export BUILD_TYPE?=
export STABLE_BUILD_TYPE?=$(BUILD_TYPE)
export CMAKE_ARGS?=
CGO_LDFLAGS?=
CUDA_LIBPATH?=/usr/local/cuda/lib64/
@ -82,6 +83,18 @@ ifeq ($(BUILD_TYPE),cublas)
export LLAMA_CUBLAS=1
endif
ifeq ($(BUILD_TYPE),hipblas)
ROCM_HOME ?= /opt/rocm
export CXX=$(ROCM_HOME)/llvm/bin/clang++
export CC=$(ROCM_HOME)/llvm/bin/clang
# Llama-stable has no hipblas support, so override it here.
export STABLE_BUILD_TYPE=
GPU_TARGETS ?= gfx900,gfx90a,gfx1030,gfx1031,gfx1100
AMDGPU_TARGETS ?= "$(GPU_TARGETS)"
CMAKE_ARGS+=-DLLAMA_HIPBLAS=ON -DAMDGPU_TARGETS="$(AMDGPU_TARGETS)" -DGPU_TARGETS="$(GPU_TARGETS)"
CGO_LDFLAGS += -O3 --rtlib=compiler-rt -unwindlib=libgcc -lhipblas -lrocblas --hip-link
endif
ifeq ($(BUILD_TYPE),metal)
CGO_LDFLAGS+=-framework Foundation -framework Metal -framework MetalKit -framework MetalPerformanceShaders
export LLAMA_METAL=1
@ -208,7 +221,7 @@ go-llama/libbinding.a: go-llama
$(MAKE) -C go-llama BUILD_TYPE=$(BUILD_TYPE) libbinding.a
go-llama-stable/libbinding.a: go-llama-stable
$(MAKE) -C go-llama-stable BUILD_TYPE=$(BUILD_TYPE) libbinding.a
$(MAKE) -C go-llama-stable BUILD_TYPE=$(STABLE_BUILD_TYPE) libbinding.a
go-piper/libpiper_binding.a:
$(MAKE) -C go-piper libpiper_binding.a example/main

2
go.mod
View File

@ -9,7 +9,7 @@ require (
github.com/go-skynet/bloomz.cpp v0.0.0-20230529155654-1834e77b83fa
github.com/go-skynet/go-bert.cpp v0.0.0-20230716133540-6abe312cded1
github.com/go-skynet/go-ggml-transformers.cpp v0.0.0-20230714203132-ffb09d7dd71e
github.com/go-skynet/go-llama.cpp v0.0.0-20230916071220-b471eb7d8c93
github.com/go-skynet/go-llama.cpp v0.0.0-20230923221031-d84f03c81a5e
github.com/gofiber/fiber/v2 v2.49.2
github.com/google/uuid v1.3.1
github.com/hashicorp/go-multierror v1.1.1

2
go.sum
View File

@ -41,6 +41,8 @@ github.com/go-skynet/go-ggml-transformers.cpp v0.0.0-20230714203132-ffb09d7dd71e
github.com/go-skynet/go-ggml-transformers.cpp v0.0.0-20230714203132-ffb09d7dd71e/go.mod h1:31j1odgFXP8hDSUVfH0zErKI5aYVP18ddYnPkwCso2A=
github.com/go-skynet/go-llama.cpp v0.0.0-20230916071220-b471eb7d8c93 h1:aJQn0Lb5TQtSXw5hK+kAUuaUjsXwDXyEItzI9U+GiOA=
github.com/go-skynet/go-llama.cpp v0.0.0-20230916071220-b471eb7d8c93/go.mod h1:VGPpMv2UMxRgBiDIFrs7AGMzrX884K0Vy8tqYu3if/Q=
github.com/go-skynet/go-llama.cpp v0.0.0-20230923221031-d84f03c81a5e h1:ubROx+GEi6onpAeQSi8L5n7lHzyMTZs7vq/m7wiUn7Y=
github.com/go-skynet/go-llama.cpp v0.0.0-20230923221031-d84f03c81a5e/go.mod h1:uejmv9IhRLb9FMhksGXgjAhvJpFnoTOX4+Dr9Lb0HSg=
github.com/go-task/slim-sprig v0.0.0-20210107165309-348f09dbbbc0/go.mod h1:fyg7847qk6SyHyPtNmDHnmrv/HOrqktSC+C9fM+CJOE=
github.com/go-task/slim-sprig v0.0.0-20230315185526-52ccab3ef572 h1:tfuBGBXKqDEevZMzYi5KSi8KkcZtzBcTgAUUtapy0OI=
github.com/go-task/slim-sprig v0.0.0-20230315185526-52ccab3ef572/go.mod h1:9Pwr4B2jHnOSGXyyzV8ROjYa2ojvAY6HCGYYfMoC3Ls=