Update Dockerfile

Dockerfile

@@ -1,22 +1,26 @@
 FROM ghcr.io/ggml-org/llama.cpp:server
 
 
-RUN apt update && apt install wget -y && rm -rf /var/lib/apt/lists/*
+RUN apt update && apt install -y wget ca-certificates && rm -rf /var/lib/apt/lists/*
 
 
-# 1. THE MAIN MODEL: Granite-4.0-H-Tiny Q4_K_M — hybrid Mamba/Transformer MoE, 7B total / ~1B active
-RUN wget "https://huggingface.co/ibm-granite/granite-4.0-h-tiny-GGUF/resolve/main/granite-4.0-h-tiny-Q4_K_M.gguf" -O /model.gguf
+# TARGET MODEL: Google Gemma 4 E2B Instruct QAT (Unsloth UD-Q4_K_XL — verified, only file in repo)
+RUN wget "https://huggingface.co/unsloth/gemma-4-E2B-it-qat-GGUF/resolve/main/gemma-4-E2B-it-qat-UD-Q4_K_XL.gguf" -O /model.gguf
 
 
-# Clear the entrypoint and use shell form CMD
-# NOTE: logs are ENABLED on purpose — do not add --log-disable until it runs fine
-ENTRYPOINT []
-CMD /app/llama-server \
-    -m /model.gguf \
-    --port 7860 --host 0.0.0.0 \
-    -t 2 --threads-batch 2 \
-    -c 4096 \
-    --parallel 1 \
-    --batch-size 256 --ubatch-size 256 \
-    --mlock --no-mmap \
-    --no-warmup
+# MTP DRAFTER: Official Gemma 4 E2B MTP head (Q8_0 — verified)
+RUN wget "https://huggingface.co/unsloth/gemma-4-E2B-it-qat-GGUF/resolve/main/MTP/gemma-4-E2B-it-Q8_0-MTP.gguf" -O /draft.gguf
+
+
+# CPU-only, MTP speculative decoding — tuned for 2 vCPU / 16 GB
+CMD ["-m", "/model.gguf", \
+     "--port", "7860", "--host", "0.0.0.0", \
+     "-t", "2", "--threads-batch", "2", \
+     "--mlock", "--no-mmap", "-c", "4096", \
+     "--parallel", "1", \
+     "--batch-size", "256", "--ubatch-size", "256", \
+     "--jinja", "--flash-attn", "on", \
+     "--spec-type", "draft-mtp", "-md", "/draft.gguf", \
+     "--spec-draft-n-max", "2", "--spec-draft-n-min", "1", \
+     "--no-warmup", \
+     "--temp", "1.0", "--top-p", "0.95", "--top-k", "64"]