feat(diffusers): add builds for nvidia-l4t

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

.github/workflows/backend.yml

@@ -87,6 +87,18 @@ jobs:
             backend: "diffusers"
             dockerfile: "./backend/Dockerfile.python"
             context: "./backend"
+          - build-type: 'l4t'
+            cuda-major-version: "12"
+            cuda-minor-version: "0"
+            platforms: 'linux/amd64'
+            tag-latest: 'auto'
+            tag-suffix: '-gpu-nvidia-l4t-diffusers'
+            runs-on: 'ubuntu-24.04-arm'
+            base-image: "nvcr.io/nvidia/l4t-jetpack:r36.4.0"
+            skip-drivers: 'true'
+            backend: "diffusers"
+            dockerfile: "./backend/Dockerfile.python"
+            context: "./backend"
           # CUDA 11 additional backends
           - build-type: 'cublas'
             cuda-major-version: "11"

Dockerfile

@@ -9,7 +9,7 @@ ENV DEBIAN_FRONTEND=noninteractive
 RUN apt-get update && \
     apt-get install -y --no-install-recommends \
         ca-certificates curl wget espeak-ng libgomp1 \
-        python3 python-is-python3 ffmpeg && \
+        python3 python-is-python3 ffmpeg libopenblas-base libopenblas-dev && \
     apt-get clean && \
     rm -rf /var/lib/apt/lists/*
 

Makefile

@@ -132,6 +132,9 @@ test: test-models/testmodel.ggml protogen-go
 	$(MAKE) test-tts
 	$(MAKE) test-stablediffusion
 
+backends/diffusers: docker-build-diffusers docker-save-diffusers build
+	./local-ai backends install "ocifile://$(abspath ./backend-images/diffusers.tar)"
+
 backends/llama-cpp: docker-build-llama-cpp docker-save-llama-cpp build
 	./local-ai backends install "ocifile://$(abspath ./backend-images/llama-cpp.tar)"
 
@@ -427,7 +430,10 @@ docker-build-transformers:
 	docker build --build-arg BUILD_TYPE=$(BUILD_TYPE) --build-arg BASE_IMAGE=$(BASE_IMAGE) -t local-ai-backend:transformers -f backend/Dockerfile.python --build-arg BACKEND=transformers .
 
 docker-build-diffusers:
-	docker build --build-arg BUILD_TYPE=$(BUILD_TYPE) --build-arg BASE_IMAGE=$(BASE_IMAGE) -t local-ai-backend:diffusers -f backend/Dockerfile.python --build-arg BACKEND=diffusers .
+	docker build --progress=plain --build-arg BUILD_TYPE=$(BUILD_TYPE) --build-arg BASE_IMAGE=$(BASE_IMAGE) -t local-ai-backend:diffusers -f backend/Dockerfile.python --build-arg BACKEND=diffusers ./backend
+
+docker-save-diffusers: backend-images
+	docker save local-ai-backend:diffusers -o backend-images/diffusers.tar
 
 docker-build-whisper:
 	docker build --build-arg BUILD_TYPE=$(BUILD_TYPE) --build-arg BASE_IMAGE=$(BASE_IMAGE) -t local-ai-backend:whisper -f backend/Dockerfile.golang --build-arg BACKEND=whisper  .

backend/index.yaml

@@ -168,6 +168,7 @@
     nvidia: "cuda12-diffusers"
     intel: "intel-diffusers"
     amd: "rocm-diffusers"
+    nvidia-l4t: "nvidia-l4t-diffusers"
 - &exllama2
   name: "exllama2"
   urls:
@@ -853,6 +854,17 @@
     nvidia: "cuda12-diffusers-development"
     intel: "intel-diffusers-development"
     amd: "rocm-diffusers-development"
+    nvidia-l4t: "nvidia-l4t-diffusers-development"
+- !!merge <<: *diffusers
+  name: "nvidia-l4t-diffusers"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-l4t-diffusers"
+  mirrors:
+    - localai/localai-backends:latest-gpu-nvidia-l4t-diffusers
+- !!merge <<: *diffusers
+  name: "nvidia-l4t-diffusers-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-l4t-diffusers"
+  mirrors:
+    - localai/localai-backends:master-gpu-nvidia-l4t-diffusers
 - !!merge <<: *diffusers
   name: "cuda12-diffusers"
   uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-12-diffusers"

backend/python/common/libbackend.sh

@@ -57,6 +57,11 @@ function init() {
 # - hipblas
 # - intel
 function getBuildProfile() {
+    if [ "x${BUILD_TYPE}" == "xl4t" ]; then
+        echo "l4t"
+        return 0
+    fi
+
     # First check if we are a cublas build, and if so report the correct build profile
     if [ x"${BUILD_TYPE}" == "xcublas" ]; then
         if [ ! -z ${CUDA_MAJOR_VERSION} ]; then

backend/python/diffusers/requirements-l4t.txt

@@ -0,0 +1,9 @@
+--extra-index-url https://pypi.jetson-ai-lab.io/jp6/cu126/
+torch
+diffusers
+transformers
+accelerate
+compel
+peft
+optimum-quanto
+numpy<2