Address Docker deployment review feedback

Author: HarshaSatyavardhan

.github/workflows/docker.yml

@@ -4,15 +4,14 @@ on:
   push:
     branches:
       - main
-      # - fix/docker-deployment
 
   pull_request:
     branches:
       - main
-  workflow_dispatch: 
+  workflow_dispatch:
 
 env:
-  IMAGE_NAME: fabnemepfl/mmirage
+  IMAGE_NAMESPACE: fabnemepfl
   REGISTRY: docker.io
 
 jobs:
@@ -25,10 +24,6 @@ jobs:
             path: docker/Dockerfile
             tag_base: amd64
             name: mmirage-git
-          - platform: ubuntu-24.04-arm
-            path: docker/Dockerfile
-            tag_base: arm64
-            name: mmirage-git
           - platform: ubuntu-latest
             path: docker/Dockerfile.cpu
             tag_base: amd64
@@ -39,7 +34,7 @@ jobs:
             name: mmirage-git-cpu
 
     runs-on: ${{ matrix.platform }}
-    environment: docker
+    environment: ${{ github.event_name != 'pull_request' && 'docker' || 'docker-ci' }}
 
     steps:
       - name: Free space (ARM)
@@ -89,5 +84,5 @@ jobs:
           file: ${{ matrix.path }}
           push: ${{ github.event_name != 'pull_request' }}
           tags: |
-            ${{ secrets.DOCKERHUB_USERNAME }}/${{ matrix.name }}:latest-${{ matrix.tag_base }}
-            ${{ secrets.DOCKERHUB_USERNAME }}/${{ matrix.name }}:${{ github.sha }}-${{ matrix.tag_base }}
+            ${{ env.REGISTRY }}/${{ env.IMAGE_NAMESPACE }}/${{ matrix.name }}:latest-${{ matrix.tag_base }}
+            ${{ env.REGISTRY }}/${{ env.IMAGE_NAMESPACE }}/${{ matrix.name }}:${{ github.sha }}-${{ matrix.tag_base }}

README.md

@@ -10,10 +10,18 @@ MMIRAGE, which stands for **M**odular **M**ultimodal **I**ntelligent **R**eforma
 
 ## How to install
 
-To install the library, you can clone it from GitHub and then use pip to install it directly. It is recommended to have already installed `torch` and `sglang` to take advantage of GPU acceleration.
+To install the library, clone it from GitHub and install it with pip. The base
+install does not include the local SGLang runtime:
 
 ```bash
 git clone git@github.com:EPFLiGHT/MMIRAGE.git
+pip install -e ./MMIRAGE
+```
+
+Install the GPU extra when using the SGLang-backed `llm` processor for local
+GPU inference:
+
+```bash
 pip install -e './MMIRAGE[gpu]'
 ```
 
@@ -43,9 +51,13 @@ The container requires an NVIDIA GPU. The `docker-compose.yml` is configured to
 
 Without these host-side prerequisites, `docker compose run` may fail to detect or use the GPU.
 
-### CPU-only (API-based LLMs)
+### CPU-only
 
-For API-based LLMs that do not require a GPU:
+The CPU image installs MMIRAGE without the GPU extra. It is suitable for
+workflows that do not instantiate the SGLang-backed `llm` processor, and is
+intended to support API-backed processors once they are available. Current
+configs that use `type: llm` require the GPU image or an install with the
+`[gpu]` extra.
 
 ```bash
 docker compose run --rm mmirage-cpu --config configs/your_config.yaml

docker/Dockerfile

@@ -8,4 +8,4 @@ WORKDIR /workspace/MMIRAGE
 
 RUN pip install --no-cache-dir .[gpu]
 
-ENTRYPOINT ["python3", "-c", "import mmirage.core.process, runpy; runpy.run_module('mmirage.shard_process', run_name='__main__')"]
+ENTRYPOINT ["python3", "-m", "mmirage.shard_process"]

docker/Dockerfile.cpu

@@ -5,4 +5,4 @@ WORKDIR /workspace/MMIRAGE
 
 RUN pip install --no-cache-dir .
 
-ENTRYPOINT ["python3", "-c", "import mmirage.core.process, runpy; runpy.run_module('mmirage.shard_process', run_name='__main__')"]
+ENTRYPOINT ["python3", "-m", "mmirage.shard_process"]

src/mmirage/core/process/processors/llm/llm_processor.py

@@ -63,7 +63,10 @@ def __init__(self, engine_args: SGLangLLMConfig, **kwargs) -> None:
         """
         super().__init__(engine_args, **kwargs)
         if not SGLANG_AVAILABLE:
-            raise RuntimeError("SGLang is not installed. Install with: pip install '.[gpu]'")
+            raise RuntimeError(
+                "SGLang is not installed. Install with: pip install 'mmirage[gpu]' "
+                "or, from a source checkout, pip install -e '.[gpu]'"
+            )
         self.llm = sgl.Engine(**asdict(engine_args.server_args))
         self.tokenizer = AutoTokenizer.from_pretrained(
             engine_args.server_args.model_path,