turn to work version

.gitea/docker.yml

@@ -0,0 +1,46 @@
+name: Build, Push and Redeploy Docker Image
+
+on:
+  push:
+    branches:
+      - main
+
+jobs:
+  build-and-push:
+    runs-on: ubuntu-latest
+
+    env:
+      REGISTRY_HOST: git.byte-mate.ru
+
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v3
+
+      - name: Set up QEMU (for multi-arch, optional)
+        uses: docker/setup-qemu-action@v3
+
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+
+      - name: Log in to Gitea Container Registry
+        run: |
+          echo "${{ secrets.GIT_PASSWORD }}" | docker login "${{ env.REGISTRY_HOST }}" -u "${{ github.actor }}" --password-stdin
+        
+      - name: Get short SHA
+        id: vars
+        run: echo "sha_short=$(git rev-parse --short HEAD)" >> "$GITHUB_ENV"
+
+      - name: Build and push
+        run: |
+          docker buildx build \
+          --no-cache \
+          --push \
+          --platform linux/arm64/v8 \
+          --tag ${{ env.REGISTRY_HOST }}/drholy/rkllm-gradio-server:${{ env.sha_short }} \
+          --tag ${{ env.REGISTRY_HOST }}/drholy/rkllm-gradio-server:latest \
+          --push \
+          .
+
+      # - name: Redeploy stack in Portainer
+      #   run: |
+      #     curl -X POST "${{ secrets.PORTAINER_WEBHOOK_URL }}"

Dockerfile

@@ -2,7 +2,6 @@
 FROM python:3.11.11
 
 # Установка метаданных
-LABEL maintainer="your-email@example.com"
 LABEL description="Docker image with Python 3.11.11 and required packages"
 
 # Обновление системы и установка базовых зависимостей
@@ -24,6 +23,8 @@ RUN pip install --upgrade pip && \
 # Проверка версии Python и pip
 RUN python --version && pip --version
 
+RUN pip install --no-cache-dir -r requirements.txt
+
 # Очистка кэша
 RUN apt-get clean && \
     rm -rf /tmp/* /var/tmp/*

README.md

@@ -53,9 +53,6 @@ http://localhost:8080
 Проект использует следующие виртуальные файловые системы:
 
 - `/app` — для приложения.
-- `/cache/cache` — для кэша.
-- `/cache/site-packages` — для пакетов Python.
-- `/offline_packages` — для офлайн пакетов.
 - `/ezrknn-llm` — для дополнительных библиотек.
 
 ## Настройка и использование моделей

app/entrypoint.sh

@@ -10,7 +10,7 @@ message_print() {
 cd /
 
 message_print "Changing to repository..."
-git clone https://github.com/DrHo1y/ezrknn-llm
+git clone https://git.home/drholy/ezrknn-llm
 cd ezrknn-llm/
 cp ./rkllm-runtime/runtime/Linux/librkllm_api/aarch64/* /usr/lib
 cp ./rkllm-runtime/runtime/Linux/librkllm_api/include/* /usr/local/include
@@ -29,10 +29,6 @@ echo "root soft nofile 16384" >> /etc/security/limits.conf
 echo "root hard nofile 1048576" >> /etc/security/limits.conf
 message_print "Done installing ezrknn-llm!"
 
-message_print "Install python packages"
-cd /offline_packages
-python -m pip install --no-index --find-links . -r requirements.txt
-
 message_print "Start Gradio server"
 cd /app
 python rkllm_server_gradio.py

app/rkllm_server_gradio.py

@@ -6,8 +6,8 @@ from model_class import *
 from mesh_utils import *
 
 # Set environment variables
-os.environ["GRADIO_SERVER_NAME"] = "0.0.0.0"
-os.environ["GRADIO_SERVER_PORT"] = "8080"
+#os.environ["GRADIO_SERVER_NAME"] = "0.0.0.0"
+#os.environ["GRADIO_SERVER_PORT"] = 8080
 os.environ["RKLLM_LOG_LEVEL"] = "1"
 # Set resource limit
 resource.setrlimit(resource.RLIMIT_NOFILE, (102400, 102400))
@@ -29,7 +29,7 @@ if __name__ == "__main__":
         print(init_msg)
         sys.stdout.flush()
         rkllm_model = RKLLMLoaderClass(model=model)
-        model_init = f"RKLLM Model, {rkllm_model.model_name} has been initialized successfully！"
+        model_init = f"RKLLM Model, {rkllm_model.model_name} has been initialized successfully!"
         print(model_init)
         complete_init = "=============================="
         print(complete_init)
@@ -106,7 +106,13 @@ if __name__ == "__main__":
     # Enable the event queue system.
     chatRKLLM.queue()
     # Start the Gradio application.
-    chatRKLLM.launch()
+    chatRKLLM.launch(
+        server_name="0.0.0.0",
+        server_port=8080,
+        share=False,
+	show_error=True,
+	quiet=False
+    )
 
     print("====================")
     print("RKLLM model inference completed, releasing RKLLM model resources...")

docker-compose.yaml

@@ -9,9 +9,6 @@ services:
     privileged: true
     volumes:
       - ./app:/app
-      - ./cache/cache:/root/.cache
-      - ./cache/site-packages:/usr/local/lib/python3.11/site-packages
-      - ./offline_packages:/offline_packages
       - ./ezrknn-llm:/ezrknn-llm
     entrypoint: ./app/entrypoint.sh
     ports:
@@ -22,8 +19,4 @@ networks:
   apps:
     name: apps
     driver: bridge
-    ipam:
-      driver: default
-      config:
-        - subnet: "172.100.0.0/24"
-          gateway: "172.100.0.1"
+

offline_packages/requirements.txt

@@ -1,8 +0,0 @@
-gradio==5.14.0
-gradio_client==1.7.0
-huggingface-hub==0.26.2
-Jinja2==3.1.4
-numpy==2.1.3
-transformers==4.46.2
-trimesh==4.5.2
-sentencepiece==0.2.0

requirements.txt

@@ -0,0 +1,8 @@
+gradio==5.25.0
+gradio_client>=1.7.0
+huggingface-hub==0.28.1
+Jinja2==3.1.4
+numpy==2.1.3
+transformers==4.46.2
+trimesh==4.5.2
+sentencepiece==0.2.0