refactor(docker): apply multi-stage build, offline caching, update model loading and variable cleanup

Dockerfile

@@ -1,22 +1,51 @@
 # Dockerfile for Sema Translation API on HuggingFace Spaces
+# Multi-stage build to handle model downloading with proper permissions
 
-# Use an official Python runtime as a parent image
+# Stage 1: Download models as root
+FROM python:3.10-slim AS model-builder
+
+# Install huggingface_hub for downloading models
+RUN pip install huggingface_hub
+
+# Download models from sematech/sema-utils
+RUN python -c "\
+from huggingface_hub import hf_hub_download; \
+hf_hub_download('sematech/sema-utils', 'spm.model'); \
+hf_hub_download('sematech/sema-utils', 'lid218e.bin'); \
+hf_hub_download('sematech/sema-utils', 'translation_models/sematrans-3.3B/model.bin'); \
+hf_hub_download('sematech/sema-utils', 'translation_models/sematrans-3.3B/config.json'); \
+hf_hub_download('sematech/sema-utils', 'translation_models/sematrans-3.3B/shared_vocabulary.txt')"
+
+# Stage 2: Build the application
 FROM python:3.10-slim
 
-# Set the working directory in the container
-WORKDIR /code
+# Set up a new user named "user" with user ID 1000
+RUN useradd -m -u 1000 user
+
+# Switch to the "user" user
+USER user
 
-# Copy the requirements file into the container at /code
-COPY ./requirements.txt /code/requirements.txt
+# Set home to the user's home directory
+ENV HOME=/home/user \
+    PATH=/home/user/.local/bin:$PATH
 
-# Install any needed packages specified in requirements.txt
-# --no-cache-dir reduces image size
-# --upgrade pip ensures we have the latest version
+# Set the working directory to the user's home directory
+WORKDIR $HOME/app
+
+# Set environment variables for HuggingFace
+ENV HF_HUB_OFFLINE=1
+ENV TRANSFORMERS_NO_ADVISORY_WARNINGS=1
+
+# Copy the requirements file and install dependencies
+COPY --chown=user ./requirements.txt requirements.txt
 RUN pip install --no-cache-dir --upgrade pip
-RUN pip install --no-cache-dir -r /code/requirements.txt
+RUN pip install --no-cache-dir --user -r requirements.txt
+
+# Copy the downloaded models from the builder stage
+COPY --chown=user --from=model-builder /root/.cache/huggingface $HOME/.cache/huggingface
 
-# Copy the application code to the working directory
-COPY ./sema_translation_api.py /code/sema_translation_api.py
+# Copy the application code
+COPY --chown=user ./sema_translation_api.py sema_translation_api.py
 
 # Expose port 7860 (HuggingFace Spaces standard)
 EXPOSE 7860

sema_translation_api.py

@@ -69,50 +69,82 @@ def get_nairobi_time():
     full_date = f"{curr_day} | {curr_date} | {curr_time}"
     return full_date, curr_time
 
-def download_models():
-    """Download models from HuggingFace Hub using default cache"""
-    print("🔄 Downloading models from sematech/sema-utils...")
+def get_model_paths():
+    """Get model paths from HuggingFace cache (models pre-downloaded in Docker)"""
+    print("🔄 Loading models from cache...")
 
     try:
-        # Download individual files from root (using default HF cache)
-        print("📥 Downloading SentencePiece model...")
-        spm_path = hf_hub_download(
-            repo_id=REPO_ID,
-            filename="spm.model"
-        )
+        # Check if we're in offline mode (Docker environment)
+        offline_mode = os.environ.get("HF_HUB_OFFLINE", "0") == "1"
+
+        if offline_mode:
+            print("📦 Running in offline mode - using cached models")
+            # In offline mode, models are already downloaded and cached
+            # We need to find them in the cache directory
+
+            # Get paths from cache using hf_hub_download with local_files_only=True
+            spm_path = hf_hub_download(
+                repo_id=REPO_ID,
+                filename="spm.model",
+                local_files_only=True
+            )
 
-        print("📥 Downloading language detection model...")
-        ft_path = hf_hub_download(
-            repo_id=REPO_ID,
-            filename="lid218e.bin"
-        )
+            ft_path = hf_hub_download(
+                repo_id=REPO_ID,
+                filename="lid218e.bin",
+                local_files_only=True
+            )
 
-        # Download translation model files individually
-        print("📥 Downloading translation model (3.3B)...")
+            # Get the translation model path
+            model_bin_path = hf_hub_download(
+                repo_id=REPO_ID,
+                filename="translation_models/sematrans-3.3B/model.bin",
+                local_files_only=True
+            )
 
-        # Download all necessary CTranslate2 files
-        model_bin_path = hf_hub_download(
-            repo_id=REPO_ID,
-            filename="translation_models/sematrans-3.3B/model.bin"
-        )
+            # The model directory is the parent of the model.bin file
+            ct_model_full_path = os.path.dirname(model_bin_path)
 
-        hf_hub_download(
-            repo_id=REPO_ID,
-            filename="translation_models/sematrans-3.3B/config.json"
-        )
+        else:
+            print("🌐 Running in online mode - downloading models")
+            # Online mode - download models (for local development)
+            spm_path = hf_hub_download(
+                repo_id=REPO_ID,
+                filename="spm.model"
+            )
 
-        hf_hub_download(
-            repo_id=REPO_ID,
-            filename="translation_models/sematrans-3.3B/shared_vocabulary.txt"
-        )
+            ft_path = hf_hub_download(
+                repo_id=REPO_ID,
+                filename="lid218e.bin"
+            )
 
-        # The model directory is the parent of the model.bin file
-        ct_model_full_path = os.path.dirname(model_bin_path)
+            # Download all necessary CTranslate2 files
+            model_bin_path = hf_hub_download(
+                repo_id=REPO_ID,
+                filename="translation_models/sematrans-3.3B/model.bin"
+            )
+
+            hf_hub_download(
+                repo_id=REPO_ID,
+                filename="translation_models/sematrans-3.3B/config.json"
+            )
+
+            hf_hub_download(
+                repo_id=REPO_ID,
+                filename="translation_models/sematrans-3.3B/shared_vocabulary.txt"
+            )
+
+            ct_model_full_path = os.path.dirname(model_bin_path)
+
+        print(f"📁 Model paths:")
+        print(f"   SentencePiece: {spm_path}")
+        print(f"   Language detection: {ft_path}")
+        print(f"   Translation model: {ct_model_full_path}")
 
         return spm_path, ft_path, ct_model_full_path
 
     except Exception as e:
-        print(f"❌ Error downloading models: {e}")
+        print(f"❌ Error loading models: {e}")
         raise e
 
 def load_models():
@@ -121,8 +153,8 @@ def load_models():
 
     print("🚀 Loading models into memory...")
 
-    # Download models first
-    spm_path, ft_path, ct_model_path = download_models()
+    # Get model paths (from cache or download)
+    spm_path, ft_path, ct_model_path = get_model_paths()
 
     # Suppress fasttext warnings
     fasttext.FastText.eprint = lambda x: None