add full build and support all vram modes

Dockerfile

@@ -26,5 +26,6 @@ ENV ENABLE_EMA="1"
 ENV VAE_PATH="models/animevae.pt"
 ENV PENULTIMATE="1"
 ENV PYTHONDONTWRITEBYTECODE=1
+ENV VRAM_MODE="medvram"
  
  CMD ["python3", "-m", "uvicorn", "--host", "0.0.0.0", "--port=6969", "main:app"]

Dockerfile.full

 ARG BASE=git-registry.mycard.moe/novelai-storage/naifu:latest
 
 FROM node:lts-bullseye as frontend
-RUN apt update && apt -y install git && npm install -g pnpm && \
+RUN apt update && apt -y install wget tar git && npm install -g pnpm && \
   git clone https://code.mycard.moe/novelai-storage/naifu-frontend.git /frontend && \
-  cd /frontend && pnpm install && pnpm run build
+  cd /frontend && pnpm install && pnpm run build && \
+  mkdir /models && cd /models && \
+  wget -O - https://cdn02.moecube.com:444/nanahira/naifu/models.tar.gz | tar -zxf -
 
 FROM $BASE
 COPY --from=frontend /frontend/build ./static
+COPY --from=frontend /models ./models
 ENV WITH_STATIC=1

docker-compose.yml

@@ -10,6 +10,7 @@ services:
       - ./hydra_node:/app/hydra_node:ro
       #- ./static:/app/static:ro
     environment:
+      VRAM_MODE: lowvram
       # TOKEN_SERVER: https://api.moecube.com/accounts/authUser
-      TOKEN: mycard
-      #WITH_STATIC: 1
+      # TOKEN: mycard
+      # WITH_STATIC: 1

hydra_node/config.py

@@ -178,7 +178,11 @@ def init_config_model():
             #attach it to the model
             model.premodules = modules
 
-    lowvram.setup_for_low_vram(model.model, True)
+    vramMode = os.getenv("VRAM_MODE", "medvram")
+    if vramMode != "highvram":
+        lowvram.setup_for_low_vram(model.model, vramMode == "medvram")
+    else:
+        model.model.to(torch.device("cuda"))
 
     config.model = model