Dockerfile

# FROM ghcr.io/ggml-org/llama.cpp:server
# about 10tps

# RUN apt-get update && apt-get install wget -y && rm -rf /var/lib/apt/lists/*
# 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 /gemma-4-E2B-it-qat-UD-Q4_K_XL.gguf
# RUN wget "https://huggingface.co/unsloth/gemma-4-E2B-it-qat-GGUF/resolve/main/mtp-gemma-4-E2B-it.gguf" -O /mtp-gemma-4-E2B-it.gguf
# RUN wget "https://huggingface.co/unsloth/gemma-4-E2B-it-qat-GGUF/resolve/main/mmproj-F16.gguf" -O /gemma-4-E2B-it-mmproj.gguf

# 90 s
# RUN curl -sS -C - -L https://huggingface.co/unsloth/gemma-4-E2B-it-qat-GGUF/resolve/main/gemma-4-E2B-it-qat-UD-Q4_K_XL.gguf -o /gemma-4-E2B-it-qat-UD-Q4_K_XL.gguf \
#  && curl -sS -C - -L https://huggingface.co/unsloth/gemma-4-E2B-it-qat-GGUF/resolve/main/mtp-gemma-4-E2B-it.gguf -o /mtp-gemma-4-E2B-it.gguf \
#  && curl -sS -C - -L https://huggingface.co/unsloth/gemma-4-E2B-it-qat-GGUF/resolve/main/mmproj-F16.gguf -o /gemma-4-E2B-it-mmproj.gguf

FROM alpine:latest
RUN apk add --no-cache curl aria2 && \
  curl -LsSf https://llama.app/install.sh | sh && \
  aria2c -c -x 16 -s 16 -k 1M https://huggingface.co/unsloth/gemma-4-E2B-it-qat-GGUF/resolve/main/gemma-4-E2B-it-qat-UD-Q4_K_XL.gguf -d/ -o gemma-4-E2B-it-qat-UD-Q4_K_XL.gguf \ 
  && aria2c -c -x 16 -s 16 -k 1M https://huggingface.co/unsloth/gemma-4-E2B-it-qat-GGUF/resolve/main/mtp-gemma-4-E2B-it.gguf -d/ -o mtp-gemma-4-E2B-it.gguf \
  && aria2c -c -x 16 -s 16 -k 1M https://huggingface.co/unsloth/gemma-4-E2B-it-qat-GGUF/resolve/main/mmproj-F16.gguf -d/ -o gemma-4-E2B-it-mmproj.gguf

#   curl -sS -C - -L https://huggingface.co/unsloth/gemma-4-E2B-it-qat-GGUF/resolve/main/gemma-4-E2B-it-qat-UD-Q4_K_XL.gguf -o /gemma-4-E2B-it-qat-UD-Q4_K_XL.gguf \
#   && curl -sS -C - -L https://huggingface.co/unsloth/gemma-4-E2B-it-qat-GGUF/resolve/main/mtp-gemma-4-E2B-it.gguf -o /mtp-gemma-4-E2B-it.gguf \
#   && curl -sS -C - -L https://huggingface.co/unsloth/gemma-4-E2B-it-qat-GGUF/resolve/main/mmproj-F16.gguf -o /gemma-4-E2B-it-mmproj.gguf

ENTRYPOINT []
CMD [ \
  "/root/.local/bin/llama", "serve", \
  "-m", "/gemma-4-E2B-it-qat-UD-Q4_K_XL.gguf", \
  "--spec-draft-model", "/mtp-gemma-4-E2B-it.gguf", \
  "--mmproj", "/gemma-4-E2B-it-mmproj.gguf", \
  "--port", "7860", \
  "--host", "0.0.0.0", \
  "-t", "2", \
  "-tb", "2", \  
  "-fa", "on", \
  "-ub", "128", \
  "-b", "512", \
  "--spec-type", "draft-mtp", \
  "--spec-draft-n-max", "3", \
  "-a", "gemma-4-e2b-it-qat-ud-q4" \
]

# let llama-server autodetect -t -tb: very slow
#
# 10 tps
#  "-t", "2", \
#  "-tb", "2", \
# 
# 7.5 tps
# -fa on
#  "-t", "2", \
#  "-tb", "2", \
#  5 tps -fa off

# https://huggingface.co/spaces/Monster/gemma-4-E2B-it-GGUF/blob/main/Dockerfile