update docker files for deploying to spaces

.github/workflows/sync-to-hugging-face-hub.yml

@@ -23,5 +23,4 @@ jobs:
       - name: Push to hub
         env:
           HF_TOKEN: ${{ secrets.HF_TOKEN }}
-          HUGGINGFACE_API_KEY: ${{ secrets.HUGGINGFACE_API_KEY }}
         run: git push https://Prgrmmer:$HF_TOKEN@huggingface.co/spaces/Prgrmmer/ai-dungeon main

Dockerfile

@@ -1,7 +1,7 @@
 FROM python:3.10-slim
 
 # Add build argument
-ARG HUGGINGFACE_API_KEY
+# ARG HUGGINGFACE_API_KEY
 ENV HUGGINGFACE_API_KEY=${HUGGINGFACE_API_KEY}
 
 # System dependencies
@@ -22,17 +22,17 @@ RUN pip install --no-cache-dir -r requirements.txt
 COPY . .
 
 # # Environment variables
-# ENV PYTHONUNBUFFERED=1
+ENV PYTHONUNBUFFERED=1
 # ENV HUGGINGFACE_API_KEY=${HUGGINGFACE_API_KEY}
 
 # HuggingFace login at container startup
-# RUN --mount=type=secret,id=hf_token \
-#     HUGGINGFACE_API_KEY=$(cat /run/secrets/hf_token) && \
-#     huggingface-cli login --token $HUGGINGFACE_API_KEY
+RUN --mount=type=secret,id=HUGGINGFACE_API_KEY \
+    HUGGINGFACE_API_KEY=$(cat /run/secrets/HUGGINGFACE_API_KEY) && \
+    huggingface-cli login --token $HUGGINGFACE_API_KEY
 
-RUN mkdir -p ~/.huggingface && \
-    echo "${HUGGINGFACE_API_KEY}" > ~/.huggingface/token && \
-    huggingface-cli login --token ${HUGGINGFACE_API_KEY}
+# RUN mkdir -p ~/.huggingface && \
+#     echo "${HUGGINGFACE_API_KEY}" > ~/.huggingface/token && \
+#     huggingface-cli login --token ${HUGGINGFACE_API_KEY}
 
 # Expose port
 EXPOSE 7860

docker-compose.yml

@@ -3,12 +3,12 @@ services:
   ai-dungeon:
     build:
       context: .
-      args:
-        - HUGGINGFACE_API_KEY=${HUGGINGFACE_API_KEY}
+      # args:
+      #   - HUGGINGFACE_API_KEY=${HUGGINGFACE_API_KEY}
     ports:
       - "7860:7860"
     environment:
-      # - PYTHONUNBUFFERED=1
+      - PYTHONUNBUFFERED=1
       - HUGGINGFACE_API_KEY=${HUGGINGFACE_API_KEY}
     secrets:
       - hf_token

helper.py

@@ -67,6 +67,8 @@ def initialize_model_pipeline(model_name, force_cpu=False):
         else:
             device = MODEL_CONFIG["main_model"]["device"]
 
+        api_key = get_huggingface_api_key()
+
         # Use 8-bit quantization for memory efficiency
         model = AutoModelForCausalLM.from_pretrained(
             model_name,
@@ -75,6 +77,8 @@ def initialize_model_pipeline(model_name, force_cpu=False):
             use_cache=True,
             device_map="auto",
             low_cpu_mem_usage=True,
+            trust_remote_code=True,
+            token=api_key,  # Add token here
         )
 
         model.config.use_cache = True
@@ -114,8 +118,6 @@ try:
 
     model_name = MODEL_CONFIG["main_model"]["name"]
 
-    api_key = get_huggingface_api_key()
-
     # Initialize the pipeline with memory management
     generator, tokenizer = initialize_model_pipeline(model_name)