Merge pull request #19 from janhq/fix_inf_sd_ggml

Fix inf sd ggml

docker-compose.yml

@@ -190,15 +190,16 @@ services:
       dockerfile: inference.Dockerfile
     # Mount the directory that contains the downloaded model.
     volumes:
-      - ./jan-inference/sd/models:/models
-      - ./jan-inference/sd/output/:/serving/output
+      - ./jan-inference/sd/models:/models/
+      - ./jan-inference/sd/output/:/output/
     command:  /bin/bash -c "python -m uvicorn main:app --proxy-headers --host 0.0.0.0 --port 8000"
     environment:
       # Specify the path to the model for the web application.
-      BASE_URL: http://0.0.0.0:8000
+      BASE_URL: http://0.0.0.0:8001
       MODEL_NAME: ${SD_MODEL_FILE}.q4_0.bin
-      MODEL_DIR: "/models"
-      SD_PATH: "/sd"
+      MODEL_DIR: /models
+      OUTPUT_DIR: /output
+      SD_PATH: /sd
       PYTHONUNBUFFERED: 1
     ports:
       - 8001:8000

jan-inference/sd/inference.Dockerfile

@@ -1,4 +1,6 @@
-FROM python:3.9.17 as build
+ARG UBUNTU_VERSION=22.04
+
+FROM ubuntu:$UBUNTU_VERSION as build
 
 RUN apt-get update && apt-get install -y build-essential git cmake
 

jan-inference/sd/main.py

@@ -8,11 +8,12 @@ from pydantic import BaseModel
 
 app = FastAPI()
 
-OUTPUT_DIR = "output"
+OUTPUT_DIR = os.environ.get("OUTPUT_DIR", "output")
 SD_PATH = os.environ.get("SD_PATH", "./sd")
 MODEL_DIR = os.environ.get("MODEL_DIR", "./models")
 MODEL_NAME = os.environ.get(
     "MODEL_NAME", "v1-5-pruned-emaonly-ggml-model-q5_0.bin")
+BASE_URL = os.environ.get("BASE_URL", "http://localhost:8000")
 
 
 class Payload(BaseModel):
@@ -51,7 +52,7 @@ def run_command(payload: Payload, filename: str):
                ]
 
     try:
-        subprocess.run(command, timeout=5*60)
+        subprocess.run(command)
     except subprocess.CalledProcessError:
         raise HTTPException(
             status_code=500, detail="Failed to execute the command.")
@@ -63,10 +64,11 @@ async def run_inference(background_tasks: BackgroundTasks, payload: Payload):
     filename = f"{uuid4()}.png"
 
     # We will use background task to run the command so it won't block
-    background_tasks.add_task(run_command, payload, filename)
+    # background_tasks.add_task(run_command, payload, filename)
+    run_command(payload, filename)
 
     # Return the expected path of the output file
-    return {"url": f'/serve/{filename}'}
+    return {"url": f'{BASE_URL}/serve/{filename}'}
 
 
 @app.get("/serve/{filename}")

jan-inference/sd/sd_cpp

@@ -1 +1 @@
-Subproject commit 0d7f04b135cd48e8d62aecd09a52eb2afa482744
+Subproject commit c8f85a4e3063e2cdb27db57b8f6167da16453e0c