Mac os native build not working

[glebnaz]

LocalAI version:
v2.4.1

Environment, CPU architecture, OS, and Version:
MBP 14 M1 PRO

Describe the bug
Not working make build and make BUILD_TYPE=metal build

To Reproduce
run in local make BUILD_TYPE=metal build (like in instruction)

Expected behavior
Error

Logs

CGO_LDFLAGS=" -lcblas -framework Accelerate -framework Foundation -framework Metal -framework MetalKit -framework MetalPerformanceShaders" C_INCLUDE_PATH=/Users/glebnaz/Documents/#main/workspace/LocalAI/sources/go-ggml-transformers LIBRARY_PATH=/Users/glebnaz/Documents/#main/workspace/LocalAI/sources/go-ggml-transformers \
	go build -ldflags " -X "github.com/go-skynet/LocalAI/internal.Version=v2.4.1-2-g62a02cd" -X "github.com/go-skynet/LocalAI/internal.Commit=62a02cd1feb7cf8a75bcbe253fb95f204f022c1f"" -tags "" -o backend-assets/grpc/falcon-ggml ./backend/go/llm/falcon-ggml/
backend/go/llm/transformers/dolly.go:11:2: falcon.go: malformed #cgo argument: -I/Users/glebnaz/Documents/#main/workspace/LocalAI/sources/go-ggml-transformers/ggml.cpp/include/
make: *** [backend-assets/grpc/falcon-ggml] Error 1

Additional context

[themeaningofmeaning]

Interesting, I'm getting the same error as well with multiple build attempts.

LocalAI version:
v2.5.1

Environment, CPU architecture, OS, and Version:
MBP 16 M1 PRO

I've tried using Docker to build this after cloning the repo and it shows "unable to locate package conda" even though the build instructions do specify conda as a required dependency. I also ran a variety of make prepare build attempts and it throws an error about golang not existing even though the mentioned gpt4all-bindings/golang does in fact exist and contains the necessary dependencies. Strange.

1. Docker log:

1.254 E: Unable to locate package conda
------
Dockerfile:60
--------------------
  59 |     
  60 | >>> RUN curl https://repo.anaconda.com/pkgs/misc/gpgkeys/anaconda.asc | gpg --dearmor > conda.gpg && \
  61 | >>>     install -o root -g root -m 644 conda.gpg /usr/share/keyrings/conda-archive-keyring.gpg && \
  62 | >>>     gpg --keyring /usr/share/keyrings/conda-archive-keyring.gpg --no-default-keyring --fingerprint 34161F5BF5EB1D4BFBBB8F0A8AEB4F8B29D82806 && \
  63 | >>>     echo "deb [arch=amd64 signed-by=/usr/share/keyrings/conda-archive-keyring.gpg] https://repo.anaconda.com/pkgs/misc/debrepo/conda stable main" > /etc/apt/sources.list.d/conda.list && \
  64 | >>>     echo "deb [arch=amd64 signed-by=/usr/share/keyrings/conda-archive-keyring.gpg] https://repo.anaconda.com/pkgs/misc/debrepo/conda stable main" | tee -a /etc/apt/sources.list.d/conda.list && \
  65 | >>>     apt-get update && \
  66 | >>>     apt-get install -y conda && apt-get clean
  67 |     
--------------------
ERROR: failed to solve: process "/bin/sh -c curl https://repo.anaconda.com/pkgs/misc/gpgkeys/anaconda.asc | gpg --dearmor > conda.gpg &&     install -o root -g root -m 644 conda.gpg /usr/share/keyrings/conda-archive-keyring.gpg &&     gpg --keyring /usr/share/keyrings/conda-archive-keyring.gpg --no-default-keyring --fingerprint 34161F5BF5EB1D4BFBBB8F0A8AEB4F8B29D82806 &&     echo \"deb [arch=amd64 signed-by=/usr/share/keyrings/conda-archive-keyring.gpg] https://repo.anaconda.com/pkgs/misc/debrepo/conda stable main\" > /etc/apt/sources.list.d/conda.list &&     echo \"deb [arch=amd64 signed-by=/usr/share/keyrings/conda-archive-keyring.gpg] https://repo.anaconda.com/pkgs/misc/debrepo/conda stable main\" | tee -a /etc/apt/sources.list.d/conda.list &&     apt-get update &&     apt-get install -y conda && apt-get clean" did not complete successfully: exit code: 100

View build details: docker-desktop://dashboard/build/desktop-linux/desktop-linux/zasirx96yr7cszlyhxyn4ddi4

2. make prepare log:

Switched to a new branch 'build'
touch get-sources
go mod edit -replace github.com/nomic-ai/gpt4all/gpt4all-bindings/golang=/Users/meaning/Desktop/uploading d1/LocalAI/sources/gpt4all/gpt4all-bindings/golang
go: open d1/LocalAI/sources/gpt4all/gpt4all-bindings/golang: no such file or directory
make: *** [replace] Error 1

[muellest]

@themeaningofmeaning: Unfortunately I get the same error =(. Any news on this?

LocalAI version:
v.2.3.1
v.2.4.0
v2.5.1

Environment, CPU architecture, OS, and Version:
MBP 14 M1 PRO

Logs

failed to solve: executor failed running [/bin/sh -c curl https://repo.anaconda.com/pkgs/misc/gpgkeys/anaconda.asc | gpg --dearmor > conda.gpg && install -o root -g root -m 644 conda.gpg /usr/share/keyrings/conda-archive-keyring.gpg && gpg --keyring /usr/share/keyrings/conda-archive-keyring.gpg --no-default-keyring --fingerprint 34161F5BF5EB1D4BFBBB8F0A8AEB4F8B29D82806 && echo "deb [arch=amd64 signed-by=/usr/share/keyrings/conda-archive-keyring.gpg] https://repo.anaconda.com/pkgs/misc/debrepo/conda stable main" > /etc/apt/sources.list.d/conda.list && echo "deb [arch=amd64 signed-by=/usr/share/keyrings/conda-archive-keyring.gpg] https://repo.anaconda.com/pkgs/misc/debrepo/conda stable main" | tee -a /etc/apt/sources.list.d/conda.list && apt-get update && apt-get install -y conda]: exit code: 100

[glebnaz]

@themeaningofmeaning i have the same error with docker. Should we wait help from author?

[themeaningofmeaning]

@glebnaz @muellest - I haven't been able to figure out a solution as it appears to be an issue with the release and Apple Silicon. I haven't tried installing conda or Miniconda before doing a Docker build....that would be one thing to test, but I think we need to wait to hear from the author. I've tried every build in the book and it's not working on the M1

[muellest]

Perhaps this is an alternative to using Docker: https://localai.io/basics/build/ (not tested yet).

[gaoyifan]

The following patch allows me to compile on the Apple Silicon platform. However, the error messages I encountered during compilation are different from @glebnaz , so please take them as reference only.

diff --git a/Makefile b/Makefile
index 6afc644..4414eaa 100644
--- a/Makefile
+++ b/Makefile
@@ -112,7 +112,7 @@ ifeq ($(BUILD_TYPE),hipblas)
 endif

 ifeq ($(BUILD_TYPE),metal)
-	CGO_LDFLAGS+=-framework Foundation -framework Metal -framework MetalKit -framework MetalPerformanceShaders
+	CGO_LDFLAGS+=-framework Foundation -framework Metal -framework MetalKit -framework MetalPerformanceShaders -framework CoreML
 	export LLAMA_METAL=1
 	export WHISPER_METAL=1
 endif

[glebnaz]

@gaoyifan For me your patch doesn't help.

go mod edit -replace github.com/nomic-ai/gpt4all/gpt4all-bindings/golang=/Users/glebnaz/Documents/#main/workspace/localai/sources/gpt4all/gpt4all-bindings/golang
go mod edit -replace github.com/go-skynet/go-ggml-transformers.cpp=/Users/glebnaz/Documents/#main/workspace/localai/sources/go-ggml-transformers
go mod edit -replace github.com/donomii/go-rwkv.cpp=/Users/glebnaz/Documents/#main/workspace/localai/sources/go-rwkv
go mod edit -replace github.com/ggerganov/whisper.cpp=/Users/glebnaz/Documents/#main/workspace/localai/sources/whisper.cpp
go mod edit -replace github.com/ggerganov/whisper.cpp/bindings/go=/Users/glebnaz/Documents/#main/workspace/localai/sources/whisper.cpp/bindings/go
go mod edit -replace github.com/go-skynet/go-bert.cpp=/Users/glebnaz/Documents/#main/workspace/localai/sources/go-bert
go mod edit -replace github.com/mudler/go-stable-diffusion=/Users/glebnaz/Documents/#main/workspace/localai/sources/go-stable-diffusion
go mod edit -replace github.com/M0Rf30/go-tiny-dream=/Users/glebnaz/Documents/#main/workspace/localai/sources/go-tiny-dream
go mod edit -replace github.com/mudler/go-piper=/Users/glebnaz/Documents/#main/workspace/localai/sources/go-piper
go mod download
touch prepare-sources
touch prepare
CGO_LDFLAGS=" -lcblas -framework Accelerate -framework Foundation -framework Metal -framework MetalKit -framework MetalPerformanceShaders -framework CoreML" C_INCLUDE_PATH=/Users/glebnaz/Documents/#main/workspace/localai/sources/go-ggml-transformers LIBRARY_PATH=/Users/glebnaz/Documents/#main/workspace/localai/sources/go-ggml-transformers \
	go build -ldflags " -X "github.com/go-skynet/LocalAI/internal.Version=v2.4.1-2-g62a02cd" -X "github.com/go-skynet/LocalAI/internal.Commit=62a02cd1feb7cf8a75bcbe253fb95f204f022c1f"" -tags "" -o backend-assets/grpc/falcon-ggml ./backend/go/llm/falcon-ggml/
backend/go/llm/transformers/dolly.go:11:2: falcon.go: malformed #cgo argument: -I/Users/glebnaz/Documents/#main/workspace/localai/sources/go-ggml-transformers/ggml.cpp/include/
make: *** [backend-assets/grpc/falcon-ggml] Error 1

but thank you for your comment

[muellest]

Anything new on this topic? - Are there any identified workarounds or solutions?

@mudler , could you please confirm whether this can be recognized as a confirmed bug?

[glebnaz]

@mudler do you have time to participate to solution?

[mudler]

I don't have a Mac (yet) to double check, but the CI doesn't show the same symptoms. Actually the falcon-ggml backend is a candidate for deprecation (see: #1126 for more context).

Maybe @dave-gray101 can chime in here, I think you fixed MacOS builds lately right?

However, in any case I think you could workaround that but selectively choose what backend to build during build time, for instance to build only llama.cpp, you can run:

make GRPC_BACKENDS=backend-assets/grpc/llama-cpp build

see also: https://localai.io/basics/build/#build-only-a-single-backend

[muellest]

@dave-gray101 Can you support/help us?

[Cyb3rDudu]

I am reproducable able to build 2.7 with BUILD_TYPE=metal BUILD_GRPC_FOR_BACKEND_LLAMA=true on Mac M2 Max(12) on Sonoma with 32GB Ram. However it's a memory heavy compilation and takes a while.
I was not able to build it by linking against installed dependencies.

EDIT:

make BUILD_TYPE=metal BUILD_GRPC_FOR_BACKEND_LLAMA=true build  5209.78s user 20787.18s system 526% cpu 1:22:15.54 total

[Cyb3rDudu]

I was able to compile today with tts,stablediffusion and tinydream backend. It builds but it's a bit of a mess. I made a PR for one issue but I'm not sure where to implement the remaining fixes.

[qdrddr]

Please consider adding Core ML model package format support to utilize Apple Silicone Nural Engine + GPU.

List of Core ML package format models

https://github.com/likedan/Awesome-CoreML-Models

[qdrddr]

Here is some additional info about running LLMs locally on Apple Silicone. Core ML is a framework that can redistribute workload across CPU, GPU & Nural Engine (ANE). ANE is available on all modern Apple Devices: iPhones & Macs (A14 or newer and M1 or newer). Ideally, we want to run LLMs on ANE only as it has optimizations for running ML tasks compared to GPU. Apple claims "deploying your Transformer models on Apple devices with an A14 or newer and M1 or newer chip to achieve up to 10 times faster and 14 times lower peak memory consumption compared to baseline implementations".

1. To utilize Core ML first, you need to convert a model from TensorFlow, PyTorch to Core ML model package format using coremltools (or simply utilize existing models in Core ML package format ).
2. Second, you must now use that converted package with an implementation designed for Apple Devices. Here is the Apple XCode reference PyTorch implementation.

https://machinelearning.apple.com/research/neural-engine-transformers

[qdrddr]

You might also be interested in another implementation Swift Transformers.

