Upload 2 files

Dockerfile

@@ -14,18 +14,15 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
 
 WORKDIR /app
 
-# ── Python deps ──────────────────────────────────────────────
-RUN pip install --no-cache-dir \
-        fastapi==0.111.0 \
-        "uvicorn[standard]==0.29.0" \
-        pydantic==2.7.1
-
-# Install llama-cpp-python CPU-only (compiled from source)
-# 0.3.4: new enough to read current GGUF format, avoids the
-# malloc/GGML_ASSERT corruption crash introduced in 0.3.9
+COPY requirements.txt .
+
+# ── Python deps (single source of truth: requirements.txt) ───
+# CMAKE_ARGS/FORCE_CMAKE apply to the whole install so llama-cpp-python
+# gets built from source as CPU-only; the other packages are unaffected.
+# See requirements.txt for why the llama-cpp-python version matters.
 RUN CMAKE_ARGS="-DGGML_CUDA=OFF -DGGML_METAL=OFF -DGGML_OPENCL=OFF" \
     FORCE_CMAKE=1 \
-    pip install --no-cache-dir llama-cpp-python==0.3.4
+    pip install --no-cache-dir -r requirements.txt
 
 # ── App code ─────────────────────────────────────────────────
 COPY app.py .

requirements.txt

@@ -1,6 +1,8 @@
 fastapi==0.111.0
 uvicorn[standard]==0.29.0
 pydantic==2.7.1
-# llama-cpp-python is installed via pip in the Dockerfile
-# with CPU-only build flags (no CUDA / Metal)
-llama-cpp-python==0.3.9
+# Must be >=0.3.9 -- that's the release that added Qwen3 architecture
+# support to llama.cpp. Anything older will fail to load a Qwen3 GGUF
+# with "Failed to load model from file".
+# Built from source by the Dockerfile with CPU-only CMAKE_ARGS (no CUDA / Metal).
+llama-cpp-python==0.3.30