fix: Docker build - install torch CPU via --index-url, not requirements file

- Dockerfile now installs torch==2.2.2 separately with --index-url (correct way)
- requirements_hf.txt no longer contains torch (avoids pip conflict)
- Model pre-download step has graceful fallback echo on failure
- README.md: fixed colorTo orange->yellow (HF validation)

Dockerfile

@@ -1,36 +1,40 @@
-# ── Hugging Face Spaces  –  Contexto FastAPI Backend ──────────────────────────
-# Base image: slim Python (HF Spaces runs on CPU by default)
+# ── Contexto Backend – Hugging Face Spaces (CPU) ──────────────────────────────
 FROM python:3.10-slim
 
 # HF Spaces requires port 7860
 EXPOSE 7860
 
-# System deps for faiss-cpu, lxml, etc.
+# System deps
 RUN apt-get update && apt-get install -y --no-install-recommends \
     gcc g++ git curl \
     && rm -rf /var/lib/apt/lists/*
 
 WORKDIR /app
 
-# Copy requirements first for Docker layer caching
-COPY requirements_hf.txt ./requirements.txt
-
-# Install Python deps  (no CUDA → CPU-only torch, much smaller)
+# ── Step 1: Install PyTorch CPU-only FIRST (separate layer for caching) ────────
 RUN pip install --no-cache-dir --upgrade pip && \
-    pip install --no-cache-dir -r requirements.txt
+    pip install --no-cache-dir \
+        torch==2.2.2 \
+        --index-url https://download.pytorch.org/whl/cpu
 
-# Copy backend source
-COPY . .
+# ── Step 2: Copy requirements and install remaining deps ───────────────────────
+COPY requirements_hf.txt ./requirements_hf.txt
+RUN pip install --no-cache-dir -r requirements_hf.txt
 
-# Pre-download the T5 model so it is cached in the image
-# (avoids cold-start download on first request)
-RUN python -c "from transformers import T5Tokenizer, T5ForConditionalGeneration; \
-    T5Tokenizer.from_pretrained('t5-small'); \
-    T5ForConditionalGeneration.from_pretrained('t5-small')" || true
+# ── Step 3: Copy all source files ─────────────────────────────────────────────
+COPY . .
 
-# Set env so transformers uses /app/.cache (writable in HF Spaces)
+# ── Step 4: Set HuggingFace cache to writable path ────────────────────────────
 ENV TRANSFORMERS_CACHE=/app/.cache/huggingface
 ENV HF_HOME=/app/.cache/huggingface
+ENV TOKENIZERS_PARALLELISM=false
+
+# Pre-download T5-small to bake into image (avoids cold-start delay)
+RUN python -c "\
+from transformers import T5Tokenizer, T5ForConditionalGeneration; \
+T5Tokenizer.from_pretrained('t5-small'); \
+T5ForConditionalGeneration.from_pretrained('t5-small'); \
+print('T5-small cached.')" || echo "Model pre-download skipped (will download at runtime)"
 
-# Run on port 7860  (HF Spaces requirement)
+# ── Run FastAPI on port 7860 ───────────────────────────────────────────────────
 CMD ["uvicorn", "src.api:app", "--host", "0.0.0.0", "--port", "7860", "--workers", "1"]

README.md

@@ -11,7 +11,7 @@ app_port: 7860
 
 # Contexto — Intent-Aware Summarization API
 
-FastAPI backend for [Contexto](https://github.com/Dks-040204/intent-aware-context-preserving-summarization-gen-ai-rag).
+FastAPI backend for Contexto.
 
 ## Endpoints
 - `GET /health` — health check

requirements_hf.txt

@@ -1,6 +1,4 @@
-# ── Contexto Backend – Hugging Face Spaces requirements ──────────────────────
-# CPU-only torch (much smaller than the full CUDA build)
-torch==2.2.2+cpu --extra-index-url https://download.pytorch.org/whl/cpu
+# ── Contexto Backend – HF Spaces requirements (torch installed separately in Dockerfile) ──
 
 # NLP core
 transformers==4.40.1
@@ -22,8 +20,13 @@ numpy
 scikit-learn
 rouge-score
 
-# Translation (for multilingual output)
+# Translation (multilingual output)
 deep-translator==1.11.4
 
-# Document parsing (used by evaluation helpers)
+# Document parsing helpers
 requests
+
+# Document reading
+python-docx
+PyMuPDF
+mammoth