Upload folder using huggingface_hub

Dockerfile

@@ -1,13 +1,14 @@
-# tinkersnot bake-off inference Space — official llama.cpp server, git added, model baked in.
-# The :server image ENTRYPOINT is /app/llama-server, so CMD = its args.
+# tinkersnot bake-off inference Space — llama.cpp server, own entrypoint for full control.
 FROM ghcr.io/ggml-org/llama.cpp:server
 
 # HF Spaces injects a `git config` build step; the minimal ggml image lacks git.
 USER root
 RUN apt-get update && apt-get install -y --no-install-recommends git && rm -rf /var/lib/apt/lists/*
 
-# Bake the GGUF at build time (visible in build logs; no slow runtime download / hang).
-# SMOKE TEST: Qwen3-0.6B (pure text, standard arch). Swap URL for other candidates later.
+# Bake the GGUF at build time (no runtime download / hang). SMOKE TEST: Qwen3-0.6B.
 ADD https://huggingface.co/ggml-org/Qwen3-0.6B-GGUF/resolve/main/Qwen3-0.6B-Q8_0.gguf /models/model.gguf
 
-CMD ["-m", "/models/model.gguf", "--host", "0.0.0.0", "--port", "7860", "-c", "8192", "--jinja"]
+# Override the base image entrypoint entirely — our script prints boot info then execs the server.
+COPY start.sh /start.sh
+RUN chmod +x /start.sh
+ENTRYPOINT ["/start.sh"]

start.sh

@@ -0,0 +1,10 @@
+#!/bin/sh
+echo "=== BOOT ==="
+echo "id=$(id)"
+echo "PATH=$PATH"
+BIN="$(command -v llama-server || echo /app/llama-server)"
+echo "BIN=$BIN"
+ls -la "$BIN" 2>/dev/null || echo "!! binary not found at $BIN"
+echo "model:"; ls -la /models 2>/dev/null
+echo "=== starting llama-server (stderr merged to stdout) ==="
+exec "$BIN" -m /models/model.gguf --host 0.0.0.0 --port 7860 -c 8192 --jinja 2>&1