Prepare public release with a minimal deployment setup (#16)

* Move deployment to deployment-next

* Add deployment setup

* Update deployment-next

* Remove vector label

* update README.md

.dockerignore

@@ -1,4 +1,6 @@
+data
 deployment
+deployment-next
 
 **/.git
 **/node_modules

.gitignore

@@ -1 +1,2 @@
 __pycache__
+data

Makefile

@@ -14,19 +14,3 @@ $(PRE_COMMIT_HOOK):
 	poetry run pre-commit install --install-hooks
 
 setup-development-environment: install-poetry $(PRE_COMMIT_HOOK)
-
-
-UP_FLAGS := up --remove-orphans --remove-orphans
-DEV_FLAGS := $(UP_FLAGS) --build
-
-up:
-	docker-compose -f deployment/docker-compose.yml $(UP_FLAGS)
-
-up-triton:
-	docker-compose -f deployment/docker-compose.yml -f deployment/docker-compose.triton.yml $(UP_FLAGS)
-
-dev:
-	docker-compose -f deployment/docker-compose.yml -f deployment/docker-compose.dev.yml $(DEV_FLAGS)
-
-dev-triton:
-	docker-compose -f deployment/docker-compose.yml -f deployment/docker-compose.triton.yml -f deployment/docker-compose.dev.yml $(DEV_FLAGS)

README.md

@@ -18,8 +18,5 @@ Assuming Linux workstation with:
 Use `make setup-development-environment` to setup basic dev environment, and `make dev` to start local development server.
 
 ## Deployment
-1. `make up-triton`
-2. Open Admin Panel [http://localhost:8501](http://localhost:8501)
-3. Types some code in editor!
 
-![Screenshot from 2023-03-25 21-31-24](https://user-images.githubusercontent.com/388154/227720212-7e7480a1-abb1-4baf-b971-d391ea8136c8.png)
+See [deployment](./deployment/README.md)

deployment-next/Makefile

@@ -0,0 +1,14 @@
+UP_FLAGS := up --remove-orphans --remove-orphans
+DEV_FLAGS := $(UP_FLAGS) --build
+
+up:
+	docker-compose -f docker-compose.yml $(UP_FLAGS)
+
+up-triton:
+	docker-compose -f docker-compose.yml -f docker-compose.triton.yml $(UP_FLAGS)
+
+dev:
+	docker-compose -f docker-compose.yml -f docker-compose.dev.yml $(DEV_FLAGS)
+
+dev-triton:
+	docker-compose -f docker-compose.yml -f docker-compose.triton.yml -f docker-compose.dev.yml $(DEV_FLAGS)

deployment-next/README.md

@@ -0,0 +1,12 @@
+## Deployment
+
+1. Start service
+   ```bash
+   docker-compose up
+   ```
+2. Test API endpoint with curl
+   ```bash
+   curl -X POST http://localhost:5000/v1/completions -H 'Content-Type: application/json' --data '{
+       "prompt": "def binarySearch(arr, left, right, x):\n    mid = (left +"
+   }'
+   ```

deployment-next/docker-compose.triton.yml

@@ -16,7 +16,7 @@ services:
     shm_size: 1gb
     volumes:
       - ./scripts:/scripts
-      - ./data/hf_cache:/root/.cache/huggingface
+      - ../data/hf_cache:/root/.cache/huggingface
     deploy:
       resources:
         reservations:

deployment-next/docker-compose.yml

@@ -0,0 +1,50 @@
+version: '3.3'
+
+services:
+  init:
+    image: tabbyml/tabby
+    container_name: tabby-init
+    command: python -m tabby.tools.model_preload --repo_id TabbyML/NeoX-70M
+    volumes:
+      - ../data/hf_cache:/root/.cache/huggingface
+
+  server:
+    image: tabbyml/tabby
+    container_name: tabby-server
+    command: uvicorn tabby.server:app --host 0.0.0.0 --port 5000
+    environment:
+      - MODEL_NAME=TabbyML/NeoX-70M
+      - EVENTS_LOG_DIR=/logs
+    ports:
+      - "5000:5000"
+    volumes:
+      - ../data/logs/tabby-server:/logs
+      - ../data/hf_cache:/root/.cache/huggingface
+    depends_on:
+      init:
+        condition: service_completed_successfully
+
+  admin:
+    image: tabbyml/tabby
+    container_name: tabby-admin
+    command: streamlit run tabby/admin/Home.py
+    ports:
+      - "8501:8501"
+
+  vector:
+    image: timberio/vector:0.28.1-alpine
+    container_name: tabby-vector
+    volumes:
+      - ./config/vector.toml:/etc/vector/vector.toml:ro
+      - ../data/logs:/logs
+
+  dagu:
+    image: tabbyml/tabby
+    container_name: tabby-dagu
+    command: dagu server --host 0.0.0.0 --port 8080
+    volumes:
+      - ./config:/config:ro
+      - ../data/repositories:/repositories
+      - ../data/dataset:/dataset
+    environment:
+      DAGU_DAGS: tabby/tasks

