Open olariuromeo opened 6 months ago
the build can sometimes fail randomly if you hit the github rate limiter. it does not throw an error until much later. I would say try again and watch the pulling of the github files and check for anything that fails while pulling. i have all of these settings the same altho i do not "make" i rebuild the image for my cpu arch. altho one thing i notice is your image is out of date. the 2.9.0 image is available. i would honestly say if you use the latest image from "quay" it will verify that it does indeed build (even master-cublas-cuda12-ffmpeg) and then you can rebuild if needed but if your cpu supports avx2 i do not think you need to rebuild.
GO_TAGS=stablediffusion,tts BUILD_TYPE=cuBLAS BUILD_GRPC_FOR_BACKEND_LLAMA=true
the build can sometimes fail randomly if you hit the github rate limiter. it does not throw an error until much later. I would say try again and watch the pulling of the github files and check for anything that fails while pulling. i have all of these settings the same altho i do not "make" i rebuild the image for my cpu arch. altho one thing i notice is your image is out of date. the 2.9.0 image is available. i would honestly say if you use the latest image from "quay" it will verify that it does indeed build (even master-cublas-cuda12-ffmpeg) and then you can rebuild if needed but if your cpu supports avx2 i do not think you need to rebuild.
GO_TAGS=stablediffusion,tts BUILD_TYPE=cuBLAS BUILD_GRPC_FOR_BACKEND_LLAMA=true
You're right, it's not mandatory to recreate the image, you can use the default one to make it work, but I want to make some changes and if I can't build an image with cuda support, I can't do them, and my resources are quite limited in the system and I would I want to make a lighter image just to work for my server. I really don't understand why I can't create a custom image for my system. . My cpu supports avx2 and I have 64 gb of ram, but the video card only has 10 gb of vram And I would like to experiment with making some images each with different characteristics and some small change, one with stablediffusion, one with ttts support and another without stablediffusion and tts support with cuda12 but none of them work properly. I can build the image without error only if I delete the replace directive from go mod, I tried with different versions of golan and python but when I build the container, and start the inference, with not cuda support although it shows me that the image has grpc backend with cublas, so it is useless, and the existing documentation is quite lacking
the build can sometimes fail randomly if you hit the github rate limiter. it does not throw an error until much later. I would say try again and watch the pulling of the github files and check for anything that fails while pulling. i have all of these settings the same altho i do not "make" i rebuild the image for my cpu arch. altho one thing i notice is your image is out of date. the 2.9.0 image is available. i would honestly say if you use the latest image from "quay" it will verify that it does indeed build (even master-cublas-cuda12-ffmpeg) and then you can rebuild if needed but if your cpu supports avx2 i do not think you need to rebuild.
GO_TAGS=stablediffusion,tts BUILD_TYPE=cuBLAS BUILD_GRPC_FOR_BACKEND_LLAMA=true
tts: enables text-to-speech with go-piper requires REBUILD=true
Have exactly the same issue on a macos. Don't think this has to do anything with github rate limiting as the file already exists locally.
the build can sometimes fail randomly if you hit the github rate limiter. it does not throw an error until much later. I would say try again and watch the pulling of the github files and check for anything that fails while pulling. i have all of these settings the same altho i do not "make" i rebuild the image for my cpu arch. altho one thing i notice is your image is out of date. the 2.9.0 image is available. i would honestly say if you use the latest image from "quay" it will verify that it does indeed build (even master-cublas-cuda12-ffmpeg) and then you can rebuild if needed but if your cpu supports avx2 i do not think you need to rebuild.
GO_TAGS=stablediffusion,tts BUILD_TYPE=cuBLAS BUILD_GRPC_FOR_BACKEND_LLAMA=true "your image is out of date" when you build a local image, it really doesn't matter what name you give to the image.
LocalAI version:
master I clone yesterday Environment, CPU architecture, OS, and Version:
Host server ubuntu 22.04, Linux office 6.5.0-17-generic #17~22.04.1-Ubuntu SMP PREEMPT_DYNAMIC Tue Jan 16 14:32:32 UTC 2 x86_64 x86_64 x86_64 GNU/Linux go , golang 1.21.7, python 3.11.5, Docker version 24.0.7, video card nvidia 3080 10gb vram, system memory 64gb ddr4. cuda 12.2
Describe the bug
the file exist :
To Reproduce
make GO_TAGS=stablediffusion,tts BUILD_TYPE=cuBLAS BUILD_GRPC_FOR_BACKEND_LLAMA=true build
Expected behavior
create an image Logs
already provided Additional context
The inference without a video card works decently, I would like to see how nvidia acceleration works. For build a new image it is necessary to delete replace directive from go.mod who has this error.