Work in progress on CoreML implementation for [whisper.cpp]. They see x3 performance improvements for some models. (ggerganov/whisper.cpp#548) you might be interested in. Example of CoreML application
https://github.com/huggingface/swift-chat

[benmarte]

I'm getting a different error when trying to follow the Build on Mac instructions, it seems like the gpt4all-backend repo changed or something and it's giving me this error:

make BUILD_TYPE=metal BUILD_GRPC_FOR_BACKEND_LLAMA=true build
git clone --recurse-submodules https://github.com/go-skynet/go-llama.cpp sources/go-llama.cpp
Cloning into 'sources/go-llama.cpp'...
remote: Enumerating objects: 775, done.
remote: Counting objects: 100% (443/443), done.
remote: Compressing objects: 100% (173/173), done.
remote: Total 775 (delta 314), reused 339 (delta 258), pack-reused 332
Receiving objects: 100% (775/775), 238.97 KiB | 3.02 MiB/s, done.
Resolving deltas: 100% (451/451), done.
Submodule 'llama.cpp' (https://github.com/ggerganov/llama.cpp) registered for path 'llama.cpp'
Cloning into '/Users/benmarte/Documents/github/LocalAI/sources/go-llama.cpp/llama.cpp'...
remote: Enumerating objects: 23832, done.        
remote: Counting objects: 100% (8015/8015), done.        
remote: Compressing objects: 100% (301/301), done.        
remote: Total 23832 (delta 7845), reused 7743 (delta 7714), pack-reused 15817        
Receiving objects: 100% (23832/23832), 43.51 MiB | 18.81 MiB/s, done.
Resolving deltas: 100% (16921/16921), done.
Submodule path 'llama.cpp': checked out 'ac43576124a75c2de6e333ac31a3444ff9eb9458'
cd sources/go-llama.cpp && git checkout -b build 2b57a8ae43e4699d3dc5d1496a1ccd42922993be && git submodule update --init --recursive --depth 1
Switched to a new branch 'build'
git clone --recurse-submodules https://github.com/nomic-ai/gpt4all sources/gpt4all
Cloning into 'sources/gpt4all'...
remote: Enumerating objects: 14638, done.
remote: Counting objects: 100% (4322/4322), done.
remote: Compressing objects: 100% (314/314), done.
remote: Total 14638 (delta 4122), reused 4044 (delta 4008), pack-reused 10316
Receiving objects: 100% (14638/14638), 14.57 MiB | 18.04 MiB/s, done.
Resolving deltas: 100% (9960/9960), done.
Submodule 'llama.cpp-mainline' (https://github.com/nomic-ai/llama.cpp.git) registered for path 'gpt4all-backend/llama.cpp-mainline'
Cloning into '/Users/benmarte/Documents/github/LocalAI/sources/gpt4all/gpt4all-backend/llama.cpp-mainline'...
remote: Enumerating objects: 15299, done.        
remote: Counting objects: 100% (4764/4764), done.        
remote: Compressing objects: 100% (157/157), done.        
remote: Total 15299 (delta 4687), reused 4607 (delta 4607), pack-reused 10535        
Receiving objects: 100% (15299/15299), 20.29 MiB | 18.50 MiB/s, done.
Resolving deltas: 100% (10663/10663), done.
Submodule path 'gpt4all-backend/llama.cpp-mainline': checked out 'a3f03b7e793ee611c4918235d4532ee535a9530d'
Submodule 'kompute' (https://github.com/nomic-ai/kompute.git) registered for path 'gpt4all-backend/llama.cpp-mainline/kompute'
Cloning into '/Users/benmarte/Documents/github/LocalAI/sources/gpt4all/gpt4all-backend/llama.cpp-mainline/kompute'...
remote: Enumerating objects: 9083, done.        
remote: Counting objects: 100% (218/218), done.        
remote: Compressing objects: 100% (130/130), done.        
remote: Total 9083 (delta 96), reused 168 (delta 78), pack-reused 8865        
Receiving objects: 100% (9083/9083), 17.57 MiB | 17.10 MiB/s, done.
Resolving deltas: 100% (5716/5716), done.
Submodule path 'gpt4all-backend/llama.cpp-mainline/kompute': checked out 'd1e3b0953cf66acc94b2e29693e221427b2c1f3f'
cd sources/gpt4all && git checkout -b build 27a8b020c36b0df8f8b82a252d261cda47cf44b8 && git submodule update --init --recursive --depth 1
fatal: not a git repository: ../../.git/modules/llama.cpp-230511
fatal: could not reset submodule index
make: *** [sources/gpt4all] Error 128

This is my system report:

Model Name:	MacBook Pro
Chip:	Apple M1 Max
Total Number of Cores:	10 (8 performance and 2 efficiency)
Memory:	64 GB
System Firmware Version:	10151.101.3
OS Loader Version:	10151.101.3

[mintyleaf]

@mudler

just using make build after installing dependencies from the docs page and exporting right paths to the brew installed cc and cxx is working for the last release tag + for the latest commit after fixing llama.cpp build

so that issue is resolved?