deployment-next/scripts/triton.sh

@@ -0,0 +1,18 @@
+#!/bin/bash
+set -e
+
+# Get model dir.
+MODEL_DIR=$(python3 <<EOF
+from huggingface_hub import snapshot_download
+
+print(snapshot_download(repo_id='$MODEL_NAME', allow_patterns='triton/**/*', local_files_only=True))
+EOF
+)
+
+# Set model dir in triton config.
+sed -i 's@${MODEL_DIR}@'$MODEL_DIR'@g' $MODEL_DIR/triton/fastertransformer/config.pbtxt
+
+# Start triton server.
+mpirun -n 1 \
+  --allow-run-as-root /opt/tritonserver/bin/tritonserver \
+  --model-repository=$MODEL_DIR/triton

deployment/README.md

@@ -1,12 +1,7 @@
 ## Deployment
 
-1. Start service
-   ```bash
-   docker-compose up
-   ```
-2. Test API endpoint with curl
-   ```bash
-   curl -X POST http://localhost:5000/v1/completions -H 'Content-Type: application/json' --data '{
-       "prompt": "def binarySearch(arr, left, right, x):\n    mid = (left +"
-   }'
-   ```
+1. `docker-compose up`
+2. Open Admin Panel [http://localhost:8501](http://localhost:8501)
+3. Types some code in editor!
+
+![Screenshot from 2023-03-25 21-31-24](https://user-images.githubusercontent.com/388154/227720212-7e7480a1-abb1-4baf-b971-d391ea8136c8.png)

deployment/docker-compose.yml

@@ -6,7 +6,7 @@ services:
     container_name: tabby-init
     command: python -m tabby.tools.model_preload --repo_id TabbyML/NeoX-70M
     volumes:
-      - ./data/hf_cache:/root/.cache/huggingface
+      - ../data/hf_cache:/root/.cache/huggingface
 
   server:
     image: tabbyml/tabby
@@ -14,15 +14,16 @@ services:
     command: uvicorn tabby.server:app --host 0.0.0.0 --port 5000
     environment:
       - MODEL_NAME=TabbyML/NeoX-70M
-      - EVENTS_LOG_DIR=/logs
+      - MODEL_BACKEND=triton
     ports:
       - "5000:5000"
     volumes:
-      - ./data/logs/tabby-server:/logs
-      - ./data/hf_cache:/root/.cache/huggingface
+      - ../data/hf_cache:/root/.cache/huggingface
     depends_on:
       init:
         condition: service_completed_successfully
+      triton:
+        condition: service_healthy
 
   admin:
     image: tabbyml/tabby
@@ -31,20 +32,28 @@ services:
     ports:
       - "8501:8501"
 
-  vector:
-    image: timberio/vector:0.28.1-alpine
-    container_name: tabby-vector
+  triton:
+    image: tabbyml/fastertransformer_backend
+    container_name: tabby-triton
+    command: /scripts/triton.sh
+    shm_size: 1gb
     volumes:
-      - ./config/vector.toml:/etc/vector/vector.toml:ro
-      - ./data/logs:/logs
-
-  dagu:
-    image: tabbyml/tabby
-    container_name: tabby-dagu
-    command: dagu server --host 0.0.0.0 --port 8080
-    volumes:
-      - ./config:/config:ro
-      - ./data/repositories:/repositories
-      - ./data/dataset:/dataset
+      - ./scripts:/scripts
+      - ../data/hf_cache:/root/.cache/huggingface
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - driver: nvidia
+              count: all
+              capabilities: [gpu]
     environment:
-      DAGU_DAGS: tabby/tasks
+      - MODEL_NAME=TabbyML/NeoX-70M
+    depends_on:
+      init:
+        condition: service_completed_successfully
+    healthcheck:
+      test: ["CMD", "curl", "-f", "http://localhost:8002/metrics"]
+      interval: 2s
+      timeout: 2s
+      start_period: 120s

tabby/admin/Home.py

@@ -5,7 +5,6 @@ from utils.service_info import ServiceInfo
 SERVICES = [
     ServiceInfo(label="server", url="http://server:5000"),
     ServiceInfo(label="triton", url="http://triton:8002/metrics"),
-    ServiceInfo(label="vector", url="http://vector:8686/health"),
 